{"meta":{"schema":"aimap.ai-disciplines/v1","schemaExtensions":"v2-additive (2026-10-21): dimensions, compositeScore, levelCapped, levelChange, biggestBreakthrough, keyPapers, latest_papers, milestone.type/why/aiImpact/fieldImpact/impactNotes/verified/datePrecision/datePrecision, stats.monthsSinceLatestMilestone/leaps/…, meta.recentLeaps","generatedAt":"2026-10-21T14:48:51+08:00","dataAsOf":"2026-10-22","title":{"zh":"AI 改造各学科进度","en":"AI × Disciplines Progress"},"levels":[{"level":0,"key":"not-started","zh":"尚未改造","en":"Not yet transformed","color":"#4b5563","criteria_zh":"综合分 < 0.75：只有零星使用或讨论，AI 尚未改变本学科的研究方法、发现、从业实践或教学。","criteria_en":"Composite < 0.75: only scattered use or discussion; AI has not changed methods, discovery, practice or teaching."},{"level":1,"key":"assistive","zh":"辅助起步","en":"Assistive","color":"#3b82f6","criteria_zh":"综合分 0.75–1.49：AI 作为辅助工具在部分研究或从业环节使用，有案例或调查，但没有以 AI 为关键的公认成果。","criteria_en":"Composite 0.75–1.49: AI is an aid in parts of research or practice, with cases or surveys, but no recognised AI-key result."},{"level":2,"key":"breakthrough","zh":"局部突破","en":"Local breakthroughs","color":"#a855f7","criteria_zh":"综合分 1.5–2.49：至少一项经同行评审或官方验证、AI 起关键作用的成果，或某一维度已达到 3 分；但常规工作流整体未变。","criteria_en":"Composite 1.5–2.49: at least one peer-reviewed/officially verified AI-key result, or one dimension at 3; routine workflow largely unchanged."},{"level":3,"key":"transforming","zh":"系统改造","en":"Systemic transformation","color":"#f59e0b","criteria_zh":"综合分 2.5–3.49（或 ≥3.5 但方法/发现/采用有一项低于 3）：AI 成为标准基础设施或业务流程，多项独立突破。","criteria_en":"Composite 2.5–3.49 (or ≥3.5 with any of method/discovery/adoption below 3): AI is standard infrastructure with multiple independent breakthroughs."},{"level":4,"key":"ai-led","zh":"AI 主导","en":"AI-led","color":"#ef4444","criteria_zh":"综合分 ≥ 3.5，且研究方法、科学发现、从业采用三项都 ≥ 3：AI 成规模自主产出核心成果并经独立验证。","criteria_en":"Composite ≥ 3.5 and method, discovery and adoption all ≥ 3: AI produces core results at scale, independently verified."}],"categories":[{"id":"formal-physical","zh":"数理科学","en":"Mathematical & physical sciences"},{"id":"life-medical","zh":"生命与医学","en":"Life & medical sciences"},{"id":"eng-computing","zh":"工程与计算","en":"Engineering & computing"},{"id":"earth-env","zh":"地球与环境","en":"Earth & environment"},{"id":"social","zh":"社会科学","en":"Social sciences"},{"id":"humanities","zh":"人文","en":"Humanities"},{"id":"arts","zh":"艺术","en":"Arts"}],"scoring":{"dimensions":[{"key":"method","zh":"研究方法","en":"Research method","weight":0.3,"scale":"0–4"},{"key":"discovery","zh":"科学发现","en":"Scientific discovery","weight":0.35,"scale":"0–4"},{"key":"adoption","zh":"从业者采用","en":"Practitioner adoption","weight":0.25,"scale":"0–4"},{"key":"education","zh":"教育/课程","en":"Education & curriculum","weight":0.1,"scale":"0–4"}],"formula":"S = 0.30·method + 0.35·discovery + 0.25·adoption + 0.10·education（教育缺证据时记 0 分；若为空则按其余权重归一）","thresholds":{"L0":"<0.75","L1":"0.75–1.49","L2":"1.5–2.49","L3":"2.5–3.49","L4":"≥3.5 且 method/discovery/adoption 均 ≥3"},"evidenceRule":"维度 ≥3 分须至少一条非厂商独立来源"},"impactTiers":[{"key":"leap","zh":"飞跃性突破","en":"Leap","rule_zh":"解决长期公开难题、首次超越最强人类/现有最佳方法、或改变学科常规做法，且经同行评审或独立专家/官方核验"},{"key":"major","zh":"重要进展","en":"Major","rule_zh":"顶刊/顶会成果、重要系统发布、规模化部署或政策，影响明显但局部"},{"key":"incremental","zh":"增量进展","en":"Incremental","rule_zh":"改进、数据集、调查、个案或早期信号"}],"impactRule":"厂商自述、有争议、未同行评审预印本或日期待核实的条目不得评为“飞跃性突破”（自动降为“重要进展”并写入 impactNotes）；例外须有独立核验并在 impactNotes 说明。","milestoneTypes":[{"key":"paper","zh":"论文","en":"Paper"},{"key":"system","zh":"系统发布","en":"System"},{"key":"benchmark","zh":"基准","en":"Benchmark"},{"key":"competition","zh":"竞赛","en":"Competition"},{"key":"deployment","zh":"落地部署","en":"Deployment"},{"key":"policy","zh":"政策","en":"Policy"},{"key":"industry","zh":"产业/采用","en":"Industry"},{"key":"clinical","zh":"临床","en":"Clinical"},{"key":"regulatory","zh":"监管","en":"Regulatory"},{"key":"prize","zh":"奖项","en":"Prize"},{"key":"dataset","zh":"数据集","en":"Dataset"},{"key":"discovery","zh":"科学发现","en":"Discovery"},{"key":"model","zh":"模型发布","en":"Model"},{"key":"legal","zh":"诉讼/判决","en":"Legal"}],"counts":{"disciplines":43,"byLevel":{"0":3,"1":8,"2":23,"3":9,"4":0},"byCategory":{"formal-physical":6,"life-medical":9,"eng-computing":8,"earth-env":3,"social":9,"humanities":5,"arts":3},"milestones":384,"milestonesV1":138,"news":352,"uniqueNews":301,"leapsAi":23,"leapsField":18,"latestPapers":314,"keyPapers":221},"recentLeaps":[{"discipline":"cybersecurity","date":"2026-10-16","title":{"zh":"Anthropic 团队用 Claude 破解 NIST P-256 等椭圆曲线种子来源之谜：1997–98 年用 drand48 以时间戳生成，领走悬赏","en":"Claude traces the NIST elliptic-curve seeds to drand48 seeded with 1997–98 timestamps; bounty claimed"},"url":"https://www.anthropic.com/research/ecdsa-curve-origins","aiImpact":"leap","fieldImpact":"major","why":{"zh":"AI：困扰密码学界 25 年的种子来源由 AI 辅助团队解出，悬赏发起人 Filippo Valsorda 已独立重实现并核验；学科：排除了 Bernstein–Lange 设想的种子操纵后门空间（自由度仅约 30 比特），恢复对 NIST 曲线的信任，但不改变其安全性","en":"AI: a 25-year mystery solved by an AI-assisted team, independently re-implemented and verified by bounty organiser Filippo Valsorda; Field: rules out seed-manipulation backdoors (≈30 bits of freedom) without changing curve security"},"notes":[{"zh":"厂商发布，但悬赏发起人 Filippo Valsorda 已用 Go 独立重实现并核验全部种子，悬赏已兑付","en":"Vendor post, but bounty organiser Filippo Valsorda independently re-implemented and verified the seeds; bounty paid"}]},{"discipline":"mathematics","date":"2026-10-11","title":{"zh":"Banks 逐行核验并重写 OpenAI 的“朗道–西格尔零点不存在”证明，给出显式常数（q₀=10¹⁰，δ₀=3.0×10⁻³）","en":"Banks: human-verified, explicit account of the OpenAI proof that Landau–Siegel zeros do not exist (arXiv 2610.15483)"},"url":"https://arxiv.org/abs/2610.15483","aiImpact":"leap","fieldImpact":"leap","why":{"zh":"AI：OpenAI 模型给出解析数论近百年核心难题的证明（2026-10-01 手稿），密苏里大学 Banks 逐行核验并称论证成立，Kerr 另给出更简证明（arXiv 2610.21099），Lean 形式化经两个独立内核复核；学科：西格尔零点阻碍算术级数素数分布与类数的有效估计，若最终确认将改写多个经典结果（尚未经期刊审稿，陶哲轩称仍在消化）","en":"AI: an OpenAI model produced a proof of a near-century-old central problem; Banks verified it line by line, Kerr gave a simpler proof (arXiv 2610.21099), and the Lean formalisation was re-checked by two kernels. Field: Siegel zeros block effective bounds for primes in progressions and class numbers (not yet journal-refereed)"},"notes":[{"zh":"预印本，但已由密苏里大学 Banks 逐行核验（本文即核验工作），Kerr 独立给出更简证明（arXiv 2610.21099），OpenAI 的 Lean 形式化经两个独立内核复核；尚未经期刊审稿","en":"Preprint, but it is itself a line-by-line verification by Banks; Kerr gave an independent simpler proof (arXiv 2610.21099) and the Lean formalisation was re-checked by two kernels; not yet journal-refereed"}]},{"discipline":"mathematics","date":"2026-10-01","title":{"zh":"Song 与 Zhang 借助人类主导的 AI 智能体证明 31 年悬而未决的 KLS 猜想（各向同性对数凹测度的 Cheeger 常数有维度无关下界）；Klartag 等三位专家独立给出证明并称其判据为“关键一步”","en":"Song & Zhang prove the 31-year-old Kannan–Lovász–Simonovits conjecture with human-directed AI agents (arXiv 2610.01447); Bizeul–Klartag–Lehec give an independent proof built on their criterion (arXiv 2610.05474)"},"url":"https://arxiv.org/abs/2610.01447","aiImpact":"major","fieldImpact":"leap","why":{"zh":"AI：微软研究院称研究者用人类主导的智能体流程探索了 100 多种思路，由人筛选并完成论证（AI 的具体贡献为微软自述，未经独立核实）；学科：KLS 是高维凸几何的核心公开问题（1995 年提出），Bizeul、Klartag、Lehec 于 10-04 独立给出证明（arXiv 2610.05474），Balasubramanian 等另给出显式常数证明（arXiv 2610.07728），并有 Lean 4 形式化；均为预印本，尚未经期刊审稿","en":"AI: per Microsoft Research a human-directed agentic process explored 100+ approaches (AI role is self-reported). Field: KLS (1995) is a central problem of high-dimensional convex geometry; Bizeul–Klartag–Lehec independently proved it (arXiv 2610.05474), another explicit-constant proof appeared (arXiv 2610.07728), plus a Lean 4 formalisation; all preprints"},"notes":[{"zh":"预印本，但数学结论已由 Bizeul、Klartag、Lehec 独立给出证明并称 Song–Zhang 判据为“关键一步”（arXiv 2610.05474），另有独立显式常数证明（arXiv 2610.07728）与 Lean 4 形式化；AI 作用为微软自述，故 AI 影响只评“重要进展”","en":"Preprint, but independently re-proved by Bizeul–Klartag–Lehec, who call the Song–Zhang criterion the crucial step (arXiv 2610.05474), plus another proof (arXiv 2610.07728) and a Lean 4 formalisation; AI role is self-reported, so AI impact is Major"}]},{"discipline":"archaeology-classics","date":"2026-06-25","title":{"zh":"维苏威挑战赛：首次完整读出一整卷赫库兰尼姆纸草（PHerc. 1667，约 20 栏、近 1.5 米）","en":"Vesuvius Challenge: first entire Herculaneum scroll read (PHerc. 1667, ~20 columns, ~1.5 m)"},"url":"https://scrollprize.org/firstscroll","aiImpact":"major","fieldImpact":"leap","why":{"zh":"近 2000 年前被火山碳化、无法展开的卷轴整卷被读出，开启大规模阅读古代失传文本的可能","en":"A carbonised 2,000-year-old scroll read in full without opening"},"notes":[]},{"discipline":"mathematics","date":"2026-05-21","title":{"zh":"DeepMind 形式化证明搜索智能体在 Lean 中自主解决 353 个公开 Erdős 问题中的 9 个、44/492 个 OEIS 猜想（arXiv 2605.22763）","en":"DeepMind formal proof-search agent autonomously resolves 9 of 353 open Erdős problems and 44/492 OEIS conjectures in Lean"},"url":"https://arxiv.org/abs/2605.22763","aiImpact":"leap","fieldImpact":"major","why":{"zh":"首次大规模评估 AI 在 Lean 中求解公开问题，结果可机器核验，单题成本几百美元","en":"First large-scale evaluation on open problems with machine-checked Lean proofs at a few hundred dollars per problem"},"notes":[{"zh":"预印本，但所有证明均经 Lean 机器核验","en":"Preprint, but all proofs are machine-checked in Lean"}]},{"discipline":"mathematics","date":"2026-05-20","title":{"zh":"单位距离猜想被推翻：OpenAI 通用推理模型构造出 n^{1+ε} 个单位距离的点集，否定 Erdős 1946 年的 n^{1+o(1)} 猜想","en":"Unit-distance conjecture disproved: an OpenAI general reasoning model constructs point sets with n^{1+ε} unit distances"},"url":"https://openai.com/index/model-disproves-discrete-geometry-conjecture/","aiImpact":"leap","fieldImpact":"leap","why":{"zh":"悬而未决约 80 年、被称为组合几何“最著名的问题”；证明经 Alon、Gowers、Sawin 等 9 位外部数学家核验并写成人类可读版本（arXiv 2605.20695），Sawin 将 ε 改进到 0.014","en":"~80-year-old flagship problem; proof checked and rewritten by nine outside mathematicians (arXiv 2605.20695); Sawin improved ε to 0.014"},"notes":[{"zh":"OpenAI 发布，但证明已由 Alon、Gowers、Sawin 等 9 位外部数学家独立核验并公开（arXiv 2605.20695）","en":"Announced by OpenAI, but independently verified by nine outside mathematicians (arXiv 2605.20695)"}]},{"discipline":"mathematics","date":"2026-05-20","title":{"zh":"《关于单位距离猜想被推翻的评注》：9 位数学家给出人类核验、简化和推广的证明（arXiv 2605.20695）","en":"“Remarks on the disproof of the unit distance conjecture”: human-verified, simplified proof by nine mathematicians"},"url":"https://arxiv.org/abs/2605.20695","aiImpact":"major","fieldImpact":"leap","why":{"zh":"独立专家核验是把 AI 结果算作数学成果的关键一步；文中指出论证依赖 Ellenberg–Venkatesh、Golod–Shafarevich 等思想","en":"Independent expert verification is what makes the AI result count as mathematics"},"notes":[{"zh":"预印本，但作者为 9 位独立数学家，本身即核验工作","en":"Preprint, but it is itself the independent verification by nine mathematicians"}]},{"discipline":"genomics","date":"2026-03-04","title":{"zh":"Evo 2：400 亿参数、百万碱基上下文，跨所有生命域建模与设计基因组（Nature）","en":"Evo 2: 40B-parameter genome model with 1 Mb context across all domains of life (Nature)"},"url":"https://doi.org/10.1038/s41586-026-10176-5","aiImpact":"leap","fieldImpact":"major","why":{"zh":"训练 9.3 万亿碱基，可预测变异效应并生成经实验验证的序列","en":"Trained on 9.3T bases; predicts variant effects and designs validated sequences"},"notes":[]},{"discipline":"medical-imaging","date":"2026-01","title":{"zh":"MASAI 终期结果：AI 辅助乳腺筛查间期癌非劣效，敏感度 80.5% vs 73.8%（The Lancet）","en":"MASAI final results: AI-supported screening non-inferior on interval cancers, sensitivity 80.5% vs 73.8% (The Lancet)"},"url":"https://doi.org/10.1016/s0140-6736(25)02464-x","aiImpact":"major","fieldImpact":"leap","why":{"zh":"10.6 万名女性的随机试验首次以间期癌这一硬终点证明 AI 筛查可安全替代双人阅片，作者建议可考虑临床实施","en":"RCT of 105,934 women shows non-inferiority on the hard endpoint of interval cancers"},"notes":[]},{"discipline":"life-sciences","date":"2025-11-05","title":{"zh":"RFdiffusion 实现原子级精确的从头抗体设计（Nature）","en":"Atomically accurate de novo design of antibodies with RFdiffusion (Nature)"},"url":"https://doi.org/10.1038/s41586-025-09721-5","aiImpact":"leap","fieldImpact":"leap","why":{"zh":"抗体此前被认为最难从头设计；冷冻电镜结构证实设计精度","en":"Antibodies were the hardest de novo design target; cryo-EM confirms accuracy"},"notes":[]},{"discipline":"mathematics","date":"2025-11-03","title":{"zh":"AlphaEvolve 数学探索论文：67 个问题，多处改进已知最佳构造，11 维接吻数下界 592→593（arXiv 2511.02864）","en":"AlphaEvolve at scale: 67 problems, several best-known constructions improved; 11-D kissing number 592→593"},"url":"https://arxiv.org/abs/2511.02864","aiImpact":"leap","fieldImpact":"major","why":{"zh":"由 Tao 等人联合署名，覆盖分析、组合、几何、数论，把 AI 构造搜索变成规模化方法","en":"Co-authored by Terence Tao; turns AI-driven construction search into a scalable method across fields"},"notes":[{"zh":"预印本已于 2026-09-30 经同行评审发表于 PNAS","en":"Peer-reviewed version published in PNAS on 2026-09-30"}]},{"discipline":"cybersecurity","date":"2025-08-08","title":{"zh":"DARPA AIxCC 决赛：Team Atlanta 夺冠，7 支队伍的自主系统发现约 80% 植入漏洞、修补 60% 以上并全部开源","en":"DARPA AIxCC final: Team Atlanta wins; autonomous systems find ~80% of planted bugs and patch 60%+, all open-sourced"},"url":"https://www.darpa.mil/news/2025/aixcc-results","aiImpact":"leap","fieldImpact":"major","why":{"zh":"政府组织、独立评分的大型赛事证明 AI 可自主发现并修补真实开源软件漏洞（原“待核实”已确认 2025-08-08）","en":"Independently scored government challenge shows autonomous find-and-patch"},"notes":[]},{"discipline":"mathematics","date":"2025-07-21","title":{"zh":"Gemini Deep Think 获 IMO 2025 官方评分金牌（35/42）","en":"Gemini Deep Think officially graded gold at IMO 2025 (35/42)"},"url":"https://deepmind.google/discover/blog/advanced-version-of-gemini-with-deep-think-officially-achieves-gold-medal-standard-at-the-international-mathematical-olympiad/","aiImpact":"leap","fieldImpact":"incremental","why":{"zh":"首个经 IMO 官方评分的 AI 金牌（35/42），用自然语言作答","en":"First AI gold officially graded by the IMO (35/42)"},"notes":[]},{"discipline":"mathematics","date":"2025-05-14","title":{"zh":"AlphaEvolve 改进 4×4 复矩阵乘法等数十个问题","en":"AlphaEvolve improves 4×4 complex matrix multiplication and dozens of problems"},"url":"https://deepmind.google/discover/blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/","aiImpact":"leap","fieldImpact":"major","why":{"zh":"4×4 复矩阵乘法用 48 次乘法，改进 Strassen 1969 年以来的结果；数十个问题追平或超过已知最佳","en":"48-multiplication 4×4 complex matrix product improves on Strassen (1969)"},"notes":[]},{"discipline":"computer-science","date":"2025-05-14","title":{"zh":"AlphaEvolve 发布：通用算法发现编码智能体","en":"AlphaEvolve: coding agent for algorithm discovery"},"url":"https://deepmind.google/discover/blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/","aiImpact":"leap","fieldImpact":"major","why":{"zh":"通用进化式编码智能体，在算法和数学问题上多处超过已知最佳","en":"General evolutionary coding agent beating best-known results"},"notes":[]},{"discipline":"weather-climate","date":"2025-02-25","title":{"zh":"ECMWF AIFS 转为业务运行","en":"ECMWF AIFS becomes operational"},"url":"https://www.ecmwf.int/en/about/media-centre/news/2025/ecmwfs-ai-forecasts-become-operational","aiImpact":"major","fieldImpact":"leap","why":{"zh":"全球顶级预报中心首次把 AI 模型转为业务运行","en":"First operational AI model at a top forecasting centre"},"notes":[]},{"discipline":"weather-climate","date":"2024-12-04","title":{"zh":"GenCast 概率集合预报（Nature）","en":"GenCast probabilistic ensemble forecasting (Nature)"},"url":"https://www.nature.com/articles/s41586-024-08252-9","aiImpact":"major","fieldImpact":"leap","why":{"zh":"概率集合预报在 97% 指标上超过 ECMWF ENS","en":"Beats ECMWF ENS on 97% of targets"},"notes":[]},{"discipline":"chemistry","date":"2024-10-09","title":{"zh":"诺贝尔化学奖授予 Baker、Hassabis、Jumper（蛋白质设计与结构预测）","en":"Chemistry Nobel to Baker, Hassabis & Jumper"},"url":"https://www.nobelprize.org/prizes/chemistry/2024/summary/","aiImpact":"leap","fieldImpact":"leap","why":{"zh":"诺贝尔化学奖授予 AlphaFold 与计算蛋白质设计，AI 成果首次获化学最高奖","en":"First chemistry Nobel for AI-driven work"},"notes":[]},{"discipline":"life-sciences","date":"2024-10-09","title":{"zh":"AlphaFold 与蛋白质设计获诺贝尔化学奖","en":"AlphaFold and protein design win the Chemistry Nobel"},"url":"https://www.nobelprize.org/prizes/chemistry/2024/summary/","aiImpact":"leap","fieldImpact":"leap","why":{"zh":"诺贝尔化学奖授予 AI 结构预测与蛋白质设计","en":"Nobel prize for AI structure prediction and design"},"notes":[]},{"discipline":"physics","date":"2024-10-08","title":{"zh":"诺贝尔物理学奖授予 Hopfield 与 Hinton（神经网络的物理基础）","en":"Nobel Prize in Physics to Hopfield & Hinton (foundations of neural networks)"},"url":"https://www.nobelprize.org/prizes/physics/2024/summary/","aiImpact":"leap","fieldImpact":"major","why":{"zh":"诺贝尔物理学奖首次授予神经网络基础研究，标志 AI 被物理学界正式承认","en":"First physics Nobel for neural-network foundations"},"notes":[]},{"discipline":"neuroscience","date":"2024-10-02","title":{"zh":"FlyWire：成年果蝇全脑连接组（Nature）","en":"FlyWire: whole-brain connectome of the adult fly (Nature)"},"url":"https://www.nature.com/articles/s41586-024-07558-y","aiImpact":"major","fieldImpact":"leap","why":{"zh":"首个成年果蝇全脑连接组（约 14 万神经元），AI 分割加人工校对完成","en":"First whole adult fly brain connectome (~140k neurons)"},"notes":[]},{"discipline":"film-animation","date":"2024-02-15","title":{"zh":"OpenAI 发布文生视频模型 Sora","en":"OpenAI unveils text-to-video model Sora"},"url":"https://openai.com/index/sora/","aiImpact":"leap","fieldImpact":"major","why":{"zh":"首个能生成一分钟高保真视频的模型，引发影视业广泛反应","en":"First minute-long high-fidelity video model"},"notes":[]},{"discipline":"archaeology-classics","date":"2024-02-05","title":{"zh":"维苏威挑战赛大奖：AI 读出赫库兰尼姆纸草段落","en":"Vesuvius Challenge Grand Prize: AI reads Herculaneum papyrus passages"},"url":"https://scrollprize.org/grandprize","aiImpact":"major","fieldImpact":"leap","why":{"zh":"AI 首次从未展开的碳化卷轴读出成段文字","en":"First passages read from an unopened carbonised scroll"},"notes":[]},{"discipline":"mathematics","date":"2023-12-14","title":{"zh":"FunSearch 用大模型找到帽集问题新构造（Nature）","en":"FunSearch finds new cap-set constructions with an LLM (Nature)"},"url":"https://www.nature.com/articles/s41586-023-06924-6","aiImpact":"leap","fieldImpact":"major","why":{"zh":"LLM 首次在公开数学问题（帽集）上找到超过人类已知最佳的新构造","en":"First LLM-found construction beating best-known human results on an open problem (cap sets)"},"notes":[]},{"discipline":"materials","date":"2023-11-29","title":{"zh":"GNoME 预测 220 万种新晶体，约 38 万种稳定（Nature）","en":"GNoME predicts 2.2M new crystals, ~380k stable (Nature)"},"url":"https://www.nature.com/articles/s41586-023-06735-9","aiImpact":"leap","fieldImpact":"major","why":{"zh":"预测 220 万种晶体、约 38 万种稳定，稳定材料已知数量扩大约一个数量级；新颖性后遭部分学者质疑","en":"2.2M crystals, ~380k stable; novelty later questioned"},"notes":[]},{"discipline":"weather-climate","date":"2023-11-14","title":{"zh":"GraphCast 中期天气预报（Science）","en":"GraphCast medium-range forecasting (Science)"},"url":"https://www.science.org/doi/10.1126/science.adi2336","aiImpact":"major","fieldImpact":"leap","why":{"zh":"90% 指标优于 HRES，1 分钟出 10 天预报","en":"Beats HRES on 90% of targets, 10-day forecast in a minute"},"notes":[]},{"discipline":"aerospace","date":"2023-08-30","title":{"zh":"Swift：深度强化学习无人机在竞速中击败人类世界冠军（Nature）","en":"Champion-level drone racing with deep RL (Nature)"},"url":"https://doi.org/10.1038/s41586-023-06419-4","aiImpact":"leap","fieldImpact":"major","why":{"zh":"首次在实体竞技运动中由 AI 击败人类世界冠军","en":"First AI to beat human world champions in a physical sport"},"notes":[]},{"discipline":"life-sciences","date":"2023-07-11","title":{"zh":"RFdiffusion 从头设计蛋白质（Nature）","en":"RFdiffusion de novo protein design (Nature)"},"url":"https://www.nature.com/articles/s41586-023-06415-8","aiImpact":"major","fieldImpact":"leap","why":{"zh":"扩散模型从头设计结合蛋白，实验成功率大幅提高","en":"Diffusion-based de novo binder design with high success rates"},"notes":[]},{"discipline":"weather-climate","date":"2023-07-05","title":{"zh":"盘古气象大模型（Nature）","en":"Pangu-Weather (Nature)"},"url":"https://www.nature.com/articles/s41586-023-06185-3","aiImpact":"leap","fieldImpact":"leap","why":{"zh":"首个在中期预报精度上超过欧洲中心业务数值模式的 AI 模型","en":"First AI model to beat ECMWF’s operational NWP on medium-range skill"},"notes":[]},{"discipline":"computer-science","date":"2022-10-05","title":{"zh":"AlphaTensor 发现更快的矩阵乘法算法（Nature）","en":"AlphaTensor discovers faster matrix multiplication (Nature)"},"url":"https://www.nature.com/articles/s41586-022-05172-4","aiImpact":"leap","fieldImpact":"major","why":{"zh":"50 年来首次在 4×4 矩阵（模 2）上改进 Strassen 算法的乘法次数","en":"First improvement on Strassen for 4×4 (mod 2) in 50 years"},"notes":[]}],"newsSources":{"collectorRun":"news/scripts/collect_all_domains.py：2026-10-20 18:41（14 天，2147 条）与 2026-10-21（90 天 2026-07-22→10-20，3359 条，0 失败源）两轮合并","samples":"research/news-summary/samples-all.json (2026-10-10, 199 条)","manual":"2026-10-20/21 WebSearch 补充，标 via=manual-websearch-2026102x","papers":"scripts/fetch_latest_papers.py：arXiv API、bioRxiv/medRxiv API、OpenAlex（近 30 天）；中文标题与要点依据摘要人工撰写"},"linkCheckLegend":{"ok":"脚本 HTTP 200","ok-crossref":"站点拒绝脚本，但 DOI/arXiv 经 Crossref/arXiv API 核实题名与日期","site-blocks-bots":"站点对脚本返回 401/403，已用搜索结果人工核对","from-feed":"来自 RSS/API 采集，链接即原文"},"notes":{"zh":"等级由四维评分按公式计算，维度证据见 dimensions；标“厂商自述”的数字未经独立核实；突破评级规则见 impactRule。","en":"Levels are computed from four dimension scores; vendor-claimed figures are not independently verified; see impactRule for breakthrough ratings."},"dailyRefresh":{"date":"2026-10-22","by":"iteration-10 每日刷新","papers":"scripts/fetch_latest_papers.py 30 天窗口 + 昨日未过期候选；中文要点见 _work/papers/zh/p6.py","news":"manual-websearch-20261022 / pool-v2-20261022"},"afternoonRefresh":{"date":"2026-10-22","by":"2026-10-22 下午轮（iteration-11）","milestonesAdded":26,"script":"research/ai-disciplines/_work/v2/pm1022/patch.py","items":"research/ai-disciplines/_work/v2/pm1022/items.py"},"auditRefresh":{"date":"2026-10-22","by":"2026-10-22 前沿栏目审查（audit-2026-10-22/frontier.md）","milestonesAdded":11,"script":"research/ai-disciplines/_work/v2/audit1022/patch.py","items":"research/ai-disciplines/_work/v2/audit1022/items.py"}},"disciplines":[{"id":"mathematics","slug":"mathematics","category":"formal-physical","categoryName":{"zh":"数理科学","en":"Mathematical & physical sciences"},"name":{"zh":"数学","en":"Mathematics"},"classification":{"oecdFos":"1.1","moe":"0701 数学"},"level":3,"levelKey":"transforming","levelLabel":{"zh":"系统改造","en":"Systemic transformation"},"trend":"up","reason":{"zh":"数学是 AI 科学发现最强的学科：2026-05 OpenAI 模型推翻 1946 年 Erdős 单位距离猜想（Alon、Gowers、Sawin 等 9 位数学家写了人类核验版），PNAS 发表 AlphaEvolve 在 67 个问题上的系统探索，DeepMind 智能体在 Lean 中独立解决 9 个 Erdős 公开问题。但从业者普遍采用和课程改革尚未发生，openai/math 还出现撤回和公开抵制，所以综合定为 L3（发现维度 4 分）。","en":"Maths has the strongest AI-discovery record: in May 2026 an OpenAI model disproved Erdős’s 1946 unit-distance conjecture (a human-verified write-up by nine mathematicians incl. Alon, Gowers, Sawin), PNAS published AlphaEvolve’s 67-problem study, and a DeepMind agent resolved 9 open Erdős problems in Lean. Broad practitioner adoption and curriculum change have not happened, and openai/math saw withdrawals and push-back, so the composite is L3 (discovery scored 4)."},"reasonSources":[{"url":"https://arxiv.org/abs/2605.20695","linkCheck":"ok"},{"url":"https://www.pnas.org/doi/10.1073/pnas.2536158123","linkCheck":"ok-crossref"},{"url":"https://arxiv.org/abs/2605.22763","linkCheck":"ok-crossref"}],"directions":{"zh":["形式化证明（Lean）","自动形式化","大模型推理求解开放问题","进化式算法搜索（AlphaEvolve）","竞赛数学"],"en":["Formal proof (Lean)","Autoformalization","LLM reasoning on open problems","Evolutionary search (AlphaEvolve)","Competition maths"]},"stats":{"milestones":24,"milestonesSince2025":18,"milestones2026":14,"milestonesLast6Months":14,"news":12,"latestProgress":"2026-10-20","latestMilestone":"2026-10-20","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":0,"leaps":{"ai":7,"field":4},"latestPapers":8,"keyPapers":14},"milestones":[{"date":"2021-02-03","title":{"zh":"拉马努金机：算法自动生成基本常数的连分数猜想（Nature）","en":"The Ramanujan Machine: algorithmically generated conjectures on fundamental constants (Nature)"},"url":"https://doi.org/10.1038/s41586-021-03229-4","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"计算机自动提出数学常数公式猜想，部分后被证明","en":"Automated conjecture generation for constants; some later proved"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"计算机自动提出数学常数公式猜想，部分后被证明","en":"Automated conjecture generation for constants; some later proved"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"计算机自动提出数学常数公式猜想，部分后被证明","en":"Automated conjecture generation for constants; some later proved"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-021-03229-4","title":"Generating conjectures on fundamental constants with the Ramanujan Machine","venue":"Nature","date":"2021-02-03"},"addedIn":"v2"},{"date":"2021-12-01","title":{"zh":"DeepMind 与数学家合作：机器学习引导直觉发现纽结理论与表示论新结果（Nature）","en":"ML guides mathematicians’ intuition to new results in knot theory and representation theory (Nature)"},"url":"https://doi.org/10.1038/s41586-021-04086-x","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"首次在顶刊证明 AI 能帮助数学家提出并证明新定理（纽结不变量关系、Kazhdan–Lusztig 多项式猜想进展）","en":"First top-journal demonstration that AI can help mathematicians conjecture and prove new theorems"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首次在顶刊证明 AI 能帮助数学家提出并证明新定理（纽结不变量关系、Kazhdan–Lusztig 多项式猜想进展）","en":"First top-journal demonstration that AI can help mathematicians conjecture and prove new theorems"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首次在顶刊证明 AI 能帮助数学家提出并证明新定理（纽结不变量关系、Kazhdan–Lusztig 多项式猜想进展）","en":"First top-journal demonstration that AI can help mathematicians conjecture and prove new theorems"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-021-04086-x","title":"Advancing mathematics by guiding human intuition with AI","venue":"Nature","date":"2021-12-01"},"addedIn":"v2"},{"date":"2023-12-14","title":{"zh":"FunSearch 用大模型找到帽集问题新构造（Nature）","en":"FunSearch finds new cap-set constructions with an LLM (Nature)"},"url":"https://www.nature.com/articles/s41586-023-06924-6","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"LLM 首次在公开数学问题（帽集）上找到超过人类已知最佳的新构造","en":"First LLM-found construction beating best-known human results on an open problem (cap sets)"},"aiImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"LLM 首次在公开数学问题（帽集）上找到超过人类已知最佳的新构造","en":"First LLM-found construction beating best-known human results on an open problem (cap sets)"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"LLM 首次在公开数学问题（帽集）上找到超过人类已知最佳的新构造","en":"First LLM-found construction beating best-known human results on an open problem (cap sets)"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-023-06924-6","title":"Mathematical discoveries from program search with large language models","venue":"Nature","date":"2023-12-14"},"addedIn":"v1"},{"date":"2024-01-17","title":{"zh":"AlphaGeometry 达到奥赛几何接近金牌水平（Nature）","en":"AlphaGeometry solves olympiad geometry near gold level (Nature)"},"url":"https://www.nature.com/articles/s41586-023-06747-5","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"奥赛几何接近金牌，但属于竞赛题而非研究问题","en":"Near-gold olympiad geometry; competition, not research"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"奥赛几何接近金牌，但属于竞赛题而非研究问题","en":"Near-gold olympiad geometry; competition, not research"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"奥赛几何接近金牌，但属于竞赛题而非研究问题","en":"Near-gold olympiad geometry; competition, not research"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-023-06747-5","title":"Solving olympiad geometry without human demonstrations","venue":"Nature","date":"2024-01-17"},"addedIn":"v1"},{"date":"2024-07-25","title":{"zh":"AlphaProof + AlphaGeometry 2 在 IMO 2024 达银牌分数","en":"AlphaProof + AlphaGeometry 2 reach IMO 2024 silver-medal score"},"url":"https://deepmind.google/discover/blog/ai-solves-imo-problems-at-silver-medal-level/","source":"Google DeepMind","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"competition","typeLabel":{"zh":"竞赛","en":"Competition"},"why":{"zh":"首次在 IMO 达到奖牌分数（银牌），但部分题耗时数天","en":"First medal-level IMO score (silver), some problems took days"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首次在 IMO 达到奖牌分数（银牌），但部分题耗时数天","en":"First medal-level IMO score (silver), some problems took days"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"首次在 IMO 达到奖牌分数（银牌），但部分题耗时数天","en":"First medal-level IMO score (silver), some problems took days"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2024-11-07","title":{"zh":"FrontierMath：由数十位数学家出题的研究级数学基准（Epoch AI）","en":"FrontierMath: research-level maths benchmark written by expert mathematicians (Epoch AI)"},"url":"https://arxiv.org/abs/2411.04872","source":"arXiv","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok-crossref","datePrecision":"day","type":"benchmark","typeLabel":{"zh":"基准","en":"Benchmark"},"why":{"zh":"发布时前沿模型正确率不足 2%，成为衡量 AI 数学研究能力的主要标尺","en":"Frontier models solved <2% at release; became the main yardstick for research-level AI maths"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"发布时前沿模型正确率不足 2%，成为衡量 AI 数学研究能力的主要标尺","en":"Frontier models solved <2% at release; became the main yardstick for research-level AI maths"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"发布时前沿模型正确率不足 2%，成为衡量 AI 数学研究能力的主要标尺","en":"Frontier models solved <2% at release; became the main yardstick for research-level AI maths"}},"impactNotes":[],"verified":{"id":"arXiv:2411.04872","title":"FrontierMath: A Benchmark for Evaluating Advanced Mathematical Reasoning in AI","venue":"arXiv","date":"2024-11-07"},"addedIn":"v2"},{"date":"2025-05-14","title":{"zh":"AlphaEvolve 改进 4×4 复矩阵乘法等数十个问题","en":"AlphaEvolve improves 4×4 complex matrix multiplication and dozens of problems"},"url":"https://deepmind.google/discover/blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/","source":"Google DeepMind","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"system","typeLabel":{"zh":"系统发布","en":"System"},"why":{"zh":"4×4 复矩阵乘法用 48 次乘法，改进 Strassen 1969 年以来的结果；数十个问题追平或超过已知最佳","en":"48-multiplication 4×4 complex matrix product improves on Strassen (1969)"},"aiImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"4×4 复矩阵乘法用 48 次乘法，改进 Strassen 1969 年以来的结果；数十个问题追平或超过已知最佳","en":"48-multiplication 4×4 complex matrix product improves on Strassen (1969)"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"4×4 复矩阵乘法用 48 次乘法，改进 Strassen 1969 年以来的结果；数十个问题追平或超过已知最佳","en":"48-multiplication 4×4 complex matrix product improves on Strassen (1969)"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2025-07-21","title":{"zh":"Gemini Deep Think 获 IMO 2025 官方评分金牌（35/42）","en":"Gemini Deep Think officially graded gold at IMO 2025 (35/42)"},"url":"https://deepmind.google/discover/blog/advanced-version-of-gemini-with-deep-think-officially-achieves-gold-medal-standard-at-the-international-mathematical-olympiad/","source":"Google DeepMind","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"competition","typeLabel":{"zh":"竞赛","en":"Competition"},"why":{"zh":"首个经 IMO 官方评分的 AI 金牌（35/42），用自然语言作答","en":"First AI gold officially graded by the IMO (35/42)"},"aiImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"首个经 IMO 官方评分的 AI 金牌（35/42），用自然语言作答","en":"First AI gold officially graded by the IMO (35/42)"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"首个经 IMO 官方评分的 AI 金牌（35/42），用自然语言作答","en":"First AI gold officially graded by the IMO (35/42)"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2025-11-03","title":{"zh":"AlphaEvolve 数学探索论文：67 个问题，多处改进已知最佳构造，11 维接吻数下界 592→593（arXiv 2511.02864）","en":"AlphaEvolve at scale: 67 problems, several best-known constructions improved; 11-D kissing number 592→593"},"url":"https://arxiv.org/abs/2511.02864","source":"arXiv（Georgiev、Gómez-Serrano、Tao、Wagner）","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"由 Tao 等人联合署名，覆盖分析、组合、几何、数论，把 AI 构造搜索变成规模化方法","en":"Co-authored by Terence Tao; turns AI-driven construction search into a scalable method across fields"},"aiImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"由 Tao 等人联合署名，覆盖分析、组合、几何、数论，把 AI 构造搜索变成规模化方法","en":"Co-authored by Terence Tao; turns AI-driven construction search into a scalable method across fields"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"由 Tao 等人联合署名，覆盖分析、组合、几何、数论，把 AI 构造搜索变成规模化方法","en":"Co-authored by Terence Tao; turns AI-driven construction search into a scalable method across fields"}},"impactNotes":[{"zh":"预印本已于 2026-09-30 经同行评审发表于 PNAS","en":"Peer-reviewed version published in PNAS on 2026-09-30"}],"verified":{"id":"arXiv:2511.02864","title":"Mathematical exploration and discovery at scale","venue":"arXiv","date":"2025-11-03"},"addedIn":"v2"},{"date":"2025-11-12","title":{"zh":"AlphaProof 论文：强化学习实现奥赛级形式化数学推理（Nature）","en":"AlphaProof paper: olympiad-level formal reasoning with RL (Nature)"},"url":"https://www.nature.com/articles/s41586-025-09833-y","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"AlphaProof 方法经同行评审发表，证明强化学习 + Lean 可行","en":"Peer-reviewed AlphaProof method (RL + Lean)"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AlphaProof 方法经同行评审发表，证明强化学习 + Lean 可行","en":"Peer-reviewed AlphaProof method (RL + Lean)"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AlphaProof 方法经同行评审发表，证明强化学习 + Lean 可行","en":"Peer-reviewed AlphaProof method (RL + Lean)"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-025-09833-y","title":"Olympiad-level formal mathematical reasoning with reinforcement learning","venue":"Nature","date":"2025-11-12"},"addedIn":"v1"},{"date":"2026-05-20","title":{"zh":"单位距离猜想被推翻：OpenAI 通用推理模型构造出 n^{1+ε} 个单位距离的点集，否定 Erdős 1946 年的 n^{1+o(1)} 猜想","en":"Unit-distance conjecture disproved: an OpenAI general reasoning model constructs point sets with n^{1+ε} unit distances"},"url":"https://openai.com/index/model-disproves-discrete-geometry-conjecture/","source":"OpenAI","flags":["vendor"],"flagLabels":[{"zh":"厂商自述","en":"Vendor claim"}],"linkCheck":"site-blocks-bots","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"悬而未决约 80 年、被称为组合几何“最著名的问题”；证明经 Alon、Gowers、Sawin 等 9 位外部数学家核验并写成人类可读版本（arXiv 2605.20695），Sawin 将 ε 改进到 0.014","en":"~80-year-old flagship problem; proof checked and rewritten by nine outside mathematicians (arXiv 2605.20695); Sawin improved ε to 0.014"},"aiImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"悬而未决约 80 年、被称为组合几何“最著名的问题”；证明经 Alon、Gowers、Sawin 等 9 位外部数学家核验并写成人类可读版本（arXiv 2605.20695），Sawin 将 ε 改进到 0.014","en":"~80-year-old flagship problem; proof checked and rewritten by nine outside mathematicians (arXiv 2605.20695); Sawin improved ε to 0.014"}},"fieldImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"悬而未决约 80 年、被称为组合几何“最著名的问题”；证明经 Alon、Gowers、Sawin 等 9 位外部数学家核验并写成人类可读版本（arXiv 2605.20695），Sawin 将 ε 改进到 0.014","en":"~80-year-old flagship problem; proof checked and rewritten by nine outside mathematicians (arXiv 2605.20695); Sawin improved ε to 0.014"}},"impactNotes":[{"zh":"OpenAI 发布，但证明已由 Alon、Gowers、Sawin 等 9 位外部数学家独立核验并公开（arXiv 2605.20695）","en":"Announced by OpenAI, but independently verified by nine outside mathematicians (arXiv 2605.20695)"}],"verified":null,"addedIn":"v2"},{"date":"2026-05-20","title":{"zh":"《关于单位距离猜想被推翻的评注》：9 位数学家给出人类核验、简化和推广的证明（arXiv 2605.20695）","en":"“Remarks on the disproof of the unit distance conjecture”: human-verified, simplified proof by nine mathematicians"},"url":"https://arxiv.org/abs/2605.20695","source":"arXiv","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"独立专家核验是把 AI 结果算作数学成果的关键一步；文中指出论证依赖 Ellenberg–Venkatesh、Golod–Shafarevich 等思想","en":"Independent expert verification is what makes the AI result count as mathematics"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"独立专家核验是把 AI 结果算作数学成果的关键一步；文中指出论证依赖 Ellenberg–Venkatesh、Golod–Shafarevich 等思想","en":"Independent expert verification is what makes the AI result count as mathematics"}},"fieldImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"独立专家核验是把 AI 结果算作数学成果的关键一步；文中指出论证依赖 Ellenberg–Venkatesh、Golod–Shafarevich 等思想","en":"Independent expert verification is what makes the AI result count as mathematics"}},"impactNotes":[{"zh":"预印本，但作者为 9 位独立数学家，本身即核验工作","en":"Preprint, but it is itself the independent verification by nine mathematicians"}],"verified":{"id":"arXiv:2605.20695","title":"Remarks on the disproof of the unit distance conjecture","venue":"arXiv","date":"2026-05-20"},"addedIn":"v2"},{"date":"2026-05-21","title":{"zh":"DeepMind 形式化证明搜索智能体在 Lean 中自主解决 353 个公开 Erdős 问题中的 9 个、44/492 个 OEIS 猜想（arXiv 2605.22763）","en":"DeepMind formal proof-search agent autonomously resolves 9 of 353 open Erdős problems and 44/492 OEIS conjectures in Lean"},"url":"https://arxiv.org/abs/2605.22763","source":"arXiv","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"首次大规模评估 AI 在 Lean 中求解公开问题，结果可机器核验，单题成本几百美元","en":"First large-scale evaluation on open problems with machine-checked Lean proofs at a few hundred dollars per problem"},"aiImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"首次大规模评估 AI 在 Lean 中求解公开问题，结果可机器核验，单题成本几百美元","en":"First large-scale evaluation on open problems with machine-checked Lean proofs at a few hundred dollars per problem"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首次大规模评估 AI 在 Lean 中求解公开问题，结果可机器核验，单题成本几百美元","en":"First large-scale evaluation on open problems with machine-checked Lean proofs at a few hundred dollars per problem"}},"impactNotes":[{"zh":"预印本，但所有证明均经 Lean 机器核验","en":"Preprint, but all proofs are machine-checked in Lean"}],"verified":{"id":"arXiv:2605.22763","title":"Advancing Mathematics Research with AI-Driven Formal Proof Search","venue":"arXiv","date":"2026-05-21"},"addedIn":"v2"},{"date":"2026-05-22","title":{"zh":"PNAS：解决 1000 美元 Erdős 奖金问题，Lean 证明由 ChatGPT 生成","en":"PNAS: $1,000 Erdős prize problem resolved, Lean proof generated by ChatGPT"},"url":"https://www.pnas.org/doi/10.1073/pnas.2531760123","source":"PNAS","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"Erdős 奖金问题解答经 PNAS 同行评审发表，Lean 证明由 ChatGPT 生成","en":"Peer-reviewed solution to an Erdős prize problem with a ChatGPT-generated Lean proof"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"Erdős 奖金问题解答经 PNAS 同行评审发表，Lean 证明由 ChatGPT 生成","en":"Peer-reviewed solution to an Erdős prize problem with a ChatGPT-generated Lean proof"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"Erdős 奖金问题解答经 PNAS 同行评审发表，Lean 证明由 ChatGPT 生成","en":"Peer-reviewed solution to an Erdős prize problem with a ChatGPT-generated Lean proof"}},"impactNotes":[],"verified":{"id":"10.1073/pnas.2531760123","title":"Forbidden Sidon subsets of perfect difference sets, featuring a human-assisted proof","venue":"Proceedings of the National Academy of Sciences","date":"2026-05-22"},"addedIn":"v1"},{"date":"2026-07-16","title":{"zh":"IMO 2026：AxiomProver 以 Lean 形式化证明拿满分 42/42","en":"IMO 2026: AxiomProver scores 42/42 with Lean proofs"},"url":"https://github.com/AxiomMath/IMO2026","source":"Axiom Math","flags":["vendor"],"flagLabels":[{"zh":"厂商自述","en":"Vendor claim"}],"linkCheck":"ok","datePrecision":"day","type":"competition","typeLabel":{"zh":"竞赛","en":"Competition"},"why":{"zh":"IMO 2026 满分并公开全部 Lean 证明，可机器核验；但 IMO 官方成绩不含 AI","en":"Perfect IMO 2026 score with public Lean proofs; IMO results exclude AI"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"IMO 2026 满分并公开全部 Lean 证明，可机器核验；但 IMO 官方成绩不含 AI","en":"Perfect IMO 2026 score with public Lean proofs; IMO results exclude AI"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"IMO 2026 满分并公开全部 Lean 证明，可机器核验；但 IMO 官方成绩不含 AI","en":"Perfect IMO 2026 score with public Lean proofs; IMO results exclude AI"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-07-16","title":{"zh":"IMO 2026：dots-note-3.0 称经组委会评分获满分 42/42","en":"IMO 2026: dots-note-3.0 reports committee-graded 42/42"},"url":"https://studio.dots.ai/dots/imo-en.html","source":"dots.ai","flags":["vendor"],"flagLabels":[{"zh":"厂商自述","en":"Vendor claim"}],"linkCheck":"ok","datePrecision":"day","type":"competition","typeLabel":{"zh":"竞赛","en":"Competition"},"why":{"zh":"厂商称满分，未公开可核验材料","en":"Vendor claim only"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"厂商称满分，未公开可核验材料","en":"Vendor claim only"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"厂商称满分，未公开可核验材料","en":"Vendor claim only"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-09-30","title":{"zh":"PNAS 正式发表 AlphaEvolve 数学发现论文（Tao 等）","en":"PNAS publishes the AlphaEvolve mathematical-discovery paper (Tao et al.)"},"url":"https://www.pnas.org/doi/10.1073/pnas.2536158123","source":"PNAS","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"规模化 AI 数学发现首次通过顶刊同行评审","en":"Large-scale AI maths discovery passes peer review in a top journal"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"规模化 AI 数学发现首次通过顶刊同行评审","en":"Large-scale AI maths discovery passes peer review in a top journal"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"规模化 AI 数学发现首次通过顶刊同行评审","en":"Large-scale AI maths discovery passes peer review in a top journal"}},"impactNotes":[],"verified":{"id":"10.1073/pnas.2536158123","title":"Mathematical discovery and exploration can be done at scale","venue":"Proceedings of the National Academy of Sciences","date":"2026-09-30"},"addedIn":"v2"},{"date":"2026-10-01","title":{"zh":"Song 与 Zhang 借助人类主导的 AI 智能体证明 31 年悬而未决的 KLS 猜想（各向同性对数凹测度的 Cheeger 常数有维度无关下界）；Klartag 等三位专家独立给出证明并称其判据为“关键一步”","en":"Song & Zhang prove the 31-year-old Kannan–Lovász–Simonovits conjecture with human-directed AI agents (arXiv 2610.01447); Bizeul–Klartag–Lehec give an independent proof built on their criterion (arXiv 2610.05474)"},"url":"https://arxiv.org/abs/2610.01447","source":"arXiv","flags":["preprint","vendor"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"},{"zh":"厂商自述","en":"Vendor claim"}],"linkCheck":"ok","datePrecision":"day","type":"discovery","typeLabel":{"zh":"科学发现","en":"Discovery"},"why":{"zh":"AI：微软研究院称研究者用人类主导的智能体流程探索了 100 多种思路，由人筛选并完成论证（AI 的具体贡献为微软自述，未经独立核实）；学科：KLS 是高维凸几何的核心公开问题（1995 年提出），Bizeul、Klartag、Lehec 于 10-04 独立给出证明（arXiv 2610.05474），Balasubramanian 等另给出显式常数证明（arXiv 2610.07728），并有 Lean 4 形式化；均为预印本，尚未经期刊审稿","en":"AI: per Microsoft Research a human-directed agentic process explored 100+ approaches (AI role is self-reported). Field: KLS (1995) is a central problem of high-dimensional convex geometry; Bizeul–Klartag–Lehec independently proved it (arXiv 2610.05474), another explicit-constant proof appeared (arXiv 2610.07728), plus a Lean 4 formalisation; all preprints"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：微软研究院称研究者用人类主导的智能体流程探索了 100 多种思路，由人筛选并完成论证（AI 的具体贡献为微软自述，未经独立核实）；学科：KLS 是高维凸几何的核心公开问题（1995 年提出），Bizeul、Klartag、Lehec 于 10-04 独立给出证明（arXiv 2610.05474），Balasubramanian 等另给出显式常数证明（arXiv 2610.07728），并有 Lean 4 形式化；均为预印本，尚未经期刊审稿","en":"AI: per Microsoft Research a human-directed agentic process explored 100+ approaches (AI role is self-reported). Field: KLS (1995) is a central problem of high-dimensional convex geometry; Bizeul–Klartag–Lehec independently proved it (arXiv 2610.05474), another explicit-constant proof appeared (arXiv 2610.07728), plus a Lean 4 formalisation; all preprints"}},"fieldImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"AI：微软研究院称研究者用人类主导的智能体流程探索了 100 多种思路，由人筛选并完成论证（AI 的具体贡献为微软自述，未经独立核实）；学科：KLS 是高维凸几何的核心公开问题（1995 年提出），Bizeul、Klartag、Lehec 于 10-04 独立给出证明（arXiv 2610.05474），Balasubramanian 等另给出显式常数证明（arXiv 2610.07728），并有 Lean 4 形式化；均为预印本，尚未经期刊审稿","en":"AI: per Microsoft Research a human-directed agentic process explored 100+ approaches (AI role is self-reported). Field: KLS (1995) is a central problem of high-dimensional convex geometry; Bizeul–Klartag–Lehec independently proved it (arXiv 2610.05474), another explicit-constant proof appeared (arXiv 2610.07728), plus a Lean 4 formalisation; all preprints"}},"impactNotes":[{"zh":"预印本，但数学结论已由 Bizeul、Klartag、Lehec 独立给出证明并称 Song–Zhang 判据为“关键一步”（arXiv 2610.05474），另有独立显式常数证明（arXiv 2610.07728）与 Lean 4 形式化；AI 作用为微软自述，故 AI 影响只评“重要进展”","en":"Preprint, but independently re-proved by Bizeul–Klartag–Lehec, who call the Song–Zhang criterion the crucial step (arXiv 2610.05474), plus another proof (arXiv 2610.07728) and a Lean 4 formalisation; AI role is self-reported, so AI impact is Major"}],"verified":{"id":"arXiv:2610.01447","title":"An $O(1)$ Bound for the KLS Constant","venue":"arXiv","date":"2026-10-01"},"addedIn":"v2-20261022pm"},{"date":"2026-10-06","title":{"zh":"OpenAI 在 GitHub 发布 openai/math：719 篇稿件、372 个结果族、约 42% 主结果 Lean 形式化","en":"OpenAI releases openai/math on GitHub: 719 manuscripts, 372 families, ~42% Lean-formalized"},"url":"https://openai.com/index/sharing-ai-progress-in-mathematics","source":"OpenAI","flags":["vendor"],"flagLabels":[{"zh":"厂商自述","en":"Vendor claim"}],"linkCheck":"site-blocks-bots","datePrecision":"day","type":"dataset","typeLabel":{"zh":"数据集","en":"Dataset"},"why":{"zh":"一次性公开 719 篇 AI 数学稿件、约 42% 主结果附 Lean 证明；但出现撤回，数学界有公开抵制","en":"719 AI maths manuscripts, ~42% Lean-backed; withdrawals and push-back"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"一次性公开 719 篇 AI 数学稿件、约 42% 主结果附 Lean 证明；但出现撤回，数学界有公开抵制","en":"719 AI maths manuscripts, ~42% Lean-backed; withdrawals and push-back"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"一次性公开 719 篇 AI 数学稿件、约 42% 主结果附 Lean 证明；但出现撤回，数学界有公开抵制","en":"719 AI maths manuscripts, ~42% Lean-backed; withdrawals and push-back"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-10-07","title":{"zh":"openai/math 因符号错误撤回 3 篇（Hodge 猜想相关），修订 14 篇","en":"openai/math withdraws 3 manuscripts over a sign error and revises 14"},"url":"https://raw.githubusercontent.com/openai/math/main/history.md","source":"openai/math history","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"dataset","typeLabel":{"zh":"数据集","en":"Dataset"},"why":{"zh":"发布次日撤回 3 篇、修订 14 篇，提示质量审查仍不足","en":"3 withdrawn and 14 revised the next day"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"发布次日撤回 3 篇、修订 14 篇，提示质量审查仍不足","en":"3 withdrawn and 14 revised the next day"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"发布次日撤回 3 篇、修订 14 篇，提示质量审查仍不足","en":"3 withdrawn and 14 revised the next day"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-10-11","title":{"zh":"Banks 逐行核验并重写 OpenAI 的“朗道–西格尔零点不存在”证明，给出显式常数（q₀=10¹⁰，δ₀=3.0×10⁻³）","en":"Banks: human-verified, explicit account of the OpenAI proof that Landau–Siegel zeros do not exist (arXiv 2610.15483)"},"url":"https://arxiv.org/abs/2610.15483","source":"arXiv","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"AI：OpenAI 模型给出解析数论近百年核心难题的证明（2026-10-01 手稿），密苏里大学 Banks 逐行核验并称论证成立，Kerr 另给出更简证明（arXiv 2610.21099），Lean 形式化经两个独立内核复核；学科：西格尔零点阻碍算术级数素数分布与类数的有效估计，若最终确认将改写多个经典结果（尚未经期刊审稿，陶哲轩称仍在消化）","en":"AI: an OpenAI model produced a proof of a near-century-old central problem; Banks verified it line by line, Kerr gave a simpler proof (arXiv 2610.21099), and the Lean formalisation was re-checked by two kernels. Field: Siegel zeros block effective bounds for primes in progressions and class numbers (not yet journal-refereed)"},"aiImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"AI：OpenAI 模型给出解析数论近百年核心难题的证明（2026-10-01 手稿），密苏里大学 Banks 逐行核验并称论证成立，Kerr 另给出更简证明（arXiv 2610.21099），Lean 形式化经两个独立内核复核；学科：西格尔零点阻碍算术级数素数分布与类数的有效估计，若最终确认将改写多个经典结果（尚未经期刊审稿，陶哲轩称仍在消化）","en":"AI: an OpenAI model produced a proof of a near-century-old central problem; Banks verified it line by line, Kerr gave a simpler proof (arXiv 2610.21099), and the Lean formalisation was re-checked by two kernels. Field: Siegel zeros block effective bounds for primes in progressions and class numbers (not yet journal-refereed)"}},"fieldImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"AI：OpenAI 模型给出解析数论近百年核心难题的证明（2026-10-01 手稿），密苏里大学 Banks 逐行核验并称论证成立，Kerr 另给出更简证明（arXiv 2610.21099），Lean 形式化经两个独立内核复核；学科：西格尔零点阻碍算术级数素数分布与类数的有效估计，若最终确认将改写多个经典结果（尚未经期刊审稿，陶哲轩称仍在消化）","en":"AI: an OpenAI model produced a proof of a near-century-old central problem; Banks verified it line by line, Kerr gave a simpler proof (arXiv 2610.21099), and the Lean formalisation was re-checked by two kernels. Field: Siegel zeros block effective bounds for primes in progressions and class numbers (not yet journal-refereed)"}},"impactNotes":[{"zh":"预印本，但已由密苏里大学 Banks 逐行核验（本文即核验工作），Kerr 独立给出更简证明（arXiv 2610.21099），OpenAI 的 Lean 形式化经两个独立内核复核；尚未经期刊审稿","en":"Preprint, but it is itself a line-by-line verification by Banks; Kerr gave an independent simpler proof (arXiv 2610.21099) and the Lean formalisation was re-checked by two kernels; not yet journal-refereed"}],"verified":{"id":"arXiv:2610.15483","title":"On the OpenAI proof that Landau-Siegel zeros do not exist","venue":"arXiv","date":"2026-10-11"},"addedIn":"v2"},{"date":"2026-10-15","title":{"zh":"Simons 基金会启动 1500 万美元计划，帮助数学家应对 AI 时代","en":"Simons Foundation launches a $15M effort to help mathematicians navigate the AI era"},"url":"https://terrytao.wordpress.com/2026/10/15/simons-foundation-launches-15m-effort-to-help-mathematicians-navigate-ai-era/","source":"Terence Tao 博客","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"policy","typeLabel":{"zh":"政策","en":"Policy"},"why":{"zh":"主要基础研究资助方首次专门投入资源审读、整合 AI 产出的数学","en":"A major funder dedicates money to vetting and absorbing AI-produced maths"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"主要基础研究资助方首次专门投入资源审读、整合 AI 产出的数学","en":"A major funder dedicates money to vetting and absorbing AI-produced maths"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"主要基础研究资助方首次专门投入资源审读、整合 AI 产出的数学","en":"A major funder dedicates money to vetting and absorbing AI-produced maths"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-10-19","title":{"zh":"Quanta：数论专家基本认可 OpenAI 的“准黎曼猜想”证明（零点实部在 1/12 与 11/12 之间），Sarnak 改课专门研读","en":"Quanta: number theorists largely accept OpenAI’s proof of a quasi-Riemann hypothesis (zeros between 1/12 and 11/12)"},"url":"https://www.quantamagazine.org/ai-proved-the-quasi-riemann-hypothesis-what-happens-now-20261019/","source":"Quanta Magazine","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok","datePrecision":"day","type":"discovery","typeLabel":{"zh":"科学发现","en":"Discovery"},"why":{"zh":"AI：Stanford 的 Lichtman 称其为“迄今最重要的 AI 数学成果”，证明综合了被忽视的亚辛形式等旧工具；学科：Radziwiłł 称黎曼猜想的大部分推论已可由准黎曼猜想推出。证明仍是预印本，人工整理稿尚未发表","en":"AI: called “the most significant AI-for-math achievement so far” by Lichtman; Field: per Radziwiłł most consequences of RH already follow from quasi-RH. Still a preprint; human write-ups unpublished"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：Stanford 的 Lichtman 称其为“迄今最重要的 AI 数学成果”，证明综合了被忽视的亚辛形式等旧工具；学科：Radziwiłł 称黎曼猜想的大部分推论已可由准黎曼猜想推出。证明仍是预印本，人工整理稿尚未发表","en":"AI: called “the most significant AI-for-math achievement so far” by Lichtman; Field: per Radziwiłł most consequences of RH already follow from quasi-RH. Still a preprint; human write-ups unpublished"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：Stanford 的 Lichtman 称其为“迄今最重要的 AI 数学成果”，证明综合了被忽视的亚辛形式等旧工具；学科：Radziwiłł 称黎曼猜想的大部分推论已可由准黎曼猜想推出。证明仍是预印本，人工整理稿尚未发表","en":"AI: called “the most significant AI-for-math achievement so far” by Lichtman; Field: per Radziwiłł most consequences of RH already follow from quasi-RH. Still a preprint; human write-ups unpublished"}},"impactNotes":[{"zh":"AI 影响原评“飞跃性突破”，因预印本，未同行评审按规则降为“重要进展”","en":"AI impact downgraded from Leap to Major because: Preprint, not peer-reviewed"},{"zh":"学科影响原评“飞跃性突破”，因预印本，未同行评审按规则降为“重要进展”","en":"Field impact downgraded from Leap to Major because: Preprint, not peer-reviewed"}],"verified":null,"addedIn":"v2"},{"date":"2026-10-20","title":{"zh":"人类数学家借助 AI 证明“双帽猜想”（球面上无正交点对的集合测度上限），主结论另有几乎完全由 AI 完成形式化的 Lean 4 证明","en":"AI-assisted proof of the double-cap conjecture, with a separate Lean 4 proof formalised almost entirely using AI tools (arXiv 2610.27523)"},"url":"https://arxiv.org/abs/2610.27523","source":"arXiv","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"AI：作者在摘要中说明 AI 协助撰写论文，主不等式和双帽结论的另一份 Lean 4 证明几乎完全由 AI 工具完成形式化与验证；学科：双帽猜想是球面上“无正交对集合”测度的长期公开问题，论文同时确定球面正交图的可测分数色数（预印本，尚未审稿）","en":"AI: per the abstract, AI assisted the paper and a separate Lean 4 proof of the main results was formalised almost entirely with AI tools. Field: settles the long-open double-cap conjecture and the measurable fractional chromatic number of the spherical orthogonality graph (preprint)"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：作者在摘要中说明 AI 协助撰写论文，主不等式和双帽结论的另一份 Lean 4 证明几乎完全由 AI 工具完成形式化与验证；学科：双帽猜想是球面上“无正交对集合”测度的长期公开问题，论文同时确定球面正交图的可测分数色数（预印本，尚未审稿）","en":"AI: per the abstract, AI assisted the paper and a separate Lean 4 proof of the main results was formalised almost entirely with AI tools. Field: settles the long-open double-cap conjecture and the measurable fractional chromatic number of the spherical orthogonality graph (preprint)"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：作者在摘要中说明 AI 协助撰写论文，主不等式和双帽结论的另一份 Lean 4 证明几乎完全由 AI 工具完成形式化与验证；学科：双帽猜想是球面上“无正交对集合”测度的长期公开问题，论文同时确定球面正交图的可测分数色数（预印本，尚未审稿）","en":"AI: per the abstract, AI assisted the paper and a separate Lean 4 proof of the main results was formalised almost entirely with AI tools. Field: settles the long-open double-cap conjecture and the measurable fractional chromatic number of the spherical orthogonality graph (preprint)"}},"impactNotes":[{"zh":"学科影响原评“飞跃性突破”，因预印本，未同行评审按规则降为“重要进展”","en":"Field impact downgraded from Leap to Major because: Preprint, not peer-reviewed"}],"verified":{"id":"arXiv:2610.27523","title":"An orthogonality inequality in $\\mathbb{RP}^{n-1}$ and the double-cap conjecture","venue":"arXiv","date":"2026-10-20"},"addedIn":"v2"}],"papersRepos":[{"title":"openai/math — 719 篇 AI 数学稿件 + Lean 形式化","url":"https://github.com/openai/math","date":"2026-10-06","kind":"repo","linkCheck":"ok"},{"title":"AxiomMath/IMO2026 — IMO 2026 六题 Lean 证明","url":"https://github.com/AxiomMath/IMO2026","date":"2026-07","kind":"repo","linkCheck":"ok"},{"title":"leanprover-community/mathlib4","url":"https://github.com/leanprover-community/mathlib4","date":null,"kind":"repo","linkCheck":"ok"},{"title":"google-deepmind/formal-conjectures","url":"https://github.com/google-deepmind/formal-conjectures","date":null,"kind":"repo","linkCheck":"ok"},{"title":"teorth/erdosproblems（Erdős 问题与 AI 贡献记录）","url":"https://github.com/teorth/erdosproblems","date":null,"kind":"repo","linkCheck":"ok"},{"title":"NanoProof: Open and Efficient Automated Theorem Proving in Lean 4 (arXiv 2610.11605)","url":"https://arxiv.org/abs/2610.11605","date":"2026-10-08","kind":"paper","linkCheck":"ok"},{"title":"An AI-Assisted Formalization of the Poincaré Conjecture (arXiv 2610.08329)","url":"https://arxiv.org/abs/2610.08329","date":"2026-10-06","kind":"paper","linkCheck":"ok"},{"title":"Mathematical discovery and exploration can be done at scale (PNAS, AlphaEvolve 67 problems)","url":"https://www.pnas.org/doi/10.1073/pnas.2536158123","date":"2026-09-30","kind":"paper","linkCheck":"ok-crossref"},{"title":"Mathematical discoveries from program search with large language models","url":"https://doi.org/10.1038/s41586-023-06924-6","date":"2023-12-14","kind":"paper","linkCheck":"ok-crossref"},{"title":"Mathematical exploration and discovery at scale","url":"https://arxiv.org/abs/2511.02864","date":"2025-11-03","kind":"paper","linkCheck":"ok-crossref"},{"title":"Remarks on the disproof of the unit distance conjecture","url":"https://arxiv.org/abs/2605.20695","date":"2026-05-20","kind":"paper","linkCheck":"ok-crossref"},{"title":"Advancing Mathematics Research with AI-Driven Formal Proof Search","url":"https://arxiv.org/abs/2605.22763","date":"2026-05-21","kind":"paper","linkCheck":"ok-crossref"},{"title":"On the OpenAI proof that Landau-Siegel zeros do not exist","url":"https://arxiv.org/abs/2610.15483","date":"2026-10-11","kind":"paper","linkCheck":"ok-crossref"},{"title":"Generating conjectures on fundamental constants with the Ramanujan Machine","url":"https://doi.org/10.1038/s41586-021-03229-4","date":"2021-02-03","kind":"paper","linkCheck":"ok-crossref"},{"title":"Advancing mathematics by guiding human intuition with AI","url":"https://doi.org/10.1038/s41586-021-04086-x","date":"2021-12-01","kind":"paper","linkCheck":"ok-crossref"},{"title":"Solving olympiad geometry without human demonstrations","url":"https://doi.org/10.1038/s41586-023-06747-5","date":"2024-01-17","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"OpenAI","url":"https://openai.com"},{"name":"Google DeepMind","url":"https://deepmind.google"},{"name":"Axiom Math","url":"https://github.com/AxiomMath"},{"name":"Lean community / mathlib","url":"https://leanprover-community.github.io/"},{"name":"Erdős Problems","url":"https://www.erdosproblems.com/"},{"name":"Epoch AI FrontierMath","url":"https://epoch.ai/frontiermath"}],"news":[{"id":"653ef197344c","date":"2026-10-20","title":"The crisis of AI-generated mathematics","url":"https://arxiv.org/abs/2608.02859","source":"arXiv math.HO 数学史与综述","lang":"en","kind":"paper","summaryZh":"arXiv math.HO 数学史与综述于2026-10-20发布的论文，涉及AI 数学，原题《The crisis of AI-generated mathematics》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"0bcb294a7eb6","date":"2026-10-20","title":"Studying proof data with proof theory: the case of AI-generated mathematics","url":"https://arxiv.org/abs/2610.25357","source":"arXiv cs.LO 逻辑与形式化","lang":"en","kind":"paper","summaryZh":"arXiv cs.LO 逻辑与形式化于2026-10-20发布的论文，涉及AI 数学，原题《Studying proof data with proof theory: the case of AI-generated mathematics》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"280083332a1c","date":"2026-10-20","title":"Formalization of R\\'{e}nyi Differential Privacy in Isabelle/HOL (Draft version)","url":"https://arxiv.org/abs/2610.25636","source":"arXiv cs.LO 逻辑与形式化","lang":"en","kind":"paper","summaryZh":"arXiv cs.LO 逻辑与形式化于2026-10-20发布的论文，涉及AI 数学、AI 科研与调研、AI 商业融资，原题《Formalization of R\\'{e}nyi Differential Privacy in Isabelle/HOL (Draft version)》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"13b1d9a2a65c","date":"2026-10-15","title":"Simons Foundation Launches $15M Effort To Help Mathematicians Navigate AI Era","url":"https://terrytao.wordpress.com/2026/10/15/simons-foundation-launches-15m-effort-to-help-mathematicians-navigate-ai-era/","source":"Terence Tao 博客","lang":"en","kind":"news","summaryZh":"Terence Tao 博客于2026-10-16发布的文章，涉及AI 数学、AI 软件开发，原题《Simons Foundation Launches $15M Effort To Help Mathematicians Navigate AI Era》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"4ea5c42c27d6","date":"2026-10-09","title":"Is this the 'mathocalypse'? Why OpenAI's latest results dump has left mathematicians in shock","url":"https://phys.org/news/2026-10-mathocalypse-openai-latest-results-dump.html","source":"Phys.org Mathematics","lang":"en","kind":"news","summaryZh":"Phys.org 解读 OpenAI 本周发布的 722 篇数学成果为何让数学家震惊，以及“mathocalypse”说法的由来。","via":"research-samples-20261010","linkCheck":"from-feed"},{"id":"309257414599","date":"2026-10-08","title":"OpenAI withdraws three preprints a day after releasing 722 manuscripts on unsolved math problems","url":"https://retractionwatch.com/2026/10/08/openai-withdraws-preprints-722-manuscripts-unsolved-math-problems/","source":"Retraction Watch","lang":"en","kind":"news","summaryZh":"OpenAI 在 GitHub 上公开 722 篇 AI 生成的数学预印本后不到 24 小时，就因错误撤回其中 3 篇（内容涉数学，与数学组可能重复）。","via":"research-samples-20261010","linkCheck":"from-feed"},{"id":"a47407dd071c","date":"2026-10-08","title":"AHM Statement on OpenAI’s October 6 Release of Mathematical Documents","url":"https://terrytao.wordpress.com/2026/10/07/ahm-statement-on-openais-october-6-release-of-mathematical-documents/","source":"Terence Tao blog","lang":"en","kind":"news","summaryZh":"陶哲轩博客转载“人类数学协会”(AHM)对 OpenAI 10 月 6 日数学文档发布的正式声明。","via":"research-samples-20261010","linkCheck":"from-feed"},{"id":"47b9255d8fa0","date":"2026-10-06","title":"Sharing AI progress in mathematics","url":"https://openai.com/index/sharing-ai-progress-in-mathematics","source":"OpenAI News","lang":"en","kind":"news","summaryZh":"OpenAI 发布内部前沿模型在数学开放问题上的一批新结果，并在 GitHub 公开 Lean 形式化证明与研究细节（厂商自述）。","via":"research-samples-20261010","linkCheck":"from-feed"},{"id":"dadc7cecbc5a","date":"2026-09-30","title":"Mathematical discovery and exploration can be done at scale (PNAS)","url":"https://www.pnas.org/doi/10.1073/pnas.2536158123","source":"PNAS","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"ok-crossref"},{"id":"13c4d4b509e2","date":"2026-05-22","title":"AI cracks 80-year-old mathematics challenge — researchers are astonished","url":"https://www.nature.com/articles/d41586-026-01651-0","source":"Nature News","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"ok"},{"id":"2983743315aa","date":"2026-05-21","title":"AI just solved an 80-year-old ‘Erdős problem,’ and mathematicians are amazed","url":"https://www.scientificamerican.com/article/ai-just-solved-an-80-year-old-erdos-problem-and-mathematicians-are-amazed/","source":"Scientific American","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"ok"},{"id":"7f625f666338","date":"2026-05-20","title":"Model disproves discrete geometry conjecture","url":"https://openai.com/index/model-disproves-discrete-geometry-conjecture/","source":"OpenAI","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"site-blocks-bots"}],"review":{"lastReviewed":"2026-10-22","cycleDays":60,"nextReview":"2026-12-20"},"urls":{"zh":"/zh-hans/disciplines/mathematics/","en":"/en/disciplines/mathematics/"},"dimensions":{"method":{"score":3,"label":{"zh":"研究方法","en":"Research method"},"evidence":"Lean 形式化 + 大模型推理已成为一线数学家（Tao、Gowers 等）公开使用的研究手段，但多数数学家尚未日常使用","source":"https://github.com/teorth/erdosproblems/wiki/AI-contributions-to-Erd%C5%91s-problems","weight":0.3},"discovery":{"score":4,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"AI 独立产出并经人类专家或 Lean 核验的研究级结果已成批出现：单位距离猜想被推翻（9 位数学家核验）、PNAS 发表 AlphaEvolve 67 题、DeepMind 智能体在 Lean 中解决 9 个 Erdős 公开问题","source":"https://arxiv.org/abs/2605.20695","weight":0.35},"adoption":{"score":2,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"有名家和社区在用，但缺少从业者使用率调查；AHM 等团体公开抵制 openai/math 的发布方式","source":"https://terrytao.wordpress.com/2026/10/07/ahm-statement-on-openais-october-6-release-of-mathematical-documents/","weight":0.25},"education":{"score":1,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"课程层面只有零星的 Lean/AI 研讨课，暂无课程体系改革的公开证据","source":null,"weight":0.1}},"compositeScore":2.9,"levelCapped":false,"levelChange":{"from":4,"to":3,"direction":"down","reason":{"zh":"数学是 AI 科学发现最强的学科：2026-05 OpenAI 模型推翻 1946 年 Erdős 单位距离猜想（Alon、Gowers、Sawin 等 9 位数学家写了人类核验版），PNAS 发表 AlphaEvolve 在 67 个问题上的系统探索，DeepMind 智能体在 Lean 中独立解决 9 个 Erdős 公开问题。但从业者普遍采用和课程改革尚未发生，openai/math 还出现撤回和公开抵制，所以综合定为 L3（发现维度 4 分）。","en":"Maths has the strongest AI-discovery record: in May 2026 an OpenAI model disproved Erdős’s 1946 unit-distance conjecture (a human-verified write-up by nine mathematicians incl. Alon, Gowers, Sawin), PNAS published AlphaEvolve’s 67-problem study, and a DeepMind agent resolved 9 open Erdős problems in Lean. Broad practitioner adoption and curriculum change have not happened, and openai/math saw withdrawals and push-back, so the composite is L3 (discovery scored 4)."}},"biggestBreakthrough":{"date":"2026-10-11","title":{"zh":"Banks 逐行核验并重写 OpenAI 的“朗道–西格尔零点不存在”证明，给出显式常数（q₀=10¹⁰，δ₀=3.0×10⁻³）","en":"Banks: human-verified, explicit account of the OpenAI proof that Landau–Siegel zeros do not exist (arXiv 2610.15483)"},"url":"https://arxiv.org/abs/2610.15483","aiImpact":"leap","fieldImpact":"leap","why":{"zh":"AI：OpenAI 模型给出解析数论近百年核心难题的证明（2026-10-01 手稿），密苏里大学 Banks 逐行核验并称论证成立，Kerr 另给出更简证明（arXiv 2610.21099），Lean 形式化经两个独立内核复核；学科：西格尔零点阻碍算术级数素数分布与类数的有效估计，若最终确认将改写多个经典结果（尚未经期刊审稿，陶哲轩称仍在消化）","en":"AI: an OpenAI model produced a proof of a near-century-old central problem; Banks verified it line by line, Kerr gave a simpler proof (arXiv 2610.21099), and the Lean formalisation was re-checked by two kernels. Field: Siegel zeros block effective bounds for primes in progressions and class numbers (not yet journal-refereed)"}},"keyPapers":[{"title":"Mathematical discoveries from program search with large language models","titleZh":"FunSearch 用大模型找到帽集问题新构造（Nature）","id":"10.1038/s41586-023-06924-6","url":"https://doi.org/10.1038/s41586-023-06924-6","venue":"Nature","date":"2023-12-14","why":{"zh":"LLM 首次在公开数学问题（帽集）上找到超过人类已知最佳的新构造","en":"First LLM-found construction beating best-known human results on an open problem (cap sets)"},"fieldImpact":"major","aiImpact":"leap"},{"title":"Mathematical exploration and discovery at scale","titleZh":"AlphaEvolve 数学探索论文：67 个问题，多处改进已知最佳构造，11 维接吻数下界 592→593（arXiv 2511.02864）","id":"arXiv:2511.02864","url":"https://arxiv.org/abs/2511.02864","venue":"arXiv","date":"2025-11-03","why":{"zh":"由 Tao 等人联合署名，覆盖分析、组合、几何、数论，把 AI 构造搜索变成规模化方法","en":"Co-authored by Terence Tao; turns AI-driven construction search into a scalable method across fields"},"fieldImpact":"major","aiImpact":"leap"},{"title":"Remarks on the disproof of the unit distance conjecture","titleZh":"《关于单位距离猜想被推翻的评注》：9 位数学家给出人类核验、简化和推广的证明（arXiv 2605.20695）","id":"arXiv:2605.20695","url":"https://arxiv.org/abs/2605.20695","venue":"arXiv","date":"2026-05-20","why":{"zh":"独立专家核验是把 AI 结果算作数学成果的关键一步；文中指出论证依赖 Ellenberg–Venkatesh、Golod–Shafarevich 等思想","en":"Independent expert verification is what makes the AI result count as mathematics"},"fieldImpact":"leap","aiImpact":"major"},{"title":"Advancing Mathematics Research with AI-Driven Formal Proof Search","titleZh":"DeepMind 形式化证明搜索智能体在 Lean 中自主解决 353 个公开 Erdős 问题中的 9 个、44/492 个 OEIS 猜想（arXiv 2605.22763）","id":"arXiv:2605.22763","url":"https://arxiv.org/abs/2605.22763","venue":"arXiv","date":"2026-05-21","why":{"zh":"首次大规模评估 AI 在 Lean 中求解公开问题，结果可机器核验，单题成本几百美元","en":"First large-scale evaluation on open problems with machine-checked Lean proofs at a few hundred dollars per problem"},"fieldImpact":"major","aiImpact":"leap"},{"title":"An $O(1)$ Bound for the KLS Constant","titleZh":"Song 与 Zhang 借助人类主导的 AI 智能体证明 31 年悬而未决的 KLS 猜想（各向同性对数凹测度的 Cheeger 常数有维度无关下界）；Klartag 等三位专家独立给出证明并称其判据为“关键一步”","id":"arXiv:2610.01447","url":"https://arxiv.org/abs/2610.01447","venue":"arXiv","date":"2026-10-01","why":{"zh":"AI：微软研究院称研究者用人类主导的智能体流程探索了 100 多种思路，由人筛选并完成论证（AI 的具体贡献为微软自述，未经独立核实）；学科：KLS 是高维凸几何的核心公开问题（1995 年提出），Bizeul、Klartag、Lehec 于 10-04 独立给出证明（arXiv 2610.05474），Balasubramanian 等另给出显式常数证明（arXiv 2610.07728），并有 Lean 4 形式化；均为预印本，尚未经期刊审稿","en":"AI: per Microsoft Research a human-directed agentic process explored 100+ approaches (AI role is self-reported). Field: KLS (1995) is a central problem of high-dimensional convex geometry; Bizeul–Klartag–Lehec independently proved it (arXiv 2610.05474), another explicit-constant proof appeared (arXiv 2610.07728), plus a Lean 4 formalisation; all preprints"},"fieldImpact":"leap","aiImpact":"major"},{"title":"On the OpenAI proof that Landau-Siegel zeros do not exist","titleZh":"Banks 逐行核验并重写 OpenAI 的“朗道–西格尔零点不存在”证明，给出显式常数（q₀=10¹⁰，δ₀=3.0×10⁻³）","id":"arXiv:2610.15483","url":"https://arxiv.org/abs/2610.15483","venue":"arXiv","date":"2026-10-11","why":{"zh":"AI：OpenAI 模型给出解析数论近百年核心难题的证明（2026-10-01 手稿），密苏里大学 Banks 逐行核验并称论证成立，Kerr 另给出更简证明（arXiv 2610.21099），Lean 形式化经两个独立内核复核；学科：西格尔零点阻碍算术级数素数分布与类数的有效估计，若最终确认将改写多个经典结果（尚未经期刊审稿，陶哲轩称仍在消化）","en":"AI: an OpenAI model produced a proof of a near-century-old central problem; Banks verified it line by line, Kerr gave a simpler proof (arXiv 2610.21099), and the Lean formalisation was re-checked by two kernels. Field: Siegel zeros block effective bounds for primes in progressions and class numbers (not yet journal-refereed)"},"fieldImpact":"leap","aiImpact":"leap"},{"title":"Generating conjectures on fundamental constants with the Ramanujan Machine","titleZh":"拉马努金机：算法自动生成基本常数的连分数猜想（Nature）","id":"10.1038/s41586-021-03229-4","url":"https://doi.org/10.1038/s41586-021-03229-4","venue":"Nature","date":"2021-02-03","why":{"zh":"计算机自动提出数学常数公式猜想，部分后被证明","en":"Automated conjecture generation for constants; some later proved"},"fieldImpact":"incremental","aiImpact":"major"},{"title":"Advancing mathematics by guiding human intuition with AI","titleZh":"DeepMind 与数学家合作：机器学习引导直觉发现纽结理论与表示论新结果（Nature）","id":"10.1038/s41586-021-04086-x","url":"https://doi.org/10.1038/s41586-021-04086-x","venue":"Nature","date":"2021-12-01","why":{"zh":"首次在顶刊证明 AI 能帮助数学家提出并证明新定理（纽结不变量关系、Kazhdan–Lusztig 多项式猜想进展）","en":"First top-journal demonstration that AI can help mathematicians conjecture and prove new theorems"},"fieldImpact":"major","aiImpact":"major"},{"title":"Solving olympiad geometry without human demonstrations","titleZh":"AlphaGeometry 达到奥赛几何接近金牌水平（Nature）","id":"10.1038/s41586-023-06747-5","url":"https://doi.org/10.1038/s41586-023-06747-5","venue":"Nature","date":"2024-01-17","why":{"zh":"奥赛几何接近金牌，但属于竞赛题而非研究问题","en":"Near-gold olympiad geometry; competition, not research"},"fieldImpact":"incremental","aiImpact":"major"},{"title":"FrontierMath: A Benchmark for Evaluating Advanced Mathematical Reasoning in AI","titleZh":"FrontierMath：由数十位数学家出题的研究级数学基准（Epoch AI）","id":"arXiv:2411.04872","url":"https://arxiv.org/abs/2411.04872","venue":"arXiv","date":"2024-11-07","why":{"zh":"发布时前沿模型正确率不足 2%，成为衡量 AI 数学研究能力的主要标尺","en":"Frontier models solved <2% at release; became the main yardstick for research-level AI maths"},"fieldImpact":"incremental","aiImpact":"major"},{"title":"Olympiad-level formal mathematical reasoning with reinforcement learning","titleZh":"AlphaProof 论文：强化学习实现奥赛级形式化数学推理（Nature）","id":"10.1038/s41586-025-09833-y","url":"https://doi.org/10.1038/s41586-025-09833-y","venue":"Nature","date":"2025-11-12","why":{"zh":"AlphaProof 方法经同行评审发表，证明强化学习 + Lean 可行","en":"Peer-reviewed AlphaProof method (RL + Lean)"},"fieldImpact":"incremental","aiImpact":"major"},{"title":"Forbidden Sidon subsets of perfect difference sets, featuring a human-assisted proof","titleZh":"PNAS：解决 1000 美元 Erdős 奖金问题，Lean 证明由 ChatGPT 生成","id":"10.1073/pnas.2531760123","url":"https://doi.org/10.1073/pnas.2531760123","venue":"Proceedings of the National Academy of Sciences","date":"2026-05-22","why":{"zh":"Erdős 奖金问题解答经 PNAS 同行评审发表，Lean 证明由 ChatGPT 生成","en":"Peer-reviewed solution to an Erdős prize problem with a ChatGPT-generated Lean proof"},"fieldImpact":"major","aiImpact":"major"},{"title":"Mathematical discovery and exploration can be done at scale","titleZh":"PNAS 正式发表 AlphaEvolve 数学发现论文（Tao 等）","id":"10.1073/pnas.2536158123","url":"https://doi.org/10.1073/pnas.2536158123","venue":"Proceedings of the National Academy of Sciences","date":"2026-09-30","why":{"zh":"规模化 AI 数学发现首次通过顶刊同行评审","en":"Large-scale AI maths discovery passes peer review in a top journal"},"fieldImpact":"major","aiImpact":"major"},{"title":"An orthogonality inequality in $\\mathbb{RP}^{n-1}$ and the double-cap conjecture","titleZh":"人类数学家借助 AI 证明“双帽猜想”（球面上无正交点对的集合测度上限），主结论另有几乎完全由 AI 完成形式化的 Lean 4 证明","id":"arXiv:2610.27523","url":"https://arxiv.org/abs/2610.27523","venue":"arXiv","date":"2026-10-20","why":{"zh":"AI：作者在摘要中说明 AI 协助撰写论文，主不等式和双帽结论的另一份 Lean 4 证明几乎完全由 AI 工具完成形式化与验证；学科：双帽猜想是球面上“无正交对集合”测度的长期公开问题，论文同时确定球面正交图的可测分数色数（预印本，尚未审稿）","en":"AI: per the abstract, AI assisted the paper and a separate Lean 4 proof of the main results was formalised almost entirely with AI tools. Field: settles the long-open double-cap conjecture and the measurable fractional chromatic number of the spherical orthogonality graph (preprint)"},"fieldImpact":"major","aiImpact":"major"}],"latest_papers":[{"id":"arXiv:2610.30027","title":"The recent counterexamples to the Eisenbud-Schreyer-Weyman Ulrich problem (2003) and to Hochster's small Cohen-Macaulay conjecture (1973): a survey, with a generalization to arbitrary dimension","titleZh":"Eisenbud–Schreyer–Weyman Ulrich 问题与 Hochster 小 Cohen–Macaulay 猜想反例综述","summaryZh":"梳理 2026 年 9 月四篇预印本和一份 AI 生成稿件如何把射影几何中的一个数值障碍推进到局部代数，得到两个长期问题的反例，并推广到任意维数。","summaryEn":"Surveys how four preprints and an AI-generated manuscript (Sept 2026) produced counterexamples to the Ulrich problem and Hochster’s small CM conjecture, generalised to all dimensions.","authors":["Cristian Anghel"],"date":"2026-10-21","url":"https://arxiv.org/abs/2610.30027","site":"arXiv","venue":null,"category":"math.AG","fetchedAt":"2026-10-22T10:01:44"},{"id":"arXiv:2610.27523","title":"An orthogonality inequality in $\\mathbb{RP}^{n-1}$ and the double-cap conjecture","titleZh":"ℝPⁿ⁻¹ 中的正交不等式与“双帽猜想”","summaryZh":"证明在实射影空间中“帽”使正交邻域测度最小，由此得到双帽猜想：球面上不含正交点对的可测集测度不超过两个半径 π/4 的对顶开帽；主结论另有一份基于不同论证、几乎完全由 AI 工具完成形式化的 Lean 4 证明。","summaryEn":"Proves caps minimise orthogonal-neighbourhood measure, yielding the double-cap conjecture; a separate Lean 4 proof was formalised almost entirely with AI tools.","authors":["Allen Ibiapina","Carlos Gomes","Alan Pio","Samuel Belo"],"date":"2026-10-20","url":"https://arxiv.org/abs/2610.27523","site":"arXiv","venue":null,"category":"math.MG","fetchedAt":"2026-10-22T10:01:44"},{"id":"arXiv:2610.26790","title":"Training with ML-labeled Data via Stochastic First- and Zeroth-Order Optimization","titleZh":"用机器学习标注的数据训练：随机一阶与零阶优化","summaryZh":"针对部分样本有真值标签、其余由机器学习模型标注（可能有错）的数据，提出随机一阶和零阶优化训练方法，实验从线性回归到大模型微调。","summaryEn":"Studies training with ML-generated labels instead of costly human ground truth, via stochastic first- and zeroth-order optimisation.","authors":["Maria-Eleni Sfyraki","Jun-Kun Wang"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.26790","site":"arXiv","venue":null,"category":"cs.LG","fetchedAt":"2026-10-22T10:01:44"},{"id":"arXiv:2610.25659","title":"Beyond Leaderboards: A Method for Application-Specific Capability Profiling of Large Language Models","titleZh":"超越排行榜：面向具体应用的大模型能力画像方法","summaryZh":"提出按具体应用评估大模型能力的方法和容器化测试平台，例如数据库迁移智能体和基于文档的聊天机器人，帮助工程师按应用、架构和成本选型。","summaryEn":"A method and containerised testbed for profiling LLM capability per application (e.g. a database-migration agent), instead of generic leaderboards.","authors":["Oleg Grynets","Vasyl Lyashkevych","Andrii Salyk","Dmytro Koziuberda"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.25659","site":"arXiv","venue":null,"category":"cs.LO","fetchedAt":"2026-10-22T10:01:44"},{"id":"arXiv:2610.26467","title":"Rózsa: Certified Equational Reasoning","titleZh":"Rózsa：可认证的等式推理","summaryZh":"为 SAIR 数学蒸馏挑战赛第二阶段开发的求解器，判断二元运算集合上一个等式是否蕴含另一个，并把证明或反例转成 Lean 证书；正式评测解出 199/200 题，排名第 6。","summaryEn":"A solver for the SAIR Mathematics Distillation Challenge that emits Lean certificates; solved 199/200, placing 6th.","authors":["Georgios Kissas"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.26467","site":"arXiv","venue":null,"category":"cs.LO","fetchedAt":"2026-10-22T10:01:44"},{"id":"arXiv:2610.25357","title":"Studying proof data with proof theory: the case of AI-generated mathematics","titleZh":"用证明论研究证明数据：以 AI 生成的数学为例","summaryZh":"提出一种分析大规模形式化验证证明（如 OpenAI 发布的证明）的方法，把证明项大小、冗余步骤数等量变成衡量 AI 生成数学的指标。","summaryEn":"Proposes proof-theoretic measures (proof-term size, redundant steps) to analyse large collections of formally verified, AI-generated proofs such as OpenAI’s.","authors":["Safal Raman Aryal"],"date":"2026-10-18","url":"https://arxiv.org/abs/2610.25357","site":"arXiv","venue":null,"category":"math.LO","fetchedAt":"2026-10-22T10:01:44"},{"id":"arXiv:2610.24737","title":"Gaussian Process Bandits with Paired Auxiliary Observations","titleZh":"带配对辅助观测的高斯过程赌博机","summaryZh":"提出 AI-GPB 方法，在贝叶斯优化中利用事先已有的低成本辅助数据降低观测噪声带来的方差，加快黑箱优化。","summaryEn":"AI-GPB uses cheap auxiliary data available beforehand to reduce variance in Bayesian optimisation of noisy black-box functions.","authors":["Weibo Huang","Qing Feng","Ruihao Zhu","Cheng Hua"],"date":"2026-10-17","url":"https://arxiv.org/abs/2610.24737","site":"arXiv","venue":null,"category":"cs.LG","fetchedAt":"2026-10-22T10:01:44"},{"id":"arXiv:2610.01447","title":"An $O(1)$ Bound for the KLS Constant","titleZh":"KLS 常数的 O(1) 上界：证明 Kannan–Lovász–Simonovits 猜想","summaryZh":"证明各向同性对数凹测度的 Cheeger 常数有与维度无关的下界（ψₙ≤C），从而解决 KLS 猜想，并给出庞加莱常数的普适上界；微软称研究中使用了人类主导的 AI 智能体。","summaryEn":"Proves ψ_n ≤ C for a universal constant, resolving the KLS conjecture, and a universal Poincaré-constant bound; Microsoft says human-directed AI agents were used.","authors":["Zhao Song","Xinzhi Zhang"],"date":"2026-10-01","url":"https://arxiv.org/abs/2610.01447","site":"arXiv","venue":null,"category":"math.PR","fetchedAt":"2026-10-22T15:20:00"}]},{"id":"statistics","slug":"statistics","category":"formal-physical","categoryName":{"zh":"数理科学","en":"Mathematical & physical sciences"},"name":{"zh":"统计学","en":"Statistics"},"classification":{"oecdFos":"1.1","moe":"0714 统计学"},"level":2,"levelKey":"breakthrough","levelLabel":{"zh":"局部突破","en":"Local breakthroughs"},"trend":"flat","reason":{"zh":"表格基础模型 TabPFN 在小样本表格预测上超过调参后的梯度提升树（Nature 2025）；“预测驱动推断”给出用 AI 预测做有效统计推断的方法（Science 2023）。统计学核心方法尚未被系统改写。","en":"Tabular foundation model TabPFN beats tuned gradient boosting on small tabular data (Nature 2025); prediction-powered inference enables valid inference with ML predictions (Science 2023). Core methods are not yet systematically rewritten."},"reasonSources":[{"url":"https://www.nature.com/articles/s41586-024-08328-6","linkCheck":"ok"},{"url":"https://www.science.org/doi/10.1126/science.adi6000","linkCheck":"ok-crossref"}],"directions":{"zh":["表格基础模型","预测驱动推断","自动化数据分析智能体"],"en":["Tabular foundation models","Prediction-powered inference","Data-analysis agents"]},"stats":{"milestones":10,"milestonesSince2025":8,"milestones2026":5,"milestonesLast6Months":5,"news":5,"latestProgress":"2026-10-20","latestMilestone":"2026-10-20","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":0,"leaps":{"ai":0,"field":0},"latestPapers":8,"keyPapers":9},"milestones":[{"date":"2023-11-10","title":{"zh":"预测驱动推断（Prediction-powered inference）发表于 Science","en":"Prediction-powered inference published in Science"},"url":"https://www.science.org/doi/10.1126/science.adi6000","source":"Science","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"给出用 AI 预测做有效置信区间的一般框架，被大量跟进","en":"General framework for valid inference with ML predictions"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"给出用 AI 预测做有效置信区间的一般框架，被大量跟进","en":"General framework for valid inference with ML predictions"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"给出用 AI 预测做有效置信区间的一般框架，被大量跟进","en":"General framework for valid inference with ML predictions"}},"impactNotes":[],"verified":{"id":"10.1126/science.adi6000","title":"Prediction-powered inference","venue":"Science","date":"2023-11-10"},"addedIn":"v1"},{"date":"2024-04-03","title":{"zh":"交叉预测驱动推断：用机器学习预测补足标注数据并保持推断有效（PNAS）","en":"Cross-prediction-powered inference (PNAS)"},"url":"https://doi.org/10.1073/pnas.2322083121","source":"PNAS","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"把 PPI 推广到无需额外标注模型的情形，扩大了“用 AI 预测做有效推断”的适用面","en":"Extends PPI to settings without a separate labelled model"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"把 PPI 推广到无需额外标注模型的情形，扩大了“用 AI 预测做有效推断”的适用面","en":"Extends PPI to settings without a separate labelled model"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"把 PPI 推广到无需额外标注模型的情形，扩大了“用 AI 预测做有效推断”的适用面","en":"Extends PPI to settings without a separate labelled model"}},"impactNotes":[],"verified":{"id":"10.1073/pnas.2322083121","title":"Cross-prediction-powered inference","venue":"Proceedings of the National Academy of Sciences","date":"2024-04-03"},"addedIn":"v2"},{"date":"2025-01-08","title":{"zh":"TabPFN：小样本表格数据基础模型（Nature）","en":"TabPFN: tabular foundation model for small data (Nature)"},"url":"https://www.nature.com/articles/s41586-024-08328-6","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"小样本表格预测超过调参梯度提升树，挑战统计/机器学习的默认工具","en":"Beats tuned gradient boosting on small tabular data"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"小样本表格预测超过调参梯度提升树，挑战统计/机器学习的默认工具","en":"Beats tuned gradient boosting on small tabular data"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"小样本表格预测超过调参梯度提升树，挑战统计/机器学习的默认工具","en":"Beats tuned gradient boosting on small tabular data"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-024-08328-6","title":"Accurate predictions on small data with a tabular foundation model","venue":"Nature","date":"2025-01-08"},"addedIn":"v1"},{"date":"2025-07-17","title":{"zh":"LAMBDA：大模型数据分析智能体，JASA 讨论论文（Donoho、Xihong Lin、孟晓犁等受邀讨论）","en":"LAMBDA: LLM data agent, JASA discussion paper with invited discussants incl. Donoho, Xihong Lin, Xiao-Li Meng"},"url":"https://doi.org/10.1080/01621459.2025.2510000","source":"JASA","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"统计学顶刊首次正面讨论数据分析走向全自动化","en":"First flagship-journal debate on fully automated data analysis"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"统计学顶刊首次正面讨论数据分析走向全自动化","en":"First flagship-journal debate on fully automated data analysis"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"统计学顶刊首次正面讨论数据分析走向全自动化","en":"First flagship-journal debate on fully automated data analysis"}},"impactNotes":[],"verified":{"id":"10.1080/01621459.2025.2510000","title":"LAMBDA: A Large Model Based Data Agent","venue":"Journal of the American Statistical Association","date":"2025-07-17"},"addedIn":"v2"},{"date":"2025-10-16","title":{"zh":"《The American Statistician》发表大模型统计/数据科学智能体综述","en":"The American Statistician survey of LLM agents for statistics & data science"},"url":"https://doi.org/10.1080/00031305.2025.2561140","source":"The American Statistician","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"主流统计期刊系统梳理数据智能体","en":"Mainstream stats journal maps the data-agent field"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"主流统计期刊系统梳理数据智能体","en":"Mainstream stats journal maps the data-agent field"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"主流统计期刊系统梳理数据智能体","en":"Mainstream stats journal maps the data-agent field"}},"impactNotes":[],"verified":{"id":"10.1080/00031305.2025.2561140","title":"A Survey on Large Language Model-based Agents for Statistics and Data Science","venue":"The American Statistician","date":"2025-10-16"},"addedIn":"v2"},{"date":"2026-04-24","title":{"zh":"JASA 2026 年 121 卷 553 期刊出 LAMBDA 及全部受邀讨论","en":"JASA 121(553) publishes LAMBDA with all invited discussions"},"url":"https://community.amstat.org/discussion/jasa-2026-discussion-paper-on-lambda-a-large-model-based-data-agent-and-and-the-future-of-automated-data-analysis","source":"ASA Connect","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"industry","typeLabel":{"zh":"产业/采用","en":"Industry"},"why":{"zh":"标志统计学界开始正式讨论 AI 对学科的冲击","en":"Marks the statistics community formally debating AI’s impact"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"标志统计学界开始正式讨论 AI 对学科的冲击","en":"Marks the statistics community formally debating AI’s impact"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"标志统计学界开始正式讨论 AI 对学科的冲击","en":"Marks the statistics community formally debating AI’s impact"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-09-01","title":{"zh":"PPI++：高效的“预测驱动推断”正式发表（Annals of Applied Statistics）","en":"PPI++: Efficient prediction-powered inference (Ann. Appl. Stat.)"},"url":"https://doi.org/10.1214/26-aoas2215","source":"The Annals of Applied Statistics","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"AI：让大量 AI 预测与少量金标准数据结合，得到有效置信区间；学科：预测驱动推断进入主流统计期刊，降低方差并适用高维问题","en":"AI: combines abundant AI predictions with scarce gold labels for valid intervals; Field: prediction-powered inference enters a mainstream statistics journal"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：让大量 AI 预测与少量金标准数据结合，得到有效置信区间；学科：预测驱动推断进入主流统计期刊，降低方差并适用高维问题","en":"AI: combines abundant AI predictions with scarce gold labels for valid intervals; Field: prediction-powered inference enters a mainstream statistics journal"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：让大量 AI 预测与少量金标准数据结合，得到有效置信区间；学科：预测驱动推断进入主流统计期刊，降低方差并适用高维问题","en":"AI: combines abundant AI predictions with scarce gold labels for valid intervals; Field: prediction-powered inference enters a mainstream statistics journal"}},"impactNotes":[],"verified":{"id":"10.1214/26-aoas2215","title":"PPI++: Efficient prediction-powered inference","venue":"The Annals of Applied Statistics","date":"2026-09-01"},"addedIn":"v2"},{"date":"2026-10-01","title":{"zh":"Chernozhukov、Newey 等：把 AI 学到的表征（文本、图像嵌入）当作控制变量做因果推断时，交叉拟合双重机器学习（DML）仍可给出有效推断，并给出误差较大时的敏感性区间","en":"Chernozhukov, Newey et al.: “Pragmatic DML with AI-Learned Representations” — valid DML inference and sensitivity bounds when AI embeddings serve as controls"},"url":"https://arxiv.org/abs/2610.01935","source":"arXiv","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"AI：文本、图像经基础模型压缩成表征后进入因果分析已很普遍；学科：证明表征误差对目标参数的扭曲等于两项误差之积，给出可直接照做的“DML 点估计 → 解释 → 敏感性区间”流程，是计量与统计推断适配 AI 特征的方法学进展（预印本，未审稿）","en":"AI: foundation-model embeddings are now routinely used as controls. Field: shows distortion equals a product of two representation errors and gives a practical DML → interpretation → sensitivity workflow (preprint)"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：文本、图像经基础模型压缩成表征后进入因果分析已很普遍；学科：证明表征误差对目标参数的扭曲等于两项误差之积，给出可直接照做的“DML 点估计 → 解释 → 敏感性区间”流程，是计量与统计推断适配 AI 特征的方法学进展（预印本，未审稿）","en":"AI: foundation-model embeddings are now routinely used as controls. Field: shows distortion equals a product of two representation errors and gives a practical DML → interpretation → sensitivity workflow (preprint)"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：文本、图像经基础模型压缩成表征后进入因果分析已很普遍；学科：证明表征误差对目标参数的扭曲等于两项误差之积，给出可直接照做的“DML 点估计 → 解释 → 敏感性区间”流程，是计量与统计推断适配 AI 特征的方法学进展（预印本，未审稿）","en":"AI: foundation-model embeddings are now routinely used as controls. Field: shows distortion equals a product of two representation errors and gives a practical DML → interpretation → sensitivity workflow (preprint)"}},"impactNotes":[],"verified":{"id":"arXiv:2610.01935","title":"Pragmatic DML with AI-Learned Representations","venue":"arXiv","date":"2026-10-01"},"addedIn":"v2-20261022audit"},{"date":"2026-10-05","title":{"zh":"PNAS：系统检验开源“AI 科学家”系统，发现基准选择不当、数据泄漏、指标误用和事后选择偏差四类统计陷阱，且在最终论文里几乎看不出来；提供完整日志和代码后审查效果明显改善","en":"PNAS: “The more you automate, the less you see” — open-source AI-scientist systems show benchmark misselection, data leakage, metric misuse and post hoc selection bias that are invisible in the final manuscript"},"url":"https://doi.org/10.1073/pnas.2610214123","source":"PNAS","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"AI：用受控实验逐一隔离四类陷阱，两套代表性开源系统都中招；学科：把统计学关心的泄漏、多重比较与选择性报告问题带进自动化科研，主张期刊从“只审论文”转向审查完整工作流（日志与代码）","en":"AI: controlled experiments isolate four pitfalls in two open-source systems. Field: brings leakage, selection bias and selective reporting into automated research and argues for auditing full workflow artefacts"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：用受控实验逐一隔离四类陷阱，两套代表性开源系统都中招；学科：把统计学关心的泄漏、多重比较与选择性报告问题带进自动化科研，主张期刊从“只审论文”转向审查完整工作流（日志与代码）","en":"AI: controlled experiments isolate four pitfalls in two open-source systems. Field: brings leakage, selection bias and selective reporting into automated research and argues for auditing full workflow artefacts"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：用受控实验逐一隔离四类陷阱，两套代表性开源系统都中招；学科：把统计学关心的泄漏、多重比较与选择性报告问题带进自动化科研，主张期刊从“只审论文”转向审查完整工作流（日志与代码）","en":"AI: controlled experiments isolate four pitfalls in two open-source systems. Field: brings leakage, selection bias and selective reporting into automated research and argues for auditing full workflow artefacts"}},"impactNotes":[],"verified":{"id":"10.1073/pnas.2610214123","title":"The more you automate, the less you see: Hidden pitfalls of autonomous AI scientists","venue":"PNAS","date":"2026-10-05"},"addedIn":"v2-20261022audit"},{"date":"2026-10-20","title":{"zh":"MR-MAS（Bioinformatics）：大模型多智能体端到端自动完成孟德尔随机化因果分析——取数、效应估计、结果解读到成稿，跨模型后端重复运行结果一致","en":"MR-MAS (Bioinformatics): an LLM multi-agent system that runs end-to-end Mendelian randomization — data retrieval, causal estimation, interpretation and write-up"},"url":"https://doi.org/10.1093/bioinformatics/btag722","source":"Bioinformatics","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"system","typeLabel":{"zh":"系统发布","en":"System"},"why":{"zh":"AI：配套结构化 MR 知识库用于方法实现和代码修复；学科：在多个场景复现了文献支持的因果信号、重复运行估计稳定，是统计因果推断流程被智能体自动化的同行评审案例，但仍属工具层面","en":"AI: a structured MR knowledge base guides implementation and code repair. Field: reproduces literature-supported causal signals with stable estimates — a peer-reviewed case of agent-automated causal-inference workflows"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：配套结构化 MR 知识库用于方法实现和代码修复；学科：在多个场景复现了文献支持的因果信号、重复运行估计稳定，是统计因果推断流程被智能体自动化的同行评审案例，但仍属工具层面","en":"AI: a structured MR knowledge base guides implementation and code repair. Field: reproduces literature-supported causal signals with stable estimates — a peer-reviewed case of agent-automated causal-inference workflows"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：配套结构化 MR 知识库用于方法实现和代码修复；学科：在多个场景复现了文献支持的因果信号、重复运行估计稳定，是统计因果推断流程被智能体自动化的同行评审案例，但仍属工具层面","en":"AI: a structured MR knowledge base guides implementation and code repair. Field: reproduces literature-supported causal signals with stable estimates — a peer-reviewed case of agent-automated causal-inference workflows"}},"impactNotes":[],"verified":{"id":"10.1093/bioinformatics/btag722","title":"MR-MAS: an end-to-end multi-agent system for autonomous Mendelian randomization analysis","venue":"Bioinformatics","date":"2026-10-20"},"addedIn":"v2-20261022audit"}],"papersRepos":[{"title":"PriorLabs/TabPFN","url":"https://github.com/PriorLabs/TabPFN","date":null,"kind":"repo","linkCheck":"ok"},{"title":"Prediction-powered inference","url":"https://doi.org/10.1126/science.adi6000","date":"2023-11-10","kind":"paper","linkCheck":"ok-crossref"},{"title":"Cross-prediction-powered inference","url":"https://doi.org/10.1073/pnas.2322083121","date":"2024-04-03","kind":"paper","linkCheck":"ok-crossref"},{"title":"Accurate predictions on small data with a tabular foundation model","url":"https://doi.org/10.1038/s41586-024-08328-6","date":"2025-01-08","kind":"paper","linkCheck":"ok-crossref"},{"title":"LAMBDA: A Large Model Based Data Agent","url":"https://doi.org/10.1080/01621459.2025.2510000","date":"2025-07-17","kind":"paper","linkCheck":"ok-crossref"},{"title":"PPI++: Efficient prediction-powered inference","url":"https://doi.org/10.1214/26-aoas2215","date":"2026-09-01","kind":"paper","linkCheck":"ok-crossref"},{"title":"A Survey on Large Language Model-based Agents for Statistics and Data Science","url":"https://doi.org/10.1080/00031305.2025.2561140","date":"2025-10-16","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"Prior Labs (TabPFN)","url":"https://priorlabs.ai"}],"news":[{"id":"fa109e275d9c","date":"2026-10-20","title":"A robust deep learning framework for smart grid stability classification using statistical validation","url":"https://www.nature.com/articles/s41598-026-74799-4","source":"Nature 学科·数学","lang":"en","kind":"news","summaryZh":"Nature 学科·数学于2026-10-20发布的文章，涉及AI 数学、AI 气候能源、AI 软件开发，原题《A robust deep learning framework for smart grid stability classification using statistical》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"fd99da729cdc","date":"2026-10-09","title":"The statistical science behind trustworthy AI","url":"https://statistics.gmu.edu/news/2026-10/statistical-science-behind-trustworthy-ai","source":"George Mason University Statistics","lang":"en","kind":"news","summaryZh":"国际华人统计学会（ICSA）2026 研讨会在乔治梅森大学举行，约 500 人讨论统计方法如何评估 AI 的不确定性、偏差与可靠性。","via":"manual-websearch-20261021","linkCheck":"ok"},{"id":"8a519467a0f7","date":"2026-10-01","title":"When the Answer Comes Too Easily: Where Statistics Goes Next","url":"https://stattrak.amstat.org/2026/10/01/when-the-answer-comes-too-easily/","source":"Amstat News · STATtr@k（美国统计协会）","lang":"en","kind":"news","summaryZh":"美国统计协会刊物评论：AI 让“答案来得太容易”，统计学下一步该做什么。","via":"manual-websearch-20261021","linkCheck":"ok"},{"id":"8244dfd04765","date":"2026-10","title":"ONS surveys and economic statistics improvement plan, quarterly progress update: October 2026","url":"https://www.ons.gov.uk/aboutus/ourstrategyandplans/onsimprovementplans/onssurveyimprovementandenhancement/onssurveysandeconomicstatisticsimprovementplanquarterlyprogressupdateoctober2026","source":"Office for National Statistics","lang":"en","kind":"news","summaryZh":"英国统计局季度进展：发布“AI 专题账户”框架，调查显示约 3 成企业在工作中使用 AI。","via":"manual-websearch-20261021","linkCheck":"ok"},{"id":"19e7b727592f","date":"2026-09-29","title":"New statistical method can flag hidden anomalies in scientific datasets","url":"https://phys.org/news/2026-09-statistical-method-flag-hidden-anomalies.html","source":"Phys.org 数学频道","lang":"en","kind":"news","summaryZh":"Phys.org 数学频道于2026-09-29发布的文章，涉及AI 数学、AI 生物/医药、AI 教育，原题《New statistical method can flag hidden anomalies in scientific datasets》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"}],"review":{"lastReviewed":"2026-10-22","cycleDays":90,"nextReview":"2027-01-19"},"urls":{"zh":"/zh-hans/disciplines/statistics/","en":"/en/disciplines/statistics/"},"dimensions":{"method":{"score":2,"label":{"zh":"研究方法","en":"Research method"},"evidence":"表格基础模型、预测驱动推断、数据分析智能体进入顶刊，但统计推断核心方法未被替代","source":"https://www.nature.com/articles/s41586-024-08328-6","weight":0.3},"discovery":{"score":2,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"TabPFN（Nature 2025）、PPI（Science 2023）是以 AI 为关键的方法学突破","source":"https://www.science.org/doi/10.1126/science.adi6000","weight":0.35},"adoption":{"score":2,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"JASA 以讨论论文形式讨论“数据分析全自动化”（2026-04），ASA 举办 AI 智能体研讨；缺少从业者使用率调查","source":"https://community.amstat.org/discussion/jasa-2026-discussion-paper-on-lambda-a-large-model-based-data-agent-and-and-the-future-of-automated-data-analysis","weight":0.25},"education":{"score":1,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"ASA 2026-09-26 更新的战略计划把“在 AI 教育中融入统计概念与方法”列为教育目标，ASA AI 战略草案计划在 JSM 2027 前发布“AI 时代统计课程”白皮书；尚无课程体系改革落地的证据","source":"https://www.amstat.org/docs/default-source/amstat-documents/abt-strategicplan.pdf?v=092626","weight":0.1}},"compositeScore":1.9,"levelCapped":false,"levelChange":{"from":2,"to":2,"direction":"same","reason":null},"biggestBreakthrough":{"date":"2025-01-08","title":{"zh":"TabPFN：小样本表格数据基础模型（Nature）","en":"TabPFN: tabular foundation model for small data (Nature)"},"url":"https://www.nature.com/articles/s41586-024-08328-6","aiImpact":"major","fieldImpact":"major","why":{"zh":"小样本表格预测超过调参梯度提升树，挑战统计/机器学习的默认工具","en":"Beats tuned gradient boosting on small tabular data"}},"keyPapers":[{"title":"Prediction-powered inference","titleZh":"预测驱动推断（Prediction-powered inference）发表于 Science","id":"10.1126/science.adi6000","url":"https://doi.org/10.1126/science.adi6000","venue":"Science","date":"2023-11-10","why":{"zh":"给出用 AI 预测做有效置信区间的一般框架，被大量跟进","en":"General framework for valid inference with ML predictions"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Cross-prediction-powered inference","titleZh":"交叉预测驱动推断：用机器学习预测补足标注数据并保持推断有效（PNAS）","id":"10.1073/pnas.2322083121","url":"https://doi.org/10.1073/pnas.2322083121","venue":"Proceedings of the National Academy of Sciences","date":"2024-04-03","why":{"zh":"把 PPI 推广到无需额外标注模型的情形，扩大了“用 AI 预测做有效推断”的适用面","en":"Extends PPI to settings without a separate labelled model"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Accurate predictions on small data with a tabular foundation model","titleZh":"TabPFN：小样本表格数据基础模型（Nature）","id":"10.1038/s41586-024-08328-6","url":"https://doi.org/10.1038/s41586-024-08328-6","venue":"Nature","date":"2025-01-08","why":{"zh":"小样本表格预测超过调参梯度提升树，挑战统计/机器学习的默认工具","en":"Beats tuned gradient boosting on small tabular data"},"fieldImpact":"major","aiImpact":"major"},{"title":"LAMBDA: A Large Model Based Data Agent","titleZh":"LAMBDA：大模型数据分析智能体，JASA 讨论论文（Donoho、Xihong Lin、孟晓犁等受邀讨论）","id":"10.1080/01621459.2025.2510000","url":"https://doi.org/10.1080/01621459.2025.2510000","venue":"Journal of the American Statistical Association","date":"2025-07-17","why":{"zh":"统计学顶刊首次正面讨论数据分析走向全自动化","en":"First flagship-journal debate on fully automated data analysis"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"PPI++: Efficient prediction-powered inference","titleZh":"PPI++：高效的“预测驱动推断”正式发表（Annals of Applied Statistics）","id":"10.1214/26-aoas2215","url":"https://doi.org/10.1214/26-aoas2215","venue":"The Annals of Applied Statistics","date":"2026-09-01","why":{"zh":"AI：让大量 AI 预测与少量金标准数据结合，得到有效置信区间；学科：预测驱动推断进入主流统计期刊，降低方差并适用高维问题","en":"AI: combines abundant AI predictions with scarce gold labels for valid intervals; Field: prediction-powered inference enters a mainstream statistics journal"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Pragmatic DML with AI-Learned Representations","titleZh":"Chernozhukov、Newey 等：把 AI 学到的表征（文本、图像嵌入）当作控制变量做因果推断时，交叉拟合双重机器学习（DML）仍可给出有效推断，并给出误差较大时的敏感性区间","id":"arXiv:2610.01935","url":"https://arxiv.org/abs/2610.01935","venue":"arXiv","date":"2026-10-01","why":{"zh":"AI：文本、图像经基础模型压缩成表征后进入因果分析已很普遍；学科：证明表征误差对目标参数的扭曲等于两项误差之积，给出可直接照做的“DML 点估计 → 解释 → 敏感性区间”流程，是计量与统计推断适配 AI 特征的方法学进展（预印本，未审稿）","en":"AI: foundation-model embeddings are now routinely used as controls. Field: shows distortion equals a product of two representation errors and gives a practical DML → interpretation → sensitivity workflow (preprint)"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"The more you automate, the less you see: Hidden pitfalls of autonomous AI scientists","titleZh":"PNAS：系统检验开源“AI 科学家”系统，发现基准选择不当、数据泄漏、指标误用和事后选择偏差四类统计陷阱，且在最终论文里几乎看不出来；提供完整日志和代码后审查效果明显改善","id":"10.1073/pnas.2610214123","url":"https://doi.org/10.1073/pnas.2610214123","venue":"PNAS","date":"2026-10-05","why":{"zh":"AI：用受控实验逐一隔离四类陷阱，两套代表性开源系统都中招；学科：把统计学关心的泄漏、多重比较与选择性报告问题带进自动化科研，主张期刊从“只审论文”转向审查完整工作流（日志与代码）","en":"AI: controlled experiments isolate four pitfalls in two open-source systems. Field: brings leakage, selection bias and selective reporting into automated research and argues for auditing full workflow artefacts"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"A Survey on Large Language Model-based Agents for Statistics and Data Science","titleZh":"《The American Statistician》发表大模型统计/数据科学智能体综述","id":"10.1080/00031305.2025.2561140","url":"https://doi.org/10.1080/00031305.2025.2561140","venue":"The American Statistician","date":"2025-10-16","why":{"zh":"主流统计期刊系统梳理数据智能体","en":"Mainstream stats journal maps the data-agent field"},"fieldImpact":"incremental","aiImpact":"incremental"},{"title":"MR-MAS: an end-to-end multi-agent system for autonomous Mendelian randomization analysis","titleZh":"MR-MAS（Bioinformatics）：大模型多智能体端到端自动完成孟德尔随机化因果分析——取数、效应估计、结果解读到成稿，跨模型后端重复运行结果一致","id":"10.1093/bioinformatics/btag722","url":"https://doi.org/10.1093/bioinformatics/btag722","venue":"Bioinformatics","date":"2026-10-20","why":{"zh":"AI：配套结构化 MR 知识库用于方法实现和代码修复；学科：在多个场景复现了文献支持的因果信号、重复运行估计稳定，是统计因果推断流程被智能体自动化的同行评审案例，但仍属工具层面","en":"AI: a structured MR knowledge base guides implementation and code repair. Field: reproduces literature-supported causal signals with stable estimates — a peer-reviewed case of agent-automated causal-inference workflows"},"fieldImpact":"incremental","aiImpact":"incremental"}],"latest_papers":[{"id":"arXiv:2610.27447","title":"Beyond Scalar Leaderboards: Adaptive Sampling for Pareto Frontier Identification in Multi-Objective LLM Arenas","titleZh":"超越单一排行榜：多目标大模型竞技场中识别帕累托前沿的自适应采样","summaryZh":"提出 VB-EGE 固定置信度自适应采样算法，用尽量少的成对比较找出在数学、代码、写作等多维度上处于帕累托前沿的模型，并证明其样本复杂度上界与极小极大最优性。","summaryEn":"VB-EGE adaptively samples arena comparisons to certify the Pareto frontier of LLMs with minimax-optimal sample complexity.","authors":["Liang Twist Shan","Jiahao Ai","Zhimei Ren"],"date":"2026-10-20","url":"https://arxiv.org/abs/2610.27447","site":"arXiv","venue":null,"category":"stat.ML","fetchedAt":"2026-10-22T10:01:48"},{"id":"arXiv:2610.25932","title":"Keep the Lineages Alive: Diversity-Preserving SMC for Diffusion LLM Decoding","titleZh":"保持谱系存活：面向扩散语言模型解码的保多样性序贯蒙特卡洛","summaryZh":"改进扩散语言模型推理时的序贯蒙特卡洛重采样步骤，在集中算力于有希望候选的同时保留多样性，以提升推理准确率。","summaryEn":"Improves the resampling step of inference-time SMC for diffusion LLMs, keeping diversity while focusing compute on promising candidates.","authors":["Minoo Ahmadi","Seyedarmin Azizi","Mehdi Kamal","Massoud Pedram"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.25932","site":"arXiv","venue":null,"category":"cs.LG","fetchedAt":"2026-10-22T10:01:48"},{"id":"arXiv:2610.23601","title":"Token-Level Attribution of Multi-Mode LLM Watermarks: Optimal Rule and Belief Weighting","titleZh":"多模式大模型水印的词元级归属：最优规则与置信加权","summaryZh":"研究当验证者持有多个水印密钥、文档混有多种水印和无水印文本时，如何在控制第一类错误的前提下做逐词元归属。","summaryEn":"Token-level attribution for multi-key Gumbel-max LLM watermarks that controls Type I error while keeping power.","authors":["Rui Zhou","Licheng Zhao","Chao Shen","Wenqiang Pu"],"date":"2026-10-16","url":"https://arxiv.org/abs/2610.23601","site":"arXiv","venue":null,"category":"cs.CR","fetchedAt":"2026-10-22T10:01:48"},{"id":"arXiv:2610.22695","title":"Chance-Constrained Personalized LLM Routing with Online Preference Learning","titleZh":"带机会约束的个性化大模型路由与在线偏好学习","summaryZh":"把在不同成本和能力的大模型之间选择的问题建模为约束上下文赌博机，在学习用户偏好的同时降低推理成本并保证回答质量。","summaryEn":"Formulates personalised LLM routing as a constrained contextual bandit that learns preferences while cutting cost and keeping quality.","authors":["Ameya Anjarlekar","Rasoul Etesami","R Srikant"],"date":"2026-10-15","url":"https://arxiv.org/abs/2610.22695","site":"arXiv","venue":null,"category":"cs.LG","fetchedAt":"2026-10-22T10:01:48"},{"id":"arXiv:2610.21837","title":"Covariance-Calibrated Prediction-Powered Inference","titleZh":"协方差校准的预测驱动推断（CC-PPI）","summaryZh":"用无标注样本上的模型预测分数校准 GMM 估计中的协方差权重，并加谱保护保证正定，让预测驱动推断在少量标注下更有效。","summaryEn":"Uses unlabeled prediction scores to calibrate the covariance weight in prediction-powered GMM inference, with spectral protection for positive definiteness.","authors":["Siyuan Chen","Yang Bai","Huichen Zhu"],"date":"2026-10-15","url":"https://arxiv.org/abs/2610.21837","site":"arXiv","venue":null,"category":"stat.ME","fetchedAt":"2026-10-22T10:01:48"},{"id":"arXiv:2610.28666","title":"Learning Checklists to Reduce Lottery Effects in Scientific Peer Review","titleZh":"学习评审清单以减少同行评审中的“抽签误差”","summaryZh":"从历史评审语料中学出分层的批评清单，提示审稿人可能遗漏的问题，从而减少少数审稿人只覆盖部分问题造成的误差。","summaryEn":"Learns hierarchical checklists of criticisms from historical reviews to reduce “reviewer lottery” error in peer review.","authors":["Xuanyou Liu","Ziyang Guo","Jason Hartline","Jessica Hullman"],"date":"2026-10-14","url":"https://arxiv.org/abs/2610.28666","site":"arXiv","venue":null,"category":"cs.DL","fetchedAt":"2026-10-22T10:01:48"},{"id":"arXiv:2610.18504","title":"How Low Can You Go: Fine-Tuning LLMs with Bayesian Optimization (LlMBO)","titleZh":"能有多低：用贝叶斯优化微调大模型（LlMBO）","summaryZh":"在只有极少可训练参数的 TinyLoRA 微调中，用贝叶斯优化代替梯度局部搜索。","summaryEn":"Replaces gradient-based local search with Bayesian optimisation for TinyLoRA fine-tuning with very few parameters.","authors":["Aditya Kulkarni","Davide Sartor","Carlos Soto","Nathan Wycoff"],"date":"2026-10-13","url":"https://arxiv.org/abs/2610.18504","site":"arXiv","venue":null,"category":"stat.ML","fetchedAt":"2026-10-22T10:01:48"},{"id":"arXiv:2610.17655","title":"textGrain watermarking is harder to detect: can sum-tests still do it optimally?","titleZh":"textGrain 水印更难检测：求和检验还能最优吗？","summaryZh":"分析新型 textGrain 大模型文本水印的统计检测问题，讨论求和型检验是否仍然最优。","summaryEn":"Studies whether sum-type tests remain optimal for detecting the entropy-calibrated textGrain LLM watermark.","authors":["Swapnaneel Bhattacharyya","Soham Bonnerjee"],"date":"2026-10-13","url":"https://arxiv.org/abs/2610.17655","site":"arXiv","venue":null,"category":"stat.ML","fetchedAt":"2026-10-22T10:01:48"}]},{"id":"physics","slug":"physics","category":"formal-physical","categoryName":{"zh":"数理科学","en":"Mathematical & physical sciences"},"name":{"zh":"物理学","en":"Physics"},"classification":{"oecdFos":"1.3","moe":"0702 物理学"},"level":2,"levelKey":"breakthrough","levelLabel":{"zh":"局部突破","en":"Local breakthroughs"},"trend":"up","reason":{"zh":"AI 已在量子纠错解码（AlphaQubit，Nature 2024）、自主推断物理规律（SciExplorer，PRX 2026）、粒子物理模型构建（Commun. Phys. 2026）上取得同行评审成果，但尚未成为主流研究方式。","en":"Peer-reviewed results in quantum error-correction decoding (AlphaQubit, Nature 2024), autonomous law discovery (SciExplorer, PRX 2026) and particle-model building (Commun. Phys. 2026); not yet the mainstream way physics is done."},"reasonSources":[{"url":"https://www.nature.com/articles/s41586-024-08148-8","linkCheck":"ok"},{"url":"https://journals.aps.org/prx/abstract/10.1103/xnqc-q6nt","linkCheck":"ok"}],"directions":{"zh":["量子纠错与控制","物理规律自主发现","机器学习势函数与模拟","对撞机数据分析"],"en":["Quantum error correction & control","Autonomous law discovery","ML potentials & simulation","Collider data analysis"]},"stats":{"milestones":11,"milestonesSince2025":8,"milestones2026":8,"milestonesLast6Months":7,"news":9,"latestProgress":"2026-10-20","latestMilestone":"2026-10-16","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":0,"leaps":{"ai":1,"field":0},"latestPapers":7,"keyPapers":7},"milestones":[{"date":"2020-04-17","title":{"zh":"“AI Feynman”：物理启发的符号回归自动发现物理公式（Science Advances）","en":"AI Feynman: physics-inspired symbolic regression (Science Advances)"},"url":"https://doi.org/10.1126/sciadv.aay2631","source":"Science Advances","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"从数据中重新发现费曼讲义中的 100 个方程，开启“AI 发现物理定律”路线","en":"Rediscovered all 100 Feynman-lecture equations from data"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"从数据中重新发现费曼讲义中的 100 个方程，开启“AI 发现物理定律”路线","en":"Rediscovered all 100 Feynman-lecture equations from data"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"从数据中重新发现费曼讲义中的 100 个方程，开启“AI 发现物理定律”路线","en":"Rediscovered all 100 Feynman-lecture equations from data"}},"impactNotes":[],"verified":{"id":"10.1126/sciadv.aay2631","title":"AI Feynman: A physics-inspired method for symbolic regression","venue":"Science Advances","date":"2020-04-17"},"addedIn":"v2"},{"date":"2024-10-08","title":{"zh":"诺贝尔物理学奖授予 Hopfield 与 Hinton（神经网络的物理基础）","en":"Nobel Prize in Physics to Hopfield & Hinton (foundations of neural networks)"},"url":"https://www.nobelprize.org/prizes/physics/2024/summary/","source":"Nobel Prize","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"prize","typeLabel":{"zh":"奖项","en":"Prize"},"why":{"zh":"诺贝尔物理学奖首次授予神经网络基础研究，标志 AI 被物理学界正式承认","en":"First physics Nobel for neural-network foundations"},"aiImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"诺贝尔物理学奖首次授予神经网络基础研究，标志 AI 被物理学界正式承认","en":"First physics Nobel for neural-network foundations"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"诺贝尔物理学奖首次授予神经网络基础研究，标志 AI 被物理学界正式承认","en":"First physics Nobel for neural-network foundations"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2024-11-20","title":{"zh":"AlphaQubit 量子纠错解码优于现有方法（Nature）","en":"AlphaQubit decodes quantum error-correcting codes better than prior decoders (Nature)"},"url":"https://www.nature.com/articles/s41586-024-08148-8","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"神经网络解码在真实 Sycamore 数据上超过此前最好的解码器，是容错量子计算的关键环节","en":"Beats state-of-the-art decoders on real Sycamore data"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"神经网络解码在真实 Sycamore 数据上超过此前最好的解码器，是容错量子计算的关键环节","en":"Beats state-of-the-art decoders on real Sycamore data"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"神经网络解码在真实 Sycamore 数据上超过此前最好的解码器，是容错量子计算的关键环节","en":"Beats state-of-the-art decoders on real Sycamore data"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-024-08148-8","title":"Learning high-accuracy error decoding for quantum processors","venue":"Nature","date":"2024-11-20"},"addedIn":"v1"},{"date":"2026-01-21","title":{"zh":"机器学习的重整化群改进格点规范作用量：梯度流“经典完美”（PRL）","en":"Machine-learned RG-improved gauge action with classically perfect gradient flow (PRL)"},"url":"https://doi.org/10.1103/k41k-2pnc","source":"Physical Review Letters","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"ML 进入格点 QCD 核心计算，粗格点下离散化误差<1%","en":"ML enters core lattice-QCD computation; <1% discretisation effects on coarse lattices"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"ML 进入格点 QCD 核心计算，粗格点下离散化误差<1%","en":"ML enters core lattice-QCD computation; <1% discretisation effects on coarse lattices"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"ML 进入格点 QCD 核心计算，粗格点下离散化误差<1%","en":"ML enters core lattice-QCD computation; <1% discretisation effects on coarse lattices"}},"impactNotes":[],"verified":{"id":"10.1103/k41k-2pnc","title":"Machine-Learned Renormalization-Group-Improved Gauge Actions and Classically Perfect Gradient Flows","venue":"Physical Review Letters","date":"2026-01-21"},"addedIn":"v2"},{"date":"2026-05-01","title":{"zh":"AMBer：强化学习智能体自动构建中微子味模型（Communications Physics）","en":"AMBer: RL agent builds neutrino-flavour models (Communications Physics)"},"url":"https://www.nature.com/articles/s42005-026-02627-2","source":"Communications Physics","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"强化学习自动构建中微子味模型，属模型搜索工具","en":"RL search over flavour models"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"强化学习自动构建中微子味模型，属模型搜索工具","en":"RL search over flavour models"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"强化学习自动构建中微子味模型，属模型搜索工具","en":"RL search over flavour models"}},"impactNotes":[],"verified":{"id":"10.1038/s42005-026-02627-2","title":"Towards AI-assisted neutrino flavor theory design","venue":"Communications Physics","date":"2026-05-01"},"addedIn":"v1"},{"date":"2026-06-17","title":{"zh":"机器学习筛选并经实验发现笼目超导体 YRu3B2、LuRu3B2（PR Research）","en":"ML-guided discovery of kagome superconductors YRu3B2 and LuRu3B2 (PR Research)"},"url":"https://doi.org/10.1103/lpqj-7hyg","source":"Physical Review Research","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"AI 预测的新超导相获磁化、比热、输运测量三重确认","en":"AI-predicted superconductors confirmed by three independent measurements"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI 预测的新超导相获磁化、比热、输运测量三重确认","en":"AI-predicted superconductors confirmed by three independent measurements"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI 预测的新超导相获磁化、比热、输运测量三重确认","en":"AI-predicted superconductors confirmed by three independent measurements"}},"impactNotes":[],"verified":{"id":"10.1103/lpqj-7hyg","title":"Machine-learning-guided discovery of kagome superconductors\n                    \n                      \n                        \n                          YRu\n                          3\n                        \n                        \n                          B\n                          2\n                        \n                      \n                    \n                    and\n                    \n                      \n                        \n                          LuRu\n                          3\n                        \n                        \n                          B\n                          2\n                        \n                      \n                    ","venue":"Physical Review Research","date":"2026-06-17"},"addedIn":"v2"},{"date":"2026-07-08","title":{"zh":"SciExplorer：大模型智能体自主推断运动方程与哈密顿量（Physical Review X）","en":"SciExplorer: LLM agent infers equations of motion and Hamiltonians (Physical Review X)"},"url":"https://journals.aps.org/prx/abstract/10.1103/xnqc-q6nt","source":"Physical Review X","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"大模型智能体从实验数据自主推断运动方程和哈密顿量（PRX）","en":"LLM agent infers equations of motion from data (PRX)"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"大模型智能体从实验数据自主推断运动方程和哈密顿量（PRX）","en":"LLM agent infers equations of motion from data (PRX)"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"大模型智能体从实验数据自主推断运动方程和哈密顿量（PRX）","en":"LLM agent infers equations of motion from data (PRX)"}},"impactNotes":[],"verified":{"id":"10.1103/xnqc-q6nt","title":"Agentic Exploration of Physics Models","venue":"Physical Review X","date":"2026-07-08"},"addedIn":"v1"},{"date":"2026-10-01","title":{"zh":"AI Theorist 为 α-RuCl3 未发表实验给出新的激子解释（预印本）","en":"AI Theorist proposes new exciton model for α-RuCl3 data (preprint)"},"url":"https://arxiv.org/abs/2610.02417","source":"arXiv","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"AI 理论家为未发表实验数据提出新解释，尚未同行评审","en":"AI theorist proposes a new model; not yet peer-reviewed"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI 理论家为未发表实验数据提出新解释，尚未同行评审","en":"AI theorist proposes a new model; not yet peer-reviewed"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI 理论家为未发表实验数据提出新解释，尚未同行评审","en":"AI theorist proposes a new model; not yet peer-reviewed"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-10-09","title":{"zh":"TheoPhysEval：50 道未发表的研究级理论物理题，GPT-6 Astra 智能体解出 49 道","en":"TheoPhysEval: 50 unpublished research-level theory problems; GPT-6 Astra agent solves 49"},"url":"https://arxiv.org/abs/2610.16733","source":"arXiv","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok","datePrecision":"day","type":"benchmark","typeLabel":{"zh":"基准","en":"Benchmark"},"why":{"zh":"AI：首个以未发表研究级题目检验自主智能体的理论物理基准，联网 98%、断网 92%；学科：作者指出题目偏“按部就班计算”，不代表能做出新发现","en":"AI: first benchmark of unpublished research-level theory problems for autonomous agents (98% with tools); Field: problems are routine calculations, not discoveries"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：首个以未发表研究级题目检验自主智能体的理论物理基准，联网 98%、断网 92%；学科：作者指出题目偏“按部就班计算”，不代表能做出新发现","en":"AI: first benchmark of unpublished research-level theory problems for autonomous agents (98% with tools); Field: problems are routine calculations, not discoveries"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：首个以未发表研究级题目检验自主智能体的理论物理基准，联网 98%、断网 92%；学科：作者指出题目偏“按部就班计算”，不代表能做出新发现","en":"AI: first benchmark of unpublished research-level theory problems for autonomous agents (98% with tools); Field: problems are routine calculations, not discoveries"}},"impactNotes":[],"verified":{"id":"arXiv:2610.16733","title":"TheoPhysEval: Assessing AI agents at autonomous theoretical physics problem solving","venue":"arXiv","date":"2026-10-09"},"addedIn":"v2"},{"date":"2026-10-15","title":{"zh":"CERN 与 Mistral 签署意向书，联合开发首个对撞机物理基础模型“大物理模型”","en":"CERN and Mistral sign a statement of intent to build the first foundation model for collider physics"},"url":"https://home.cern/cern-and-mistral-join-forces-to-accelerate-ai-for-science/","source":"CERN","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"industry","typeLabel":{"zh":"产业/采用","en":"Industry"},"why":{"zh":"AI：用数十亿次模拟 LHC 碰撞预训练单一模型，覆盖重建、粒子鉴别与异常信号；学科：粒子物理首次由 CERN 官方推动用基础模型替代分段重建链，尚处意向阶段","en":"AI: one model pretrained on billions of simulated LHC collisions; Field: CERN officially pursues replacing the staged reconstruction chain (statement of intent only)"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：用数十亿次模拟 LHC 碰撞预训练单一模型，覆盖重建、粒子鉴别与异常信号；学科：粒子物理首次由 CERN 官方推动用基础模型替代分段重建链，尚处意向阶段","en":"AI: one model pretrained on billions of simulated LHC collisions; Field: CERN officially pursues replacing the staged reconstruction chain (statement of intent only)"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：用数十亿次模拟 LHC 碰撞预训练单一模型，覆盖重建、粒子鉴别与异常信号；学科：粒子物理首次由 CERN 官方推动用基础模型替代分段重建链，尚处意向阶段","en":"AI: one model pretrained on billions of simulated LHC collisions; Field: CERN officially pursues replacing the staged reconstruction chain (statement of intent only)"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-10-16","title":{"zh":"Anthropic：Claude 在自动化循环中设计出更优量子电路，椭圆曲线点加法门数减少约 17 倍","en":"Anthropic: Claude designs more efficient quantum circuits, ~17× fewer gates for elliptic-curve point addition"},"url":"https://www.anthropic.com/research/designing-efficient-quantum-circuits","source":"Anthropic","flags":["vendor","preprint"],"flagLabels":[{"zh":"厂商自述","en":"Vendor claim"},{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok","datePrecision":"day","type":"discovery","typeLabel":{"zh":"科学发现","en":"Discovery"},"why":{"zh":"AI：模型从自然语言规格出发，在 8 项任务上产出优于已有设计的电路（arXiv 2610.22367）；学科：降低容错量子算法的门数和比特开销，但属厂商自述、尚未同行评审","en":"AI: from plain-language specs the model beat prior circuits on 8 tasks (arXiv 2610.22367); Field: cuts gate/qubit costs for fault-tolerant algorithms (vendor, not peer-reviewed)"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：模型从自然语言规格出发，在 8 项任务上产出优于已有设计的电路（arXiv 2610.22367）；学科：降低容错量子算法的门数和比特开销，但属厂商自述、尚未同行评审","en":"AI: from plain-language specs the model beat prior circuits on 8 tasks (arXiv 2610.22367); Field: cuts gate/qubit costs for fault-tolerant algorithms (vendor, not peer-reviewed)"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：模型从自然语言规格出发，在 8 项任务上产出优于已有设计的电路（arXiv 2610.22367）；学科：降低容错量子算法的门数和比特开销，但属厂商自述、尚未同行评审","en":"AI: from plain-language specs the model beat prior circuits on 8 tasks (arXiv 2610.22367); Field: cuts gate/qubit costs for fault-tolerant algorithms (vendor, not peer-reviewed)"}},"impactNotes":[],"verified":null,"addedIn":"v2"}],"papersRepos":[{"title":"The AI Theorist reveals excitonic structure in α-RuCl3 (arXiv 2610.02417)","url":"https://arxiv.org/abs/2610.02417","date":"2026-10-01","kind":"paper","linkCheck":"ok"},{"title":"AI Feynman: A physics-inspired method for symbolic regression","url":"https://doi.org/10.1126/sciadv.aay2631","date":"2020-04-17","kind":"paper","linkCheck":"ok-crossref"},{"title":"Learning high-accuracy error decoding for quantum processors","url":"https://doi.org/10.1038/s41586-024-08148-8","date":"2024-11-20","kind":"paper","linkCheck":"ok-crossref"},{"title":"Machine-Learned Renormalization-Group-Improved Gauge Actions and Classically Perfect Gradient Flows","url":"https://doi.org/10.1103/k41k-2pnc","date":"2026-01-21","kind":"paper","linkCheck":"ok-crossref"},{"title":"Machine-learning-guided discovery of kagome superconductors\n                    \n                      \n                        \n                          YRu\n                          3\n                        \n                        \n                          B\n                          2\n                        \n                      \n                    \n                    and\n                    \n                      \n                        \n                          LuRu\n                          3\n                        \n                        \n                          B\n                          2\n                        \n                      \n                    ","url":"https://doi.org/10.1103/lpqj-7hyg","date":"2026-06-17","kind":"paper","linkCheck":"ok-crossref"},{"title":"Agentic Exploration of Physics Models","url":"https://doi.org/10.1103/xnqc-q6nt","date":"2026-07-08","kind":"paper","linkCheck":"ok-crossref"},{"title":"TheoPhysEval: Assessing AI agents at autonomous theoretical physics problem solving","url":"https://arxiv.org/abs/2610.16733","date":"2026-10-09","kind":"paper","linkCheck":"ok-crossref"},{"title":"Towards AI-assisted neutrino flavor theory design","url":"https://doi.org/10.1038/s42005-026-02627-2","date":"2026-05-01","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"Google Quantum AI","url":"https://quantumai.google"},{"name":"Max Planck Institute for the Science of Light","url":"https://mpl.mpg.de"},{"name":"CERN","url":"https://home.cern"},{"name":"IAIFI","url":"https://iaifi.org"}],"news":[{"id":"6b8d0d0f060c","date":"2026-10-20","title":"NVIDIA Runs Record Fault-Tolerant Quantum Simulation at Jülich Supercomputing Centre","url":"https://www.hpcwire.com/off-the-wire/nvidia-runs-record-fault-tolerant-quantum-simulation-at-julich-supercomputing-centre/","source":"HPCwire","lang":"en","kind":"news","summaryZh":"HPCwire于2026-10-20发布的文章，涉及AI 芯片硬件，原题《NVIDIA Runs Record Fault-Tolerant Quantum Simulation at Jülich Supercomputing Centre》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"12d10c346fbc","date":"2026-10-20","title":"Measurement-efficient energy estimator for quantum chemistry of photodynamic materials","url":"https://arxiv.org/abs/2610.26689","source":"arXiv physics.chem-ph","lang":"en","kind":"paper","summaryZh":"arXiv physics.chem-ph于2026-10-20发布的论文，涉及AI 化学材料、AI 气候能源、AI 软件开发，原题《Measurement-efficient energy estimator for quantum chemistry of photodynamic materials》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"c861f4d741de","date":"2026-10-19","title":"从百比特芯片到整机交付，这家量子计算公司正将量子算力推向可调用服务","url":"https://www.qbitai.com/2026/10/504277.html","source":"量子位","lang":"zh","kind":"news","summaryZh":"量子位于2026-10-19发布的文章，涉及AI 芯片硬件，原题《从百比特芯片到整机交付，这家量子计算公司正将量子算力推向可调用服务》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"43fe977bbf8e","date":"2026-10-19","title":"NVIDIA Runs Record Simulation of Fault-Tolerant Quantum Computing at Jülich Supercomputing Centre","url":"https://nvidianews.nvidia.com/news/nvidia-runs-record-simulation-of-fault-tolerant-quantum-computing-at-julich-supercomputing-centre","source":"NVIDIA Newsroom","lang":"en","kind":"news","summaryZh":"NVIDIA Newsroom于2026-10-20发布的文章，涉及AI 芯片硬件，原题《NVIDIA Runs Record Simulation of Fault-Tolerant Quantum Computing at Jülich Supercomputing》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"c53a997b4db8","date":"2026-10-19","title":"光量子计算，距离真正的产业化还有多远？","url":"https://www.leiphone.com/category/chips/1LbhTsMb0OyPhsw6.html","source":"雷峰网","lang":"zh","kind":"news","summaryZh":"雷峰网于2026-10-19发布的文章，涉及AI 综合、AI 科研与调研、AI 化学材料，原题《光量子计算，距离真正的产业化还有多远？》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"a1b3fb3caae8","date":"2026-10-16","title":"Can AI help discover new food physics? Researchers pair AI with human expertise","url":"https://techxplore.com/news/2026-10-ai-food-physics-pair-human.html","source":"Phys.org 机器学习与AI","lang":"en","kind":"news","summaryZh":"Phys.org 机器学习与AI于2026-10-17发布的文章，涉及AI 科研与调研、AI 安全，原题《Can AI help discover new food physics? Researchers pair AI with human expertise》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"6bbbb40e1e0b","date":"2026-10-16","title":"Physics-guided AI reconstructs hidden patterns inside ultrathin magnets from external fields","url":"https://phys.org/news/2026-10-physics-ai-reconstructs-hidden-patterns.html","source":"Phys.org 材料科学","lang":"en","kind":"news","summaryZh":"Phys.org 材料科学于2026-10-16发布的文章，涉及AI 化学材料、AI 气候能源，原题《Physics-guided AI reconstructs hidden patterns inside ultrathin magnets from external fiel》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"4610d2d935ba","date":"2026-08-10","title":"With a feel for physics, AI models simulate a wider range of real-world scenarios","url":"https://news.mit.edu/2026/ai-models-simulate-wider-range-of-real-world-scenarios-0810","source":"MIT News：Artificial intelligence","lang":"en","kind":"news","summaryZh":"MIT News：Artificial intelligence于2026-08-11发布的文章，涉及AI 科研与调研，原题《With a feel for physics, AI models simulate a wider range of real-world scenarios》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"1fb27f68e42a","date":"2026-07-22","title":"Towards a quantum computer that learns from its errors","url":"https://research.google/blog/towards-a-quantum-computer-that-learns-from-its-errors/","source":"Google Research Blog","lang":"en","kind":"news","summaryZh":"Google Research Blog于2026-07-23发布的文章，涉及AI 科研与调研，原题《Towards a quantum computer that learns from its errors》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"}],"review":{"lastReviewed":"2026-10-21","cycleDays":90,"nextReview":"2027-01-19"},"urls":{"zh":"/zh-hans/disciplines/physics/","en":"/en/disciplines/physics/"},"dimensions":{"method":{"score":2,"label":{"zh":"研究方法","en":"Research method"},"evidence":"机器学习势、神经网络解码、格点 QCD 中的 ML 作用量已进入主流期刊","source":"https://link.aps.org/doi/10.1103/k41k-2pnc","weight":0.3},"discovery":{"score":2,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"AI 预测的新超导体获实验证实；智能体自主推断物理规律（PRX 2026）；但没有改变学科格局的发现","source":"https://doi.org/10.1103/lpqj-7hyg","weight":0.35},"adoption":{"score":2,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"粒子物理和天体物理大型实验普遍用 ML 做触发与分析；理论物理采用有限","source":null,"weight":0.25},"education":{"score":1,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到课程层面的证据","source":null,"weight":0.1}},"compositeScore":1.9,"levelCapped":false,"levelChange":{"from":2,"to":2,"direction":"same","reason":null},"biggestBreakthrough":{"date":"2024-10-08","title":{"zh":"诺贝尔物理学奖授予 Hopfield 与 Hinton（神经网络的物理基础）","en":"Nobel Prize in Physics to Hopfield & Hinton (foundations of neural networks)"},"url":"https://www.nobelprize.org/prizes/physics/2024/summary/","aiImpact":"leap","fieldImpact":"major","why":{"zh":"诺贝尔物理学奖首次授予神经网络基础研究，标志 AI 被物理学界正式承认","en":"First physics Nobel for neural-network foundations"}},"keyPapers":[{"title":"AI Feynman: A physics-inspired method for symbolic regression","titleZh":"“AI Feynman”：物理启发的符号回归自动发现物理公式（Science Advances）","id":"10.1126/sciadv.aay2631","url":"https://doi.org/10.1126/sciadv.aay2631","venue":"Science Advances","date":"2020-04-17","why":{"zh":"从数据中重新发现费曼讲义中的 100 个方程，开启“AI 发现物理定律”路线","en":"Rediscovered all 100 Feynman-lecture equations from data"},"fieldImpact":"incremental","aiImpact":"major"},{"title":"Learning high-accuracy error decoding for quantum processors","titleZh":"AlphaQubit 量子纠错解码优于现有方法（Nature）","id":"10.1038/s41586-024-08148-8","url":"https://doi.org/10.1038/s41586-024-08148-8","venue":"Nature","date":"2024-11-20","why":{"zh":"神经网络解码在真实 Sycamore 数据上超过此前最好的解码器，是容错量子计算的关键环节","en":"Beats state-of-the-art decoders on real Sycamore data"},"fieldImpact":"major","aiImpact":"major"},{"title":"Machine-Learned Renormalization-Group-Improved Gauge Actions and Classically Perfect Gradient Flows","titleZh":"机器学习的重整化群改进格点规范作用量：梯度流“经典完美”（PRL）","id":"10.1103/k41k-2pnc","url":"https://doi.org/10.1103/k41k-2pnc","venue":"Physical Review Letters","date":"2026-01-21","why":{"zh":"ML 进入格点 QCD 核心计算，粗格点下离散化误差<1%","en":"ML enters core lattice-QCD computation; <1% discretisation effects on coarse lattices"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Machine-learning-guided discovery of kagome superconductors\n                    \n                      \n                        \n                          YRu\n                          3\n                        \n                        \n                          B\n                          2\n                        \n                      \n                    \n                    and\n                    \n                      \n                        \n                          LuRu\n                          3\n                        \n                        \n                          B\n                          2\n                        \n                      \n                    ","titleZh":"机器学习筛选并经实验发现笼目超导体 YRu3B2、LuRu3B2（PR Research）","id":"10.1103/lpqj-7hyg","url":"https://doi.org/10.1103/lpqj-7hyg","venue":"Physical Review Research","date":"2026-06-17","why":{"zh":"AI 预测的新超导相获磁化、比热、输运测量三重确认","en":"AI-predicted superconductors confirmed by three independent measurements"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Agentic Exploration of Physics Models","titleZh":"SciExplorer：大模型智能体自主推断运动方程与哈密顿量（Physical Review X）","id":"10.1103/xnqc-q6nt","url":"https://doi.org/10.1103/xnqc-q6nt","venue":"Physical Review X","date":"2026-07-08","why":{"zh":"大模型智能体从实验数据自主推断运动方程和哈密顿量（PRX）","en":"LLM agent infers equations of motion from data (PRX)"},"fieldImpact":"incremental","aiImpact":"major"},{"title":"TheoPhysEval: Assessing AI agents at autonomous theoretical physics problem solving","titleZh":"TheoPhysEval：50 道未发表的研究级理论物理题，GPT-6 Astra 智能体解出 49 道","id":"arXiv:2610.16733","url":"https://arxiv.org/abs/2610.16733","venue":"arXiv","date":"2026-10-09","why":{"zh":"AI：首个以未发表研究级题目检验自主智能体的理论物理基准，联网 98%、断网 92%；学科：作者指出题目偏“按部就班计算”，不代表能做出新发现","en":"AI: first benchmark of unpublished research-level theory problems for autonomous agents (98% with tools); Field: problems are routine calculations, not discoveries"},"fieldImpact":"incremental","aiImpact":"major"},{"title":"Towards AI-assisted neutrino flavor theory design","titleZh":"AMBer：强化学习智能体自动构建中微子味模型（Communications Physics）","id":"10.1038/s42005-026-02627-2","url":"https://doi.org/10.1038/s42005-026-02627-2","venue":"Communications Physics","date":"2026-05-01","why":{"zh":"强化学习自动构建中微子味模型，属模型搜索工具","en":"RL search over flavour models"},"fieldImpact":"incremental","aiImpact":"incremental"}],"latest_papers":[{"id":"arXiv:2610.29850","title":"Automated Discovery of Physically Interpretable Subgrid Length Scales for Detached Eddy Simulation via Agentic AI","titleZh":"DEScover：用智能体 AI 自动发现分离涡模拟中可解释的亚格子长度尺度","summaryZh":"用户只给出建模思路，四个大模型智能体分工提出、评审、实现并评估候选公式，自动探索湍流分离涡模拟（DES）的亚格子长度尺度。","summaryEn":"Four LLM agents propose, review, implement and evaluate candidate subgrid length scales for detached eddy simulation from a single modelling idea.","authors":["Jiawei Chen","Xiang I. A. Yang","Ping He"],"date":"2026-10-21","url":"https://arxiv.org/abs/2610.29850","site":"arXiv","venue":null,"category":"physics.flu-dyn","fetchedAt":"2026-10-22T10:01:55"},{"id":"arXiv:2610.23779","title":"Mesoscopic learning dynamics of extensive-width neural networks","titleZh":"宽神经网络的介观学习动力学","summaryZh":"在网络宽度与输入维度同比例增长的情形下，研究随机梯度下降如何让两层神经网络学习内部表示。","summaryEn":"Analyses how SGD learns internal representations in two-layer networks when width and input dimension grow proportionally.","authors":["Shu Zhou","Jean Barbier","Sebastian Goldt"],"date":"2026-10-16","url":"https://arxiv.org/abs/2610.23779","site":"arXiv","venue":null,"category":"stat.ML","fetchedAt":"2026-10-22T10:01:55"},{"id":"arXiv:2610.23776","title":"Active Learning for Interatomic Potentials: A Review of Uncertainty-Driven Data Efficiency","titleZh":"原子间势的主动学习：不确定性驱动的数据效率综述","summaryZh":"综述机器学习原子间势的主动学习方法：在候选生成、数据选择、参考计算和模型更新之间形成闭环，以较少数据达到目标精度。","summaryEn":"Reviews uncertainty-driven active learning that closes the loop for building accurate ML interatomic potentials.","authors":["Ilya V. Chepkasov","Alexander V. Shapeev","Alexander G. Kvashnin"],"date":"2026-10-16","url":"https://arxiv.org/abs/2610.23776","site":"arXiv","venue":null,"category":"physics.comp-ph","fetchedAt":"2026-10-22T10:01:55"},{"id":"arXiv:2610.23080","title":"Properties and challenges of machine learning closures for ten-moment fluid models in the case of Landau damping","titleZh":"朗道阻尼情形下十矩流体模型机器学习闭合的性质与挑战","summaryZh":"用机器学习模型发现等离子体十矩流体模型的热流闭合关系，作为传统 Hammett-Perkins 类解析闭合的替代。","summaryEn":"Uses ML to discover heat-flux closures for ten-moment plasma fluid models as an alternative to analytic Hammett–Perkins closures.","authors":["Emanuel Jeß","Simon Lautenbach","Sophia Köhne","George Miloshevich"],"date":"2026-10-16","url":"https://arxiv.org/abs/2610.23080","site":"arXiv","venue":null,"category":"physics.plasm-ph","fetchedAt":"2026-10-22T10:01:55"},{"id":"arXiv:2610.21395","title":"When do quantum views help? Complementary classical-quantum representations in multi-view fusion","titleZh":"量子视图何时有用？多视图融合中互补的经典-量子表示","summaryZh":"提出把经典与量子机器学习表示融合，在两者捕捉互补关系时提升预测效果。","summaryEn":"Proposes fusing classical and quantum ML representations when they capture complementary predictive relationships.","authors":["Azadeh Alavi","Fatemeh Kouchmeshki","Hossein Akhoundi","Yongli Re"],"date":"2026-10-15","url":"https://arxiv.org/abs/2610.21395","site":"arXiv","venue":null,"category":"quant-ph","fetchedAt":"2026-10-22T10:01:55"},{"id":"arXiv:2610.14999","title":"DiffGA: Steering Frozen Diffusion Models for Materials Design via Evolutionary Search","titleZh":"DiffGA：用进化搜索引导冻结的扩散模型做材料设计","summaryZh":"不重新训练扩散模型，而用进化搜索把候选结构生成引导到黑箱、可变或不可微的材料性能目标。","summaryEn":"Steers a frozen diffusion model with evolutionary search towards black-box materials-design objectives without retraining.","authors":["Shivang Agarwal","Rodrigo Wang","Kevin Ryczko","Aayush R. Singh"],"date":"2026-10-10","url":"https://arxiv.org/abs/2610.14999","site":"arXiv","venue":null,"category":"cond-mat.mtrl-sci","fetchedAt":"2026-10-22T10:01:55"},{"id":"arXiv:2610.14891","title":"Pressure-transferable universal machine-learning potentials: the WBM-MatPES benchmark and Tuga-FF","titleZh":"可跨压力迁移的通用机器学习势：WBM-MatPES 基准与 Tuga-FF","summaryZh":"发布 18.2 万个结构的 WBM-MatPES 数据集和参考凸包，用于评估通用机器学习原子间势在不同压力下的精度与迁移性。","summaryEn":"Releases WBM-MatPES (182k structures) and reference hulls to benchmark universal ML interatomic potentials across pressures.","authors":["Pierre-Paul De Breuck","Thalis H. B. da Silva","Antoine Loew","Silvana Botti"],"date":"2026-10-10","url":"https://arxiv.org/abs/2610.14891","site":"arXiv","venue":null,"category":"cond-mat.mtrl-sci","fetchedAt":"2026-10-22T10:01:55"}]},{"id":"astronomy","slug":"astronomy","category":"formal-physical","categoryName":{"zh":"数理科学","en":"Mathematical & physical sciences"},"name":{"zh":"天文学","en":"Astronomy & astrophysics"},"classification":{"oecdFos":"1.3","moe":"0704 天文学"},"level":2,"levelKey":"breakthrough","levelLabel":{"zh":"局部突破","en":"Local breakthroughs"},"trend":"flat","reason":{"zh":"机器学习已能批量发现系外行星候选与超新星：TESS 数据中找出约 1 万个新候选（ApJS 2026），NASA ExoMiner++ 给出约 7000 个候选；多数仍需人工审查和后续观测确认。","en":"ML finds exoplanet candidates and supernovae at scale (≈10,000 new TESS candidates, ApJS 2026; NASA ExoMiner++ ≈7,000), though most still need vetting and follow-up."},"reasonSources":[{"url":"https://iopscience.iop.org/article/10.3847/1538-4365/ae5b6c","linkCheck":"ok"},{"url":"https://science.nasa.gov/open-science/deep-learning-exoplanets-tess/","linkCheck":"ok"}],"directions":{"zh":["系外行星搜寻","瞬变源实时分类","巡天数据异常检测"],"en":["Exoplanet search","Real-time transient classification","Survey anomaly detection"]},"stats":{"milestones":9,"milestonesSince2025":6,"milestones2026":6,"milestonesLast6Months":5,"news":3,"latestProgress":"2026-10-15","latestMilestone":"2026-10-15","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":1,"leaps":{"ai":0,"field":0},"latestPapers":8,"keyPapers":5},"milestones":[{"date":"2018-01-30","title":{"zh":"神经网络在 Kepler 数据中发现 Kepler-90i（AJ；NASA 2017-12-14 公布）","en":"Neural network finds Kepler-90i in Kepler data (AJ)"},"url":"https://iopscience.iop.org/article/10.3847/1538-3881/aa9e09","source":"The Astronomical Journal","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"神经网络在 Kepler 数据中找到第 8 颗行星，Kepler-90 成为与太阳系并列行星最多的系统","en":"Neural net finds Kepler-90i, tying the Solar System’s planet count"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"神经网络在 Kepler 数据中找到第 8 颗行星，Kepler-90 成为与太阳系并列行星最多的系统","en":"Neural net finds Kepler-90i, tying the Solar System’s planet count"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"神经网络在 Kepler 数据中找到第 8 颗行星，Kepler-90 成为与太阳系并列行星最多的系统","en":"Neural net finds Kepler-90i, tying the Solar System’s planet count"}},"impactNotes":[],"verified":{"id":"10.3847/1538-3881/aa9e09","title":"Identifying Exoplanets with Deep Learning: A Five-planet Resonant Chain around Kepler-80 and an Eighth Planet around Kepler-90","venue":"The Astronomical Journal","date":"2018-01-30"},"addedIn":"v1"},{"date":"2022-02-17","title":{"zh":"ExoMiner：深度学习一次验证 301 颗 Kepler 新行星（ApJ）","en":"ExoMiner validates 301 new Kepler exoplanets (ApJ)"},"url":"https://doi.org/10.3847/1538-4357/ac4399","source":"The Astrophysical Journal","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"NASA 用神经网络一次性统计验证 301 颗行星，确立 AI 行星验证流程","en":"NASA neural network statistically validates 301 planets at once"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"NASA 用神经网络一次性统计验证 301 颗行星，确立 AI 行星验证流程","en":"NASA neural network statistically validates 301 planets at once"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"NASA 用神经网络一次性统计验证 301 颗行星，确立 AI 行星验证流程","en":"NASA neural network statistically validates 301 planets at once"}},"impactNotes":[],"verified":{"id":"10.3847/1538-4357/ac4399","title":"ExoMiner: A Highly Accurate and Explainable Deep Learning Classifier That Validates 301 New Exoplanets","venue":"The Astrophysical Journal","date":"2022-02-17"},"addedIn":"v2"},{"date":"2023-01-10","title":{"zh":"综述：深度学习对星系巡天分析的影响（PASA Dawes Review）","en":"Dawes Review: the impact of deep learning on galaxy-survey analysis (PASA)"},"url":"https://doi.org/10.1017/pasa.2022.55","source":"PASA","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"系统梳理深度学习成为巡天标准工具的过程","en":"Maps deep learning becoming a standard survey tool"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"系统梳理深度学习成为巡天标准工具的过程","en":"Maps deep learning becoming a standard survey tool"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"系统梳理深度学习成为巡天标准工具的过程","en":"Maps deep learning becoming a standard survey tool"}},"impactNotes":[],"verified":{"id":"10.1017/pasa.2022.55","title":"The Dawes Review 10: The impact of deep learning for the analysis of galaxy surveys","venue":"Publications of the Astronomical Society of Australia","date":"2023-01-10"},"addedIn":"v2"},{"date":"2026-01-22","title":{"zh":"NASA ExoMiner++ 在 TESS 数据中给出约 7000 个候选","en":"NASA ExoMiner++ flags ~7,000 TESS candidates"},"url":"https://science.nasa.gov/open-science/deep-learning-exoplanets-tess/","source":"NASA Science","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"dataset","typeLabel":{"zh":"数据集","en":"Dataset"},"why":{"zh":"约 7000 个 TESS 候选，扩大后续观测目标池","en":"~7,000 TESS candidates"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"约 7000 个 TESS 候选，扩大后续观测目标池","en":"~7,000 TESS candidates"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"约 7000 个 TESS 候选，扩大后续观测目标池","en":"~7,000 TESS candidates"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-04-28","title":{"zh":"T16 行星搜寻：TESS 第 1 周期约 1 万个新行星候选（ApJS）","en":"T16 planet hunt: ~10,000 new TESS Cycle-1 candidates (ApJS)"},"url":"https://iopscience.iop.org/article/10.3847/1538-4365/ae5b6c","source":"ApJS","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"约 1 万个新候选，规模远超人工搜寻","en":"~10,000 new candidates"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"约 1 万个新候选，规模远超人工搜寻","en":"~10,000 new candidates"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"约 1 万个新候选，规模远超人工搜寻","en":"~10,000 new candidates"}},"impactNotes":[],"verified":{"id":"10.3847/1538-4365/ae5b6c","title":"The T16 Planet Hunt: 10,000 New Planet Candidates from TESS Cycle 1 and the Confirmation of a Hot Jupiter around TIC 183374187*","venue":"The Astrophysical Journal Supplement Series","date":"2026-04-28"},"addedIn":"v1"},{"date":"2026-09","title":{"zh":"NOMAI：Fink 平台实时识别超亮超新星（A&A）","en":"NOMAI: real-time superluminous-supernova classifier in Fink (A&A)"},"url":"https://www.aanda.org/articles/aa/full_html/2026/09/aa60399-26/aa60399-26.html","source":"A&A","flags":["tbv"],"flagLabels":[{"zh":"日期或细节待核实","en":"Date/details to be verified"}],"linkCheck":"site-blocks-bots","datePrecision":"month","type":"deployment","typeLabel":{"zh":"落地部署","en":"Deployment"},"why":{"zh":"实时警报代理中用 ML 识别超亮超新星","en":"ML in real-time alert brokering"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"实时警报代理中用 ML 识别超亮超新星","en":"ML in real-time alert brokering"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"实时警报代理中用 ML 识别超亮超新星","en":"ML in real-time alert brokering"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-10-08","title":{"zh":"Anthropic 用 Claude 生成首张完整紫外全天图，补全约三分之一未观测天区（经校准预测）","en":"Anthropic uses Claude to produce the first complete UV sky map, filling ~1/3 unobserved sky with calibrated predictions"},"url":"https://www.anthropic.com/research/the-missing-map-of-the-sky","source":"Anthropic","flags":["vendor"],"flagLabels":[{"zh":"厂商自述","en":"Vendor claim"}],"linkCheck":"ok","datePrecision":"day","type":"dataset","typeLabel":{"zh":"数据集","en":"Dataset"},"why":{"zh":"大模型智能体完成天文数据产品的端到端制作；预测天区不是真实观测，需谨慎使用","en":"Agent-built data product; predicted regions are not observations"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"大模型智能体完成天文数据产品的端到端制作；预测天区不是真实观测，需谨慎使用","en":"Agent-built data product; predicted regions are not observations"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"大模型智能体完成天文数据产品的端到端制作；预测天区不是真实观测，需谨慎使用","en":"Agent-built data product; predicted regions are not observations"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-10-14","title":{"zh":"业余天文学家借助 AI 发现一颗可能隐藏的系外行星","en":"Amateur astronomer uses AI to find a possible hidden exoplanet"},"url":"https://www.scientificamerican.com/article/amateur-astronomer-harnesses-ai-to-find-a-possible-hidden-exoplanet/","source":"Scientific American","flags":[],"flagLabels":[],"linkCheck":"site-blocks-bots","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"AI 工具让非专业人员也能参与前沿搜寻","en":"AI tools let amateurs join frontier searches"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI 工具让非专业人员也能参与前沿搜寻","en":"AI tools let amateurs join frontier searches"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI 工具让非专业人员也能参与前沿搜寻","en":"AI tools let amateurs join frontier searches"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-10-15","title":{"zh":"NA-SODINN 深度学习重分析 VLT/SPHERE SHINE 巡天 150 颗恒星，找到 13 个此前未报告的亚恒星伴星候选（A&A）","en":"Deep learning (NA-SODINN) re-analysis of the SHINE F150 sample finds 13 new substellar candidates (A&A)"},"url":"https://doi.org/10.1051/0004-6361/202660463","source":"Astronomy & Astrophysics","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"discovery","typeLabel":{"zh":"科学发现","en":"Discovery"},"why":{"zh":"AI：监督式深度学习在角差分成像序列中找出经典方法漏掉的弱信号，并找回全部已知伴星；学科：目前仅 Smethells 20 周围的候选值得优先跟进观测，属候选而非确认","en":"AI: recovers all known companions and finds faint signals missed by classical pipelines. Field: candidates only; one (Smethells 20) remains strong"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：监督式深度学习在角差分成像序列中找出经典方法漏掉的弱信号，并找回全部已知伴星；学科：目前仅 Smethells 20 周围的候选值得优先跟进观测，属候选而非确认","en":"AI: recovers all known companions and finds faint signals missed by classical pipelines. Field: candidates only; one (Smethells 20) remains strong"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：监督式深度学习在角差分成像序列中找出经典方法漏掉的弱信号，并找回全部已知伴星；学科：目前仅 Smethells 20 周围的候选值得优先跟进观测，属候选而非确认","en":"AI: recovers all known companions and finds faint signals missed by classical pipelines. Field: candidates only; one (Smethells 20) remains strong"}},"impactNotes":[],"verified":{"id":"10.1051/0004-6361/202660463","title":"New substellar candidates identified through deep learning in the F150 sample of the large-scale SHINE direct imaging survey","venue":"Astronomy & Astrophysics","date":"2026-10-15"},"addedIn":"v2-20261022pm"}],"papersRepos":[{"title":"SNAD: enabling discovery in the era of big data (arXiv 2608.30477)","url":"https://arxiv.org/abs/2608.30477","date":"2026-08","kind":"paper","linkCheck":"ok"},{"title":"Identifying Exoplanets with Deep Learning: A Five-planet Resonant Chain around Kepler-80 and an Eighth Planet around Kepler-90","url":"https://doi.org/10.3847/1538-3881/aa9e09","date":"2018-01-30","kind":"paper","linkCheck":"ok-crossref"},{"title":"ExoMiner: A Highly Accurate and Explainable Deep Learning Classifier That Validates 301 New Exoplanets","url":"https://doi.org/10.3847/1538-4357/ac4399","date":"2022-02-17","kind":"paper","linkCheck":"ok-crossref"},{"title":"The T16 Planet Hunt: 10,000 New Planet Candidates from TESS Cycle 1 and the Confirmation of a Hot Jupiter around TIC 183374187*","url":"https://doi.org/10.3847/1538-4365/ae5b6c","date":"2026-04-28","kind":"paper","linkCheck":"ok-crossref"},{"title":"The Dawes Review 10: The impact of deep learning for the analysis of galaxy surveys","url":"https://doi.org/10.1017/pasa.2022.55","date":"2023-01-10","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"NASA","url":"https://science.nasa.gov"},{"name":"Vera C. Rubin Observatory","url":"https://rubinobservatory.org"},{"name":"Fink broker","url":"https://fink-broker.org"}],"news":[{"id":"9e41130d4a12","date":"2026-10-14","title":"Amateur astronomer harnesses AI to find a possible hidden exoplanet","url":"https://www.scientificamerican.com/article/amateur-astronomer-harnesses-ai-to-find-a-possible-hidden-exoplanet/","source":"Scientific American","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"site-blocks-bots"},{"id":"2999c3c463fc","date":"2026-10-08","title":"Using Claude Science to produce the first complete map of the sky in UV light","url":"https://www.anthropic.com/research/the-missing-map-of-the-sky","source":"Anthropic","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"ok"},{"id":"c9bfef970774","date":"2026-10-07","title":"Largest 2D Map of the Universe Helps Scientists Discover New Cosmic Magnifying Glasses","url":"https://noirlab.edu/public/blog/desi-legacy-imaging-surveys/","source":"NSF NOIRLab","lang":"en","kind":"news","summaryZh":"团队用残差神经网络在 DESI Legacy 巡天数据中筛出引力透镜候选，光谱确认了 70 个新引力透镜（发表于 ApJ）。","via":"manual-websearch-20261021","linkCheck":"ok"}],"review":{"lastReviewed":"2026-10-22","cycleDays":90,"nextReview":"2027-01-19"},"urls":{"zh":"/zh-hans/disciplines/astronomy/","en":"/en/disciplines/astronomy/"},"dimensions":{"method":{"score":3,"label":{"zh":"研究方法","en":"Research method"},"evidence":"深度学习已是巡天数据分类、瞬变源和系外行星搜寻的标准方法","source":"https://iopscience.iop.org/article/10.3847/1538-4365/ae5b6c","weight":0.3},"discovery":{"score":2,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"批量候选体发现（约 1 万个 TESS 候选、强引力透镜）需后续观测确认","source":"https://www.osti.gov/pages/biblio/3818034","weight":0.35},"adoption":{"score":3,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"Rubin 天文台等实时警报代理（Fink 等）用 ML 做业务化分类","source":"https://fink-broker.org","weight":0.25},"education":{"score":1,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到课程证据","source":null,"weight":0.1}},"compositeScore":2.45,"levelCapped":false,"levelChange":{"from":2,"to":2,"direction":"same","reason":null},"biggestBreakthrough":{"date":"2018-01-30","title":{"zh":"神经网络在 Kepler 数据中发现 Kepler-90i（AJ；NASA 2017-12-14 公布）","en":"Neural network finds Kepler-90i in Kepler data (AJ)"},"url":"https://iopscience.iop.org/article/10.3847/1538-3881/aa9e09","aiImpact":"major","fieldImpact":"major","why":{"zh":"神经网络在 Kepler 数据中找到第 8 颗行星，Kepler-90 成为与太阳系并列行星最多的系统","en":"Neural net finds Kepler-90i, tying the Solar System’s planet count"}},"keyPapers":[{"title":"Identifying Exoplanets with Deep Learning: A Five-planet Resonant Chain around Kepler-80 and an Eighth Planet around Kepler-90","titleZh":"神经网络在 Kepler 数据中发现 Kepler-90i（AJ；NASA 2017-12-14 公布）","id":"10.3847/1538-3881/aa9e09","url":"https://doi.org/10.3847/1538-3881/aa9e09","venue":"The Astronomical Journal","date":"2018-01-30","why":{"zh":"神经网络在 Kepler 数据中找到第 8 颗行星，Kepler-90 成为与太阳系并列行星最多的系统","en":"Neural net finds Kepler-90i, tying the Solar System’s planet count"},"fieldImpact":"major","aiImpact":"major"},{"title":"ExoMiner: A Highly Accurate and Explainable Deep Learning Classifier That Validates 301 New Exoplanets","titleZh":"ExoMiner：深度学习一次验证 301 颗 Kepler 新行星（ApJ）","id":"10.3847/1538-4357/ac4399","url":"https://doi.org/10.3847/1538-4357/ac4399","venue":"The Astrophysical Journal","date":"2022-02-17","why":{"zh":"NASA 用神经网络一次性统计验证 301 颗行星，确立 AI 行星验证流程","en":"NASA neural network statistically validates 301 planets at once"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"The T16 Planet Hunt: 10,000 New Planet Candidates from TESS Cycle 1 and the Confirmation of a Hot Jupiter around TIC 183374187*","titleZh":"T16 行星搜寻：TESS 第 1 周期约 1 万个新行星候选（ApJS）","id":"10.3847/1538-4365/ae5b6c","url":"https://doi.org/10.3847/1538-4365/ae5b6c","venue":"The Astrophysical Journal Supplement Series","date":"2026-04-28","why":{"zh":"约 1 万个新候选，规模远超人工搜寻","en":"~10,000 new candidates"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"The Dawes Review 10: The impact of deep learning for the analysis of galaxy surveys","titleZh":"综述：深度学习对星系巡天分析的影响（PASA Dawes Review）","id":"10.1017/pasa.2022.55","url":"https://doi.org/10.1017/pasa.2022.55","venue":"Publications of the Astronomical Society of Australia","date":"2023-01-10","why":{"zh":"系统梳理深度学习成为巡天标准工具的过程","en":"Maps deep learning becoming a standard survey tool"},"fieldImpact":"incremental","aiImpact":"incremental"},{"title":"New substellar candidates identified through deep learning in the F150 sample of the large-scale SHINE direct imaging survey","titleZh":"NA-SODINN 深度学习重分析 VLT/SPHERE SHINE 巡天 150 颗恒星，找到 13 个此前未报告的亚恒星伴星候选（A&A）","id":"10.1051/0004-6361/202660463","url":"https://doi.org/10.1051/0004-6361/202660463","venue":"Astronomy & Astrophysics","date":"2026-10-15","why":{"zh":"AI：监督式深度学习在角差分成像序列中找出经典方法漏掉的弱信号，并找回全部已知伴星；学科：目前仅 Smethells 20 周围的候选值得优先跟进观测，属候选而非确认","en":"AI: recovers all known companions and finds faint signals missed by classical pipelines. Field: candidates only; one (Smethells 20) remains strong"},"fieldImpact":"incremental","aiImpact":"incremental"}],"latest_papers":[{"id":"arXiv:2610.29654","title":"Probabilistic Machine Learning for Short-Term Atmospheric Seeing Forecasting","titleZh":"用概率机器学习做短时大气视宁度预报","summaryZh":"对比多种概率机器学习模型，提前最多两小时预测天文观测中的大气视宁度，并重点评估不确定性量化。","summaryEn":"Benchmarks probabilistic ML models for forecasting astronomical seeing up to two hours ahead, focusing on uncertainty quantification.","authors":["Mary-Joe Medlej","Rahul Srinivasan","Simon Prunet","Aziz Ziad"],"date":"2026-10-21","url":"https://arxiv.org/abs/2610.29654","site":"arXiv","venue":null,"category":"cs.LG","fetchedAt":"2026-10-22T10:02:13"},{"id":"arXiv:2610.29412","title":"Data-driven characterisation of variable source light curves with multimodal representation learning","titleZh":"用多模态表示学习刻画变源光变曲线","summaryZh":"提出与巡天无关、无需标注的框架，为 ZTF、鲁宾 LSST 等产生的海量变星光变曲线学习表示并做特征刻画，不再局限于预定义分类体系。","summaryEn":"A survey-agnostic, label-free multimodal representation framework to characterise variable-source light curves from ZTF and Rubin LSST.","authors":["M. Madarász","G. Marton","M. Audard"],"date":"2026-10-21","url":"https://arxiv.org/abs/2610.29412","site":"arXiv","venue":null,"category":"astro-ph.IM","fetchedAt":"2026-10-22T10:02:13"},{"id":"arXiv:2610.28740","title":"A Hybrid Neural Network Emulator for M1 Neutrino Reaction Rates in Binary Neutron Star Merger Simulations","titleZh":"双中子星并合模拟中 M1 中微子反应率的混合神经网络替代器","summaryZh":"只用神经网络模拟计算量最大的二维反应过程，其余仍用原库计算，在保持精度的同时加速中微子输运模拟。","summaryEn":"A hybrid neural emulator replaces only the costly 2D neutrino reactions in binary neutron star merger simulations.","authors":["Yi Qiu","David Radice","Raffaele Ferrari","Albino Perego"],"date":"2026-10-20","url":"https://arxiv.org/abs/2610.28740","site":"arXiv","venue":null,"category":"astro-ph.HE","fetchedAt":"2026-10-22T10:02:13"},{"id":"arXiv:2610.28482","title":"The \"Rubin 2036: Opportunities for Discovery\" Workshop: A Community Roadmap for the Future of the Vera C. Rubin Observatory After the LSST","titleZh":"“Rubin 2036”研讨会：LSST 之后鲁宾天文台的社区路线图","summaryZh":"约 100 名研究者讨论鲁宾天文台 2030 年代中期完成 LSST 后的规划，其中一个方向是用鲁宾影像预训练机器学习基础模型，用于源检测、去混叠与分类。","summaryEn":"Community roadmap for Rubin after LSST, including a foundation model pretrained on Rubin imaging for detection, deblending and classification.","authors":["Andrés Alejandro Plazas Malagón","Steve Allen","Shreya Anand","Muhammad Awais"],"date":"2026-10-20","url":"https://arxiv.org/abs/2610.28482","site":"arXiv","venue":null,"category":"astro-ph.IM","fetchedAt":"2026-10-22T10:02:13"},{"id":"arXiv:2610.27252","title":"J-PLUS: Identifying Outlier Spectral Energy Distributions in the Fourth Data Release Using Unsupervised Machine Learning","titleZh":"J-PLUS 第四次数据发布中用无监督机器学习识别异常光谱能量分布","summaryZh":"用自编码器异常检测和 t-SNE，从 12 波段测光数据中筛出偏离主流模式的天体，供光谱跟踪发现稀有或新型天体。","summaryEn":"Autoencoder anomaly detection on 12-band J-PLUS photometry flags outlier SEDs for spectroscopic follow-up.","authors":["F. Quispe-Huaynasi","F. Roig","V. Loaiza-Tacuri","V. M. Placco"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.27252","site":"arXiv","venue":null,"category":"astro-ph.IM","fetchedAt":"2026-10-22T10:02:13"},{"id":"arXiv:2610.27068","title":"ASTROLABE: A 3D Particle-Level Surrogate for Cosmology Simulation with Physical Flow Matching","titleZh":"ASTROLABE：基于物理流匹配的三维粒子级宇宙学模拟替代模型","summaryZh":"直接在 N 个粒子的相空间坐标上做流匹配生成，不依赖体素网格，可高分辨率替代昂贵的暗物质 N 体模拟扫描。","summaryEn":"A flow-matching emulator acting on particle phase-space coordinates, replacing costly dark-matter N-body scans at high resolution.","authors":["Ibrahim Elsharkawy","David Curtin","Yonatan Kahn","Neal Weiner"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.27068","site":"arXiv","venue":null,"category":"astro-ph.CO","fetchedAt":"2026-10-22T10:02:13"},{"id":"arXiv:2610.24882","title":"Single-Transit Parameter Estimation with Calibrated Uncertainties","titleZh":"带校准不确定度的单次凌星参数估计","summaryZh":"针对系外行星巡天中只观测到一次凌星的事件，提出自动化参数估计方法并给出校准的不确定度。","summaryEn":"Automated parameter estimation with calibrated uncertainties for single-transit exoplanet events.","authors":["Helem Salinas","Rafael Brahm","Greg Olmschenk","Melinda Soares-Furtado"],"date":"2026-10-17","url":"https://arxiv.org/abs/2610.24882","site":"arXiv","venue":null,"category":"astro-ph.EP","fetchedAt":"2026-10-22T10:02:13"},{"id":"arXiv:2610.24596","title":"Attitude Estimation of Falcon 9 Rocket Body Based on Photometric Characteristics","titleZh":"基于光度特征的猎鹰 9 号火箭箭体姿态估计","summaryZh":"针对太空碎片监测，提出根据光度变化估计火箭残骸姿态的方法，不依赖特定目标模型。","summaryEn":"Estimates the attitude of a Falcon 9 rocket body from photometric signatures for space-debris monitoring.","authors":["Hui Li","Rong-Wang Li","Ru-Feng Tang","Yu-Qiang Li"],"date":"2026-10-17","url":"https://arxiv.org/abs/2610.24596","site":"arXiv","venue":null,"category":"astro-ph.IM","fetchedAt":"2026-10-22T10:02:13"}]},{"id":"chemistry","slug":"chemistry","category":"formal-physical","categoryName":{"zh":"数理科学","en":"Mathematical & physical sciences"},"name":{"zh":"化学","en":"Chemistry"},"classification":{"oecdFos":"1.4","moe":"0703 化学"},"level":3,"levelKey":"transforming","levelLabel":{"zh":"系统改造","en":"Systemic transformation"},"trend":"up","reason":{"zh":"上调为 L3：2024 年诺贝尔化学奖授予 AI 蛋白质结构预测与设计；逆合成和机器学习势已是常规工具；2026 年自主实验平台相继发现新反应和新配体（Nat. Commun. 2026-06、2026-10），低成本自驱实验室出现。日常实验仍以人工为主。","en":"Raised to L3: the 2024 chemistry Nobel went to AI structure prediction and design; retrosynthesis and ML potentials are routine; in 2026 autonomous platforms discovered new reactions and ligands (Nat. Commun. Jun & Oct 2026) and low-cost self-driving labs appeared. Day-to-day lab work is still largely manual."},"reasonSources":[{"url":"https://www.nobelprize.org/prizes/chemistry/2024/summary/","linkCheck":"ok"},{"url":"https://www.nature.com/articles/s41467-026-78271-9","linkCheck":"ok-crossref"},{"url":"https://www.nature.com/articles/s41467-026-74425-x","linkCheck":"ok-crossref"}],"directions":{"zh":["自驱实验室","反应预测与逆合成","化学大模型智能体"],"en":["Self-driving labs","Reaction prediction & retrosynthesis","Chemistry LLM agents"]},"stats":{"milestones":9,"milestonesSince2025":4,"milestones2026":4,"milestonesLast6Months":3,"news":12,"latestProgress":"2026-10-20","latestMilestone":"2026-10-13","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":0,"leaps":{"ai":1,"field":1},"latestPapers":7,"keyPapers":8},"milestones":[{"date":"2018-03-29","title":{"zh":"深度神经网络 + 符号 AI 规划化学合成（Nature，Segler 等）","en":"Planning chemical syntheses with deep neural networks and symbolic AI (Nature)"},"url":"https://doi.org/10.1038/nature25978","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"双盲测试中化学家无法区分 AI 与文献给出的合成路线，奠定 AI 逆合成","en":"In blind tests chemists could not tell AI routes from literature routes"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"双盲测试中化学家无法区分 AI 与文献给出的合成路线，奠定 AI 逆合成","en":"In blind tests chemists could not tell AI routes from literature routes"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"双盲测试中化学家无法区分 AI 与文献给出的合成路线，奠定 AI 逆合成","en":"In blind tests chemists could not tell AI routes from literature routes"}},"impactNotes":[],"verified":{"id":"10.1038/nature25978","title":"Planning chemical syntheses with deep neural networks and symbolic AI","venue":"Nature","date":"2018-03-29"},"addedIn":"v2"},{"date":"2020-07-08","title":{"zh":"移动机器人化学家 8 天自主完成 688 次实验，找到更好的光催化剂（Nature）","en":"A mobile robotic chemist runs 688 experiments in 8 days and finds better photocatalysts (Nature)"},"url":"https://doi.org/10.1038/s41586-020-2442-2","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"首个在普通实验室里自主移动操作的机器人化学家","en":"First mobile autonomous robot chemist in a standard lab"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首个在普通实验室里自主移动操作的机器人化学家","en":"First mobile autonomous robot chemist in a standard lab"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首个在普通实验室里自主移动操作的机器人化学家","en":"First mobile autonomous robot chemist in a standard lab"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-020-2442-2","title":"A mobile robotic chemist","venue":"Nature","date":"2020-07-08"},"addedIn":"v2"},{"date":"2023-12-20","title":{"zh":"Coscientist：大模型自主设计并执行化学实验（Nature）","en":"Coscientist: LLM autonomously plans and runs chemistry experiments (Nature)"},"url":"https://www.nature.com/articles/s41586-023-06792-0","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"大模型首次自主设计、规划并执行真实化学实验（钯催化偶联）","en":"First LLM system to autonomously plan and run real chemistry experiments"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"大模型首次自主设计、规划并执行真实化学实验（钯催化偶联）","en":"First LLM system to autonomously plan and run real chemistry experiments"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"大模型首次自主设计、规划并执行真实化学实验（钯催化偶联）","en":"First LLM system to autonomously plan and run real chemistry experiments"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-023-06792-0","title":"Autonomous chemical research with large language models","venue":"Nature","date":"2023-12-20"},"addedIn":"v1"},{"date":"2024-05-08","title":{"zh":"ChemCrow：给大模型接入 18 种化学工具，自主规划并合成驱虫剂和催化剂（Nat. Mach. Intell.）","en":"ChemCrow augments an LLM with 18 chemistry tools (Nat. Mach. Intell.)"},"url":"https://doi.org/10.1038/s42256-024-00832-8","source":"Nature Machine Intelligence","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"大模型智能体在真实实验室执行合成","en":"LLM agent executes real syntheses"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"大模型智能体在真实实验室执行合成","en":"LLM agent executes real syntheses"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"大模型智能体在真实实验室执行合成","en":"LLM agent executes real syntheses"}},"impactNotes":[],"verified":{"id":"10.1038/s42256-024-00832-8","title":"Augmenting large language models with chemistry tools","venue":"Nature Machine Intelligence","date":"2024-05-08"},"addedIn":"v2"},{"date":"2024-10-09","title":{"zh":"诺贝尔化学奖授予 Baker、Hassabis、Jumper（蛋白质设计与结构预测）","en":"Chemistry Nobel to Baker, Hassabis & Jumper"},"url":"https://www.nobelprize.org/prizes/chemistry/2024/summary/","source":"Nobel Prize","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"prize","typeLabel":{"zh":"奖项","en":"Prize"},"why":{"zh":"诺贝尔化学奖授予 AlphaFold 与计算蛋白质设计，AI 成果首次获化学最高奖","en":"First chemistry Nobel for AI-driven work"},"aiImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"诺贝尔化学奖授予 AlphaFold 与计算蛋白质设计，AI 成果首次获化学最高奖","en":"First chemistry Nobel for AI-driven work"}},"fieldImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"诺贝尔化学奖授予 AlphaFold 与计算蛋白质设计，AI 成果首次获化学最高奖","en":"First chemistry Nobel for AI-driven work"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-04-13","title":{"zh":"RoboChem-Flex：约 5000 美元的开源自驱化学实验室（Nature Synthesis）","en":"RoboChem-Flex: a ~US$5,000 open self-driving chemistry lab (Nature Synthesis)"},"url":"https://doi.org/10.1038/s44160-026-01053-0","source":"Nature Synthesis","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"system","typeLabel":{"zh":"系统发布","en":"System"},"why":{"zh":"把闭环自主优化的门槛降到普通课题组可负担","en":"Puts closed-loop optimisation within reach of ordinary labs"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"把闭环自主优化的门槛降到普通课题组可负担","en":"Puts closed-loop optimisation within reach of ordinary labs"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"把闭环自主优化的门槛降到普通课题组可负担","en":"Puts closed-loop optimisation within reach of ordinary labs"}},"impactNotes":[],"verified":{"id":"10.1038/s44160-026-01053-0","title":"A flexible and affordable self-driving laboratory for automated reaction optimization","venue":"Nature Synthesis","date":"2026-04-13"},"addedIn":"v2"},{"date":"2026-05-04","title":{"zh":"PoLARIS 微流控自驱实验室优化多元素双钙钛矿纳米片（Nat. Commun.）","en":"PoLARIS self-driving microfluidic lab optimises double-perovskite nanoplatelets (Nat. Commun.)"},"url":"https://www.nature.com/articles/s41467-026-72765-2","source":"Nature Communications","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"微流控自驱实验室优化六元素纳米片并推断反应机理","en":"Microfluidic SDL optimises six-element nanoplatelets"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"微流控自驱实验室优化六元素纳米片并推断反应机理","en":"Microfluidic SDL optimises six-element nanoplatelets"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"微流控自驱实验室优化六元素纳米片并推断反应机理","en":"Microfluidic SDL optimises six-element nanoplatelets"}},"impactNotes":[],"verified":{"id":"10.1038/s41467-026-72765-2","title":"Autonomous microfluidic experimentation for exploring reaction inference and synthesizing double perovskite nanoplatelets","venue":"Nature Communications","date":"2026-05-04"},"addedIn":"v1"},{"date":"2026-06-20","title":{"zh":"Flex-Cat：自主实验室 680 次实验发现条件可编程的“灵活配体”，转化频率提升 2.5 倍以上（Nat. Commun.）","en":"Flex-Cat autonomous lab: 680 experiments, programmable “flexible ligands”, >2.5× turnover (Nat. Commun.)"},"url":"https://doi.org/10.1038/s41467-026-74425-x","source":"Nature Communications","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"自主实验从调条件扩展到发现催化剂新行为，并放大 10 倍验证","en":"Moves from tuning to discovering new catalyst behaviour, validated at 10× scale"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"自主实验从调条件扩展到发现催化剂新行为，并放大 10 倍验证","en":"Moves from tuning to discovering new catalyst behaviour, validated at 10× scale"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"自主实验从调条件扩展到发现催化剂新行为，并放大 10 倍验证","en":"Moves from tuning to discovering new catalyst behaviour, validated at 10× scale"}},"impactNotes":[],"verified":{"id":"10.1038/s41467-026-74425-x","title":"An autonomous lab for data-driven homogeneous catalysis","venue":"Nature Communications","date":"2026-06-20"},"addedIn":"v2"},{"date":"2026-10-13","title":{"zh":"强化学习机器人 30 天 3000 多次决策，发现 4 个高对映选择性新反应（Nat. Commun.）","en":"RL-guided robot makes 3,000+ decisions in 30 days and discovers four highly enantioselective reactions (Nat. Commun.)"},"url":"https://doi.org/10.1038/s41467-026-78271-9","source":"Nature Communications","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"首次让自主平台在手性催化中发现新反应，而非只优化已知反应","en":"Autonomous discovery of new asymmetric reactions, not just optimisation"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首次让自主平台在手性催化中发现新反应，而非只优化已知反应","en":"Autonomous discovery of new asymmetric reactions, not just optimisation"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首次让自主平台在手性催化中发现新反应，而非只优化已知反应","en":"Autonomous discovery of new asymmetric reactions, not just optimisation"}},"impactNotes":[],"verified":{"id":"10.1038/s41467-026-78271-9","title":"Reinforcement learning guided robotic platform that investigates and discovers asymmetric reactions","venue":"Nature Communications","date":"2026-10-13"},"addedIn":"v2"}],"papersRepos":[{"title":"ChemCrow: Augmenting LLMs with chemistry tools (Nat. Mach. Intell. 2024)","url":"https://www.nature.com/articles/s42256-024-00832-8","date":"2024-05","kind":"paper","linkCheck":"ok"},{"title":"Planning chemical syntheses with deep neural networks and symbolic AI","url":"https://doi.org/10.1038/nature25978","date":"2018-03-29","kind":"paper","linkCheck":"ok-crossref"},{"title":"A mobile robotic chemist","url":"https://doi.org/10.1038/s41586-020-2442-2","date":"2020-07-08","kind":"paper","linkCheck":"ok-crossref"},{"title":"Autonomous chemical research with large language models","url":"https://doi.org/10.1038/s41586-023-06792-0","date":"2023-12-20","kind":"paper","linkCheck":"ok-crossref"},{"title":"Augmenting large language models with chemistry tools","url":"https://doi.org/10.1038/s42256-024-00832-8","date":"2024-05-08","kind":"paper","linkCheck":"ok-crossref"},{"title":"A flexible and affordable self-driving laboratory for automated reaction optimization","url":"https://doi.org/10.1038/s44160-026-01053-0","date":"2026-04-13","kind":"paper","linkCheck":"ok-crossref"},{"title":"An autonomous lab for data-driven homogeneous catalysis","url":"https://doi.org/10.1038/s41467-026-74425-x","date":"2026-06-20","kind":"paper","linkCheck":"ok-crossref"},{"title":"Reinforcement learning guided robotic platform that investigates and discovers asymmetric reactions","url":"https://doi.org/10.1038/s41467-026-78271-9","date":"2026-10-13","kind":"paper","linkCheck":"ok-crossref"},{"title":"Autonomous microfluidic experimentation for exploring reaction inference and synthesizing double perovskite nanoplatelets","url":"https://doi.org/10.1038/s41467-026-72765-2","date":"2026-05-04","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"Carnegie Mellon University","url":"https://www.cmu.edu"},{"name":"IBM RXN","url":"https://rxn.app.accelerate.science"}],"news":[{"id":"82497db5b87a","date":"2026-10-20","title":"Otter.ai raises $320 million from General Catalyst’s growth fund for its meeting tool","url":"https://techcrunch.com/2026/10/20/otter-ai-raises-320-million-from-general-catalysts-growth-fund-for-its-meeting-tool/","source":"TechCrunch AI","lang":"en","kind":"news","summaryZh":"TechCrunch AI于2026-10-20发布的文章，涉及AI 化学材料、AI 商业融资，原题《Otter.ai raises $320 million from General Catalyst’s growth fund for its meeting tool》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"d08b5efe1a58","date":"2026-10-20","title":"Aptamers as a sandbox for lab-in-the-loop agentic molecular discovery","url":"https://arxiv.org/abs/2610.24806","source":"arXiv q-bio.BM（生物分子）","lang":"en","kind":"paper","summaryZh":"arXiv q-bio.BM（生物分子）于2026-10-20发布的论文，涉及AI 生物/医药，原题《Aptamers as a sandbox for lab-in-the-loop agentic molecular discovery》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"78b3f2f5855d","date":"2026-10-20","title":"Distillation, Self-Training, and the Fragility of Learned Chemical Reasoning","url":"https://arxiv.org/abs/2610.24199","source":"arXiv cs.AI","lang":"en","kind":"paper","summaryZh":"arXiv cs.AI于2026-10-20发布的论文，涉及AI 科研与调研、AI 教育、AI 化学材料，原题《Distillation, Self-Training, and the Fragility of Learned Chemical Reasoning》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"12d10c346fbc","date":"2026-10-20","title":"Measurement-efficient energy estimator for quantum chemistry of photodynamic materials","url":"https://arxiv.org/abs/2610.26689","source":"arXiv physics.chem-ph","lang":"en","kind":"paper","summaryZh":"arXiv physics.chem-ph于2026-10-20发布的论文，涉及AI 化学材料、AI 气候能源、AI 软件开发，原题《Measurement-efficient energy estimator for quantum chemistry of photodynamic materials》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"4f2bfcfd9737","date":"2026-10-19","title":"'24/7 unmanned synthesis platform' uses robots and light to produce drug candidates","url":"https://phys.org/news/2026-10-unmanned-synthesis-platform-robots-drug.html","source":"Phys.org 化学","lang":"en","kind":"news","summaryZh":"Phys.org 化学于2026-10-20发布的文章，涉及AI 化学材料、AI 医疗、AI 机器人，原题《'24/7 unmanned synthesis platform' uses robots and light to produce drug candidates》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"095e9f3605da","date":"2026-10-14","title":"CAS and Novartis collaborate to advance access to reaction data and AI research","url":"https://cen.acs.org/acs-news/cas-novartis-collaborate-advance-access/104/web/2026/10?sc=230901_cenrssfeed_eng_latestnewsrss_cen","source":"C&EN 最新","lang":"en","kind":"news","summaryZh":"C&EN 最新于2026-10-14发布的文章，涉及AI 化学材料，原题《CAS and Novartis collaborate to advance access to reaction data and AI research》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"e337f1ea1716","date":"2026-10-12","title":"Copernic deploys AI to discover better catalysts for making chemicals","url":"https://cen.acs.org/business/agriculture/copernic-catalysts-start-ups-agriculture-fertilizer/104/web/2026/10?sc=230901_cenrssfeed_eng_latestnewsrss_cen","source":"C&EN 最新","lang":"en","kind":"news","summaryZh":"C&EN 最新于2026-10-12发布的文章，涉及AI 化学材料，原题《Copernic deploys AI to discover better catalysts for making chemicals》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"d07f1777148a","date":"2026-10-09","title":"Advancing chemical reaction prediction in data-scarce drug discovery with active and geometric deep learning","url":"https://www.nature.com/articles/s43588-026-01063-0","source":"Nature 学科·计算化学","lang":"en","kind":"news","summaryZh":"Nature 学科·计算化学于2026-10-09发布的文章，涉及AI 化学材料、AI 生物/医药，原题《Advancing chemical reaction prediction in data-scarce drug discovery with active and geome》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"89ec0a42fa1b","date":"2026-10-06","title":"Inductive Bio launches AI tools for day-to-day medicinal chemistry","url":"https://cen.acs.org/business/informatics/inductive-bio-medicinal-chemistry-ai-tools/104/web/2026/10?sc=230901_cenrssfeed_eng_latestnewsrss_cen","source":"C&EN","lang":"en","kind":"news","summaryZh":"Inductive Bio 推出面向药物化学家日常工作的 AI 工具，从“生成分子”延伸到实验室实际决策。","via":"research-samples-20261010","linkCheck":"from-feed"},{"id":"2f5a5f8b9182","date":"2026-10-05","title":"Designing enzymes for new-to-nature chemistry and non-natural substrates with AlphaProtein Novo","url":"https://www.biorxiv.org/content/10.64898/2026.10.01.756017v1","source":"bioRxiv (synthetic biology)","lang":"en","kind":"news","summaryZh":"bioRxiv：AlphaProtein Novo 从头设计能催化非天然反应、作用于非天然底物的酶。","via":"research-samples-20261010","linkCheck":"from-feed"},{"id":"72020960144b","date":"2026-10-05","title":"Editorial: What can chemists learn from how AI is shaking up mathematics?","url":"https://cen.acs.org/research-integrity/editorial-artificial-intelligence-chemists/104/web/2026/10?sc=230901_cenrssfeed_eng_latestnewsrss_cen","source":"C&EN 最新","lang":"en","kind":"news","summaryZh":"C&EN 最新于2026-10-05发布的文章，涉及AI 化学材料、AI 数学、AI 创意媒体，原题《Editorial: What can chemists learn from how AI is shaking up mathematics?》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"f68564af95f9","date":"2026-04-27","title":"A chemistry lab that runs itself to find the perfect reaction","url":"https://www.nature.com/articles/d41586-026-01283-4","source":"Nature","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"ok"}],"review":{"lastReviewed":"2026-10-21","cycleDays":60,"nextReview":"2026-12-20"},"urls":{"zh":"/zh-hans/disciplines/chemistry/","en":"/en/disciplines/chemistry/"},"dimensions":{"method":{"score":3,"label":{"zh":"研究方法","en":"Research method"},"evidence":"逆合成规划、机器学习势、贝叶斯优化自驱实验室已成常规方法","source":"https://doi.org/10.1038/nature25978","weight":0.3},"discovery":{"score":3,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"2024 诺贝尔化学奖授予 AI 蛋白质结构预测与设计；自主平台 2026 年发现新的高对映选择性反应、新催化配体","source":"https://www.nobelprize.org/prizes/chemistry/2024/summary/","weight":0.35},"adoption":{"score":2,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"自驱实验室成本降至约 5000 美元（RoboChem-Flex），但普及率无调查数据","source":"https://www.nature.com/articles/s44160-026-01053-0","weight":0.25},"education":{"score":1,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到课程层面证据","source":null,"weight":0.1}},"compositeScore":2.55,"levelCapped":false,"levelChange":{"from":2,"to":3,"direction":"up","reason":{"zh":"上调为 L3：2024 年诺贝尔化学奖授予 AI 蛋白质结构预测与设计；逆合成和机器学习势已是常规工具；2026 年自主实验平台相继发现新反应和新配体（Nat. Commun. 2026-06、2026-10），低成本自驱实验室出现。日常实验仍以人工为主。","en":"Raised to L3: the 2024 chemistry Nobel went to AI structure prediction and design; retrosynthesis and ML potentials are routine; in 2026 autonomous platforms discovered new reactions and ligands (Nat. Commun. Jun & Oct 2026) and low-cost self-driving labs appeared. Day-to-day lab work is still largely manual."}},"biggestBreakthrough":{"date":"2024-10-09","title":{"zh":"诺贝尔化学奖授予 Baker、Hassabis、Jumper（蛋白质设计与结构预测）","en":"Chemistry Nobel to Baker, Hassabis & Jumper"},"url":"https://www.nobelprize.org/prizes/chemistry/2024/summary/","aiImpact":"leap","fieldImpact":"leap","why":{"zh":"诺贝尔化学奖授予 AlphaFold 与计算蛋白质设计，AI 成果首次获化学最高奖","en":"First chemistry Nobel for AI-driven work"}},"keyPapers":[{"title":"Planning chemical syntheses with deep neural networks and symbolic AI","titleZh":"深度神经网络 + 符号 AI 规划化学合成（Nature，Segler 等）","id":"10.1038/nature25978","url":"https://doi.org/10.1038/nature25978","venue":"Nature","date":"2018-03-29","why":{"zh":"双盲测试中化学家无法区分 AI 与文献给出的合成路线，奠定 AI 逆合成","en":"In blind tests chemists could not tell AI routes from literature routes"},"fieldImpact":"major","aiImpact":"major"},{"title":"A mobile robotic chemist","titleZh":"移动机器人化学家 8 天自主完成 688 次实验，找到更好的光催化剂（Nature）","id":"10.1038/s41586-020-2442-2","url":"https://doi.org/10.1038/s41586-020-2442-2","venue":"Nature","date":"2020-07-08","why":{"zh":"首个在普通实验室里自主移动操作的机器人化学家","en":"First mobile autonomous robot chemist in a standard lab"},"fieldImpact":"major","aiImpact":"major"},{"title":"Autonomous chemical research with large language models","titleZh":"Coscientist：大模型自主设计并执行化学实验（Nature）","id":"10.1038/s41586-023-06792-0","url":"https://doi.org/10.1038/s41586-023-06792-0","venue":"Nature","date":"2023-12-20","why":{"zh":"大模型首次自主设计、规划并执行真实化学实验（钯催化偶联）","en":"First LLM system to autonomously plan and run real chemistry experiments"},"fieldImpact":"major","aiImpact":"major"},{"title":"Augmenting large language models with chemistry tools","titleZh":"ChemCrow：给大模型接入 18 种化学工具，自主规划并合成驱虫剂和催化剂（Nat. Mach. Intell.）","id":"10.1038/s42256-024-00832-8","url":"https://doi.org/10.1038/s42256-024-00832-8","venue":"Nature Machine Intelligence","date":"2024-05-08","why":{"zh":"大模型智能体在真实实验室执行合成","en":"LLM agent executes real syntheses"},"fieldImpact":"incremental","aiImpact":"major"},{"title":"A flexible and affordable self-driving laboratory for automated reaction optimization","titleZh":"RoboChem-Flex：约 5000 美元的开源自驱化学实验室（Nature Synthesis）","id":"10.1038/s44160-026-01053-0","url":"https://doi.org/10.1038/s44160-026-01053-0","venue":"Nature Synthesis","date":"2026-04-13","why":{"zh":"把闭环自主优化的门槛降到普通课题组可负担","en":"Puts closed-loop optimisation within reach of ordinary labs"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"An autonomous lab for data-driven homogeneous catalysis","titleZh":"Flex-Cat：自主实验室 680 次实验发现条件可编程的“灵活配体”，转化频率提升 2.5 倍以上（Nat. Commun.）","id":"10.1038/s41467-026-74425-x","url":"https://doi.org/10.1038/s41467-026-74425-x","venue":"Nature Communications","date":"2026-06-20","why":{"zh":"自主实验从调条件扩展到发现催化剂新行为，并放大 10 倍验证","en":"Moves from tuning to discovering new catalyst behaviour, validated at 10× scale"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Reinforcement learning guided robotic platform that investigates and discovers asymmetric reactions","titleZh":"强化学习机器人 30 天 3000 多次决策，发现 4 个高对映选择性新反应（Nat. Commun.）","id":"10.1038/s41467-026-78271-9","url":"https://doi.org/10.1038/s41467-026-78271-9","venue":"Nature Communications","date":"2026-10-13","why":{"zh":"首次让自主平台在手性催化中发现新反应，而非只优化已知反应","en":"Autonomous discovery of new asymmetric reactions, not just optimisation"},"fieldImpact":"major","aiImpact":"major"},{"title":"Autonomous microfluidic experimentation for exploring reaction inference and synthesizing double perovskite nanoplatelets","titleZh":"PoLARIS 微流控自驱实验室优化多元素双钙钛矿纳米片（Nat. Commun.）","id":"10.1038/s41467-026-72765-2","url":"https://doi.org/10.1038/s41467-026-72765-2","venue":"Nature Communications","date":"2026-05-04","why":{"zh":"微流控自驱实验室优化六元素纳米片并推断反应机理","en":"Microfluidic SDL optimises six-element nanoplatelets"},"fieldImpact":"incremental","aiImpact":"incremental"}],"latest_papers":[{"id":"doi:10.5445/ir/1000198180","title":"Breaking Bottlenecks in Solid Electrolyte Discovery With Large Artificial Intelligence Models","titleZh":"用大型 AI 模型突破固态电解质发现瓶颈","summaryZh":"提出以大型 AI 模型驱动的自主固态电解质发现框架，统筹离子输运、缺陷化学、力学和界面稳定性等多目标优化。","summaryEn":"Outlines a framework for autonomous solid-electrolyte discovery driven by large AI models across transport, defect, mechanical and interface objectives.","authors":["Eric Jianfeng Cheng","Min Hong","Zhiquan Zeng","Chuanyu Liu"],"date":"2026-10-20","url":"https://doi.org/10.5445/ir/1000198180","site":"OpenAlex","venue":"KITopen","category":null,"fetchedAt":"2026-10-22T10:02:18"},{"id":"arXiv:2610.27230","title":"Magnolia: A Persistent-Memory Agent Framework for Computational Chemistry","titleZh":"Magnolia：面向计算化学的持久记忆智能体框架","summaryZh":"整合对接到量子化学的计算流程、实验记录与对话日志，用三层记忆把会话观察沉淀为经人工确认的跨项目规则，并在四个案例中演示。","summaryEn":"A persistent-memory LLM agent for computational chemistry that distils observations into human-validated cross-project rules.","authors":["Tao Jiang","Benjamin Navet","Yousouf Taghzouti","Fabien Gandon"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.27230","site":"arXiv","venue":null,"category":"cs.AI","fetchedAt":"2026-10-22T10:02:18"},{"id":"arXiv:2610.26701","title":"Challenging Atomistic Foundation Models on Disordered Carbons","titleZh":"用无序碳挑战原子尺度基础模型","summaryZh":"用纯碳及含氧/氢的无序碳体系压力测试 6 个流行的基础机器学习原子间势，检验其超出训练域的可靠性。","summaryEn":"Stress-tests six popular foundation ML interatomic potentials on disordered carbons to assess out-of-domain reliability.","authors":["Joseph Hart","Samuel G. H. Brookes","Fabian L. Thiemann","Christoph Schran"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.26701","site":"arXiv","venue":null,"category":"physics.chem-ph","fetchedAt":"2026-10-22T10:02:18"},{"id":"doi:10.25258/ijddt.16.75s.134","title":"Global Research Trends in Artificial Intelligence and Machine Learning for Drug Delivery Technology: A Bibliometric and Science Mapping Analysis (2015–2026)","titleZh":"药物递送技术中人工智能与机器学习的全球研究趋势：文献计量分析（2015–2026）","summaryZh":"用文献计量方法梳理 2015–2026 年 AI/ML 在药物释放动力学预测、制剂优化和新载体发现等方面的研究趋势。","summaryEn":"Bibliometric mapping of AI/ML research in drug delivery (2015–2026).","authors":[],"date":"2026-10-18","url":"https://doi.org/10.25258/ijddt.16.75s.134","site":"OpenAlex","venue":"International Journal of Drug Delivery Technology","category":null,"fetchedAt":"2026-10-22T10:02:18"},{"id":"arXiv:2610.24806","title":"Aptamers as a sandbox for lab-in-the-loop agentic molecular discovery","titleZh":"以适配体为沙盒：实验室闭环的智能体分子发现","summaryZh":"用蛋白结合适配体设计（每轮 1–2 周、每个候选 1–3 美元）检验大模型推理智能体能否借助湿实验数据迭代加速从头分子发现。","summaryEn":"Uses aptamer design (1–2 week cycles, $1–3 per candidate) to test whether LLM agents iterating with wet-lab data speed up de novo discovery.","authors":["Zachary Cutts","John Coller","Rachel Chen","Hua Tang"],"date":"2026-10-17","url":"https://arxiv.org/abs/2610.24806","site":"arXiv","venue":null,"category":"q-bio.BM","fetchedAt":"2026-10-22T10:02:18"},{"id":"doi:10.1021/acsnano.6c04548","title":"Experimental Design Signatures Emerge from Topic Modeling of Graphene Oxide Literature","titleZh":"氧化石墨烯文献主题建模揭示实验设计特征","summaryZh":"对数十万篇氧化石墨烯文献做主题建模，构建结构化知识图谱，提取实验设计规律。","summaryEn":"Topic modelling of the vast graphene-oxide literature reveals experimental design signatures.","authors":["Henry Julio Kobs","Henrique Ferreira","Yang Liu","Amauri Jardim de Paula"],"date":"2026-10-17","url":"https://doi.org/10.1021/acsnano.6c04548","site":"OpenAlex","venue":"ACS Nano","category":null,"fetchedAt":"2026-10-22T10:02:18"},{"id":"doi:10.1038/s41598-026-75562-5","title":"SpecLLM: LLM-based VNIR spectral interpretation","titleZh":"SpecLLM：基于大模型的可见-近红外光谱解读","summaryZh":"把光谱数据转为大模型可理解的特征级描述（峰位、强度等），让大模型像光谱学家一样解读可见-近红外光谱。","summaryEn":"Makes VNIR spectra accessible to LLMs through feature-level descriptions for interpretation.","authors":["Michał Romaszewski","Anna Strzoda","Marzena Halama","Przemysław Głomb"],"date":"2026-10-17","url":"https://doi.org/10.1038/s41598-026-75562-5","site":"OpenAlex","venue":"Scientific Reports","category":null,"fetchedAt":"2026-10-22T10:02:18"}]},{"id":"materials","slug":"materials","category":"formal-physical","categoryName":{"zh":"数理科学","en":"Mathematical & physical sciences"},"name":{"zh":"材料科学","en":"Materials science"},"classification":{"oecdFos":"2.5","moe":"0805 材料科学与工程"},"level":2,"levelKey":"breakthrough","levelLabel":{"zh":"局部突破","en":"Local breakthroughs"},"trend":"flat","reason":{"zh":"GNoME 预测 220 万种晶体、MatterGen 生成式设计新材料（Nature 2023/2025），A-Lab 自主合成；但 A-Lab 论文 2026 年 1 月被更正（成功数由 41/58 降至 36/57），“AI 发现新材料”的成色仍有争议。","en":"GNoME predicted 2.2M crystals and MatterGen designs materials generatively (Nature 2023/2025), A-Lab synthesises autonomously; but the A-Lab paper was corrected in Jan 2026 (41/58 → 36/57), so “AI-discovered materials” claims remain contested."},"reasonSources":[{"url":"https://www.nature.com/articles/s41586-023-06735-9","linkCheck":"ok"},{"url":"https://cen.acs.org/research-integrity/Nature-robot-chemist-paper-corrected/104/web/2026/01","linkCheck":"ok"}],"directions":{"zh":["晶体结构预测","生成式材料设计","自主合成与表征","机器学习势"],"en":["Crystal structure prediction","Generative materials design","Autonomous synthesis","ML interatomic potentials"]},"stats":{"milestones":10,"milestonesSince2025":7,"milestones2026":6,"milestonesLast6Months":2,"news":8,"latestProgress":"2026-10-20","latestMilestone":"2026-10-02","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":0,"leaps":{"ai":1,"field":0},"latestPapers":8,"keyPapers":9},"milestones":[{"date":"2023-11-29","title":{"zh":"GNoME 预测 220 万种新晶体，约 38 万种稳定（Nature）","en":"GNoME predicts 2.2M new crystals, ~380k stable (Nature)"},"url":"https://www.nature.com/articles/s41586-023-06735-9","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"预测 220 万种晶体、约 38 万种稳定，稳定材料已知数量扩大约一个数量级；新颖性后遭部分学者质疑","en":"2.2M crystals, ~380k stable; novelty later questioned"},"aiImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"预测 220 万种晶体、约 38 万种稳定，稳定材料已知数量扩大约一个数量级；新颖性后遭部分学者质疑","en":"2.2M crystals, ~380k stable; novelty later questioned"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"预测 220 万种晶体、约 38 万种稳定，稳定材料已知数量扩大约一个数量级；新颖性后遭部分学者质疑","en":"2.2M crystals, ~380k stable; novelty later questioned"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-023-06735-9","title":"Scaling deep learning for materials discovery","venue":"Nature","date":"2023-11-29"},"addedIn":"v1"},{"date":"2023-11-29","title":{"zh":"A-Lab 自主实验室 17 天合成新化合物（Nature；2026 年更正）","en":"A-Lab synthesises new compounds in 17 days (Nature; corrected 2026)"},"url":"https://www.nature.com/articles/s41586-023-06734-w","source":"Nature","flags":["contested"],"flagLabels":[{"zh":"有争议","en":"Contested"}],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"自主实验室 17 天合成新化合物；2026-01 更正为 36/57，结论有争议","en":"Autonomous synthesis; corrected to 36/57 in Jan 2026"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"自主实验室 17 天合成新化合物；2026-01 更正为 36/57，结论有争议","en":"Autonomous synthesis; corrected to 36/57 in Jan 2026"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"自主实验室 17 天合成新化合物；2026-01 更正为 36/57，结论有争议","en":"Autonomous synthesis; corrected to 36/57 in Jan 2026"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-023-06734-w","title":"An autonomous laboratory for the accelerated synthesis of inorganic materials","venue":"Nature","date":"2023-11-29"},"addedIn":"v1"},{"date":"2023-12-29","title":{"zh":"MACE-MP-0：覆盖周期表的通用机器学习原子间势基础模型","en":"MACE-MP-0: a foundation ML interatomic potential across the periodic table"},"url":"https://arxiv.org/abs/2401.00096","source":"arXiv","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok","datePrecision":"day","type":"system","typeLabel":{"zh":"系统发布","en":"System"},"why":{"zh":"一个模型即可模拟大部分无机材料体系，大幅降低原子模拟门槛","en":"One model simulates most inorganic systems"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"一个模型即可模拟大部分无机材料体系，大幅降低原子模拟门槛","en":"One model simulates most inorganic systems"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"一个模型即可模拟大部分无机材料体系，大幅降低原子模拟门槛","en":"One model simulates most inorganic systems"}},"impactNotes":[],"verified":{"id":"arXiv:2401.00096","title":"A foundation model for atomistic materials chemistry","venue":"arXiv","date":"2023-12-29"},"addedIn":"v2"},{"date":"2025-01-16","title":{"zh":"MatterGen 生成式设计无机材料（Nature）","en":"MatterGen generative design of inorganic materials (Nature)"},"url":"https://www.nature.com/articles/s41586-025-08628-5","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"生成式模型按目标性质设计新材料并合成验证一例","en":"Generative design with one synthesised validation"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"生成式模型按目标性质设计新材料并合成验证一例","en":"Generative design with one synthesised validation"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"生成式模型按目标性质设计新材料并合成验证一例","en":"Generative design with one synthesised validation"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-025-08628-5","title":"A generative model for inorganic materials design","venue":"Nature","date":"2025-01-16"},"addedIn":"v1"},{"date":"2026-01-27","title":{"zh":"BEE-NET：从 130 万结构中筛选，2 个此前未知的化合物经实验证实超导（npj Comput. Mater.）","en":"BEE-NET screens 1.3M structures; two previously unknown superconductors confirmed experimentally"},"url":"https://doi.org/10.1038/s41524-026-01964-8","source":"npj Computational Materials","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"AI 预测新超导体获实验证实","en":"AI-predicted superconductors confirmed in the lab"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI 预测新超导体获实验证实","en":"AI-predicted superconductors confirmed in the lab"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI 预测新超导体获实验证实","en":"AI-predicted superconductors confirmed in the lab"}},"impactNotes":[],"verified":{"id":"10.1038/s41524-026-01964-8","title":"Developing a complete AI-accelerated workflow for superconductor discovery","venue":"npj Computational Materials","date":"2026-01-27"},"addedIn":"v2"},{"date":"2026-01-29","title":{"zh":"A-Lab 论文更正：成功合成数改为 36/57","en":"A-Lab paper corrected: successes revised to 36 of 57"},"url":"https://cen.acs.org/research-integrity/Nature-robot-chemist-paper-corrected/104/web/2026/01","source":"C&EN","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"更正说明“AI 发现材料”需要更严格的表征和验证","en":"Correction shows need for stricter validation"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"更正说明“AI 发现材料”需要更严格的表征和验证","en":"Correction shows need for stricter validation"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"更正说明“AI 发现材料”需要更严格的表征和验证","en":"Correction shows need for stricter validation"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-02-02","title":{"zh":"DiffSyn：扩散模型规划合成路线，据此合成出热稳定性更好的新沸石（Nat. Comput. Sci.）","en":"DiffSyn plans synthesis routes; a new zeolite with better thermal stability is made"},"url":"https://doi.org/10.1038/s43588-025-00949-9","source":"Nature Computational Science","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"AI 从“预测结构”推进到“告诉你怎么合成”","en":"From predicting structures to planning how to make them"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI 从“预测结构”推进到“告诉你怎么合成”","en":"From predicting structures to planning how to make them"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI 从“预测结构”推进到“告诉你怎么合成”","en":"From predicting structures to planning how to make them"}},"impactNotes":[],"verified":{"id":"10.1038/s43588-025-00949-9","title":"DiffSyn: a generative diffusion approach to materials synthesis planning","venue":"Nature Computational Science","date":"2026-02-02"},"addedIn":"v2"},{"date":"2026-04-14","title":{"zh":"机器学习设计增材制造专用低模量钛合金，经激光粉床熔融验证（Nat. Commun.）","en":"ML-designed AM-specific low-modulus Ti alloy validated by laser powder-bed fusion"},"url":"https://doi.org/10.1038/s41467-026-71927-6","source":"Nature Communications","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"AI 设计合金进入可制造的植入物工程验证","en":"AI-designed alloy validated for manufacturable implants"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI 设计合金进入可制造的植入物工程验证","en":"AI-designed alloy validated for manufacturable implants"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI 设计合金进入可制造的植入物工程验证","en":"AI-designed alloy validated for manufacturable implants"}},"impactNotes":[],"verified":{"id":"10.1038/s41467-026-71927-6","title":"Machine learning driven discovery of low modulus biomedical titanium alloys for additive manufacturing","venue":"Nature Communications","date":"2026-04-14"},"addedIn":"v2"},{"date":"2026-08-17","title":{"zh":"贝叶斯“共导航”：AFM 实验与动力学蒙特卡洛模型自主协同（ACS Nano）","en":"Bayesian conavigation couples AFM experiments with a kMC model (ACS Nano)"},"url":"https://doi.org/10.1021/acsnano.6c03268","source":"ACS Nano","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"实验与模拟自主协同，属方法改进","en":"Method improvement"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"实验与模拟自主协同，属方法改进","en":"Method improvement"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"实验与模拟自主协同，属方法改进","en":"Method improvement"}},"impactNotes":[],"verified":{"id":"10.1021/acsnano.6c03268","title":"Bayesian\nConavigation of a Computational Physical\nModel and Atomic Force Microscopy Experiment to Autonomously Survey\na Combinatorial Materials Library","venue":"ACS Nano","date":"2026-08-17"},"addedIn":"v1"},{"date":"2026-10-02","title":{"zh":"AI 加速钙钛矿纳米晶规模化合成：成本降至 2.16 美元/克，太阳能电池效率 19.37%（Nature Communications）","en":"AI-accelerated scalable synthesis of nanocrystals for cost-effective photovoltaics (Nat. Commun.)"},"url":"https://www.nature.com/articles/s41467-026-78268-4","source":"Nature Communications","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"AI：机器学习与自动化流动合成闭环优化工艺；学科：把实验室材料推向低成本规模化制备","en":"AI: ML plus automated flow synthesis optimise the process; Field: pushes lab materials toward low-cost scale-up"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：机器学习与自动化流动合成闭环优化工艺；学科：把实验室材料推向低成本规模化制备","en":"AI: ML plus automated flow synthesis optimise the process; Field: pushes lab materials toward low-cost scale-up"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：机器学习与自动化流动合成闭环优化工艺；学科：把实验室材料推向低成本规模化制备","en":"AI: ML plus automated flow synthesis optimise the process; Field: pushes lab materials toward low-cost scale-up"}},"impactNotes":[],"verified":{"id":"10.1038/s41467-026-78268-4","title":"AI-accelerated scalable synthesis of nanocrystals for cost-effective photovoltaics","venue":"Nature Communications","date":"2026-10-02"},"addedIn":"v2"}],"papersRepos":[{"title":"google-deepmind/materials_discovery (GNoME)","url":"https://github.com/google-deepmind/materials_discovery","date":null,"kind":"repo","linkCheck":"ok"},{"title":"microsoft/mattergen","url":"https://github.com/microsoft/mattergen","date":null,"kind":"repo","linkCheck":"ok"},{"title":"Scaling deep learning for materials discovery","url":"https://doi.org/10.1038/s41586-023-06735-9","date":"2023-11-29","kind":"paper","linkCheck":"ok-crossref"},{"title":"An autonomous laboratory for the accelerated synthesis of inorganic materials","url":"https://doi.org/10.1038/s41586-023-06734-w","date":"2023-11-29","kind":"paper","linkCheck":"ok-crossref"},{"title":"A foundation model for atomistic materials chemistry","url":"https://arxiv.org/abs/2401.00096","date":"2023-12-29","kind":"paper","linkCheck":"ok-crossref"},{"title":"A generative model for inorganic materials design","url":"https://doi.org/10.1038/s41586-025-08628-5","date":"2025-01-16","kind":"paper","linkCheck":"ok-crossref"},{"title":"Developing a complete AI-accelerated workflow for superconductor discovery","url":"https://doi.org/10.1038/s41524-026-01964-8","date":"2026-01-27","kind":"paper","linkCheck":"ok-crossref"},{"title":"DiffSyn: a generative diffusion approach to materials synthesis planning","url":"https://doi.org/10.1038/s43588-025-00949-9","date":"2026-02-02","kind":"paper","linkCheck":"ok-crossref"},{"title":"Machine learning driven discovery of low modulus biomedical titanium alloys for additive manufacturing","url":"https://doi.org/10.1038/s41467-026-71927-6","date":"2026-04-14","kind":"paper","linkCheck":"ok-crossref"},{"title":"AI-accelerated scalable synthesis of nanocrystals for cost-effective photovoltaics","url":"https://doi.org/10.1038/s41467-026-78268-4","date":"2026-10-02","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"Google DeepMind","url":"https://deepmind.google"},{"name":"Microsoft Research AI for Science","url":"https://www.microsoft.com/en-us/research/lab/microsoft-research-ai-for-science/"},{"name":"Lawrence Berkeley National Lab","url":"https://www.lbl.gov"},{"name":"Materials Project","url":"https://next-gen.materialsproject.org"}],"news":[{"id":"49d6da1288e4","date":"2026-10-20","title":"硬氪首发｜用AI智能体辅助新材料研发，睿质科技完成数千万元天使轮融资","url":"https://36kr.com/p/4033523733042053?f=rss","source":"36氪","lang":"zh","kind":"news","summaryZh":"36氪于2026-10-20发布的文章，涉及AI 化学材料、AI 商业融资、AI 科研与调研，原题《硬氪首发｜用AI智能体辅助新材料研发，睿质科技完成数千万元天使轮融资》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"1ef652e1f212","date":"2026-10-20","title":"Multi-property Feynman-Kac Steering of Crystal Diffusion Models","url":"https://arxiv.org/abs/2610.24828","source":"arXiv cond-mat.mtrl-sci","lang":"en","kind":"paper","summaryZh":"arXiv cond-mat.mtrl-sci于2026-10-20发布的论文，涉及AI 化学材料，原题《Multi-property Feynman-Kac Steering of Crystal Diffusion Models》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"2d006a416f43","date":"2026-10-20","title":"Crystal Structure Based Graph Neural Network And LLM Multi-Modal Approach for Material Magnetic Ordering Prediction","url":"https://arxiv.org/abs/2610.25775","source":"arXiv cond-mat.mtrl-sci","lang":"en","kind":"paper","summaryZh":"arXiv cond-mat.mtrl-sci于2026-10-20发布的论文，涉及AI 化学材料，原题《Crystal Structure Based Graph Neural Network And LLM Multi-Modal Approach for Material Mag》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"6f154627796a","date":"2026-10-20","title":"Data-Efficient Indentation Size Effect Mitigation in Steels Using Machine Learning and a Physics-Constrained Neural Network","url":"https://arxiv.org/abs/2604.27775","source":"arXiv cond-mat.mtrl-sci","lang":"en","kind":"paper","summaryZh":"arXiv cond-mat.mtrl-sci于2026-10-20发布的论文，涉及AI 化学材料，原题《Data-Efficient Indentation Size Effect Mitigation in Steels Using Machine Learning and a P》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"f9fab6451ee7","date":"2026-10-19","title":"Giving AI a 'crystal clear' view of the materials world by mapping geometry to text","url":"https://phys.org/news/2026-10-ai-crystal-view-materials-world.html","source":"Phys.org 化学","lang":"en","kind":"news","summaryZh":"Phys.org 化学于2026-10-20发布的文章，涉及AI 化学材料，原题《Giving AI a 'crystal clear' view of the materials world by mapping geometry to text》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"631c78f18e27","date":"2026-10-18","title":"Giving AI a “crystal clear” view of the materials world","url":"https://news.mit.edu/2026/giving-ai-crystal-clear-view-of-materials-world-1018","source":"MIT News：Artificial intelligence","lang":"en","kind":"news","summaryZh":"MIT News：Artificial intelligence于2026-10-18发布的文章，涉及AI 科研与调研、AI 化学材料，原题《Giving AI a “crystal clear” view of the materials world》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"4ca6cb0f4f59","date":"2026-09-22","title":"Ken Crutchfield: Lessons From Steel — Why LLMs Will Become Commodities","url":"https://www.lawnext.com/2026/09/ken-crutchfield-lessons-from-steel-why-llms-will-become-commodities.html","source":"LawNext / LawSites","lang":"en","kind":"news","summaryZh":"LawNext / LawSites于2026-09-23发布的文章，涉及AI 法律政策、AI 医疗、AI 商业融资，原题《Ken Crutchfield: Lessons From Steel — Why LLMs Will Become Commodities》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"6a0eb488f0a1","date":"2026-09-17","title":"Google backs Swedish green-steel project to tackle rising emissions","url":"https://www.canarymedia.com/articles/green-steel/google-backs-stegra-swedish-green-steel-project","source":"Canary Media","lang":"en","kind":"news","summaryZh":"Canary Media于2026-09-17发布的文章，涉及AI 气候能源，原题《Google backs Swedish green-steel project to tackle rising emissions》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"}],"review":{"lastReviewed":"2026-10-21","cycleDays":90,"nextReview":"2027-01-19"},"urls":{"zh":"/zh-hans/disciplines/materials/","en":"/en/disciplines/materials/"},"dimensions":{"method":{"score":3,"label":{"zh":"研究方法","en":"Research method"},"evidence":"机器学习原子间势和生成式模型已是计算材料学标准工具","source":"https://arxiv.org/abs/2401.00096","weight":0.3},"discovery":{"score":2,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"AI 预测的新超导体、沸石、合金获实验证实；但 A-Lab 被更正、GNoME 新颖性受质疑，大规模“AI 发现”成色有争议","source":"https://www.nature.com/articles/s41586-023-06734-w","weight":0.35},"adoption":{"score":2,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"工业界开始使用（MatterGen 开源），缺少普及率数据","source":null,"weight":0.25},"education":{"score":1,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到课程证据","source":null,"weight":0.1}},"compositeScore":2.2,"levelCapped":false,"levelChange":{"from":2,"to":2,"direction":"same","reason":null},"biggestBreakthrough":{"date":"2023-11-29","title":{"zh":"GNoME 预测 220 万种新晶体，约 38 万种稳定（Nature）","en":"GNoME predicts 2.2M new crystals, ~380k stable (Nature)"},"url":"https://www.nature.com/articles/s41586-023-06735-9","aiImpact":"leap","fieldImpact":"major","why":{"zh":"预测 220 万种晶体、约 38 万种稳定，稳定材料已知数量扩大约一个数量级；新颖性后遭部分学者质疑","en":"2.2M crystals, ~380k stable; novelty later questioned"}},"keyPapers":[{"title":"Scaling deep learning for materials discovery","titleZh":"GNoME 预测 220 万种新晶体，约 38 万种稳定（Nature）","id":"10.1038/s41586-023-06735-9","url":"https://doi.org/10.1038/s41586-023-06735-9","venue":"Nature","date":"2023-11-29","why":{"zh":"预测 220 万种晶体、约 38 万种稳定，稳定材料已知数量扩大约一个数量级；新颖性后遭部分学者质疑","en":"2.2M crystals, ~380k stable; novelty later questioned"},"fieldImpact":"major","aiImpact":"leap"},{"title":"An autonomous laboratory for the accelerated synthesis of inorganic materials","titleZh":"A-Lab 自主实验室 17 天合成新化合物（Nature；2026 年更正）","id":"10.1038/s41586-023-06734-w","url":"https://doi.org/10.1038/s41586-023-06734-w","venue":"Nature","date":"2023-11-29","why":{"zh":"自主实验室 17 天合成新化合物；2026-01 更正为 36/57，结论有争议","en":"Autonomous synthesis; corrected to 36/57 in Jan 2026"},"fieldImpact":"incremental","aiImpact":"major"},{"title":"A foundation model for atomistic materials chemistry","titleZh":"MACE-MP-0：覆盖周期表的通用机器学习原子间势基础模型","id":"arXiv:2401.00096","url":"https://arxiv.org/abs/2401.00096","venue":"arXiv","date":"2023-12-29","why":{"zh":"一个模型即可模拟大部分无机材料体系，大幅降低原子模拟门槛","en":"One model simulates most inorganic systems"},"fieldImpact":"major","aiImpact":"major"},{"title":"A generative model for inorganic materials design","titleZh":"MatterGen 生成式设计无机材料（Nature）","id":"10.1038/s41586-025-08628-5","url":"https://doi.org/10.1038/s41586-025-08628-5","venue":"Nature","date":"2025-01-16","why":{"zh":"生成式模型按目标性质设计新材料并合成验证一例","en":"Generative design with one synthesised validation"},"fieldImpact":"major","aiImpact":"major"},{"title":"Developing a complete AI-accelerated workflow for superconductor discovery","titleZh":"BEE-NET：从 130 万结构中筛选，2 个此前未知的化合物经实验证实超导（npj Comput. Mater.）","id":"10.1038/s41524-026-01964-8","url":"https://doi.org/10.1038/s41524-026-01964-8","venue":"npj Computational Materials","date":"2026-01-27","why":{"zh":"AI 预测新超导体获实验证实","en":"AI-predicted superconductors confirmed in the lab"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"DiffSyn: a generative diffusion approach to materials synthesis planning","titleZh":"DiffSyn：扩散模型规划合成路线，据此合成出热稳定性更好的新沸石（Nat. Comput. Sci.）","id":"10.1038/s43588-025-00949-9","url":"https://doi.org/10.1038/s43588-025-00949-9","venue":"Nature Computational Science","date":"2026-02-02","why":{"zh":"AI 从“预测结构”推进到“告诉你怎么合成”","en":"From predicting structures to planning how to make them"},"fieldImpact":"major","aiImpact":"major"},{"title":"Machine learning driven discovery of low modulus biomedical titanium alloys for additive manufacturing","titleZh":"机器学习设计增材制造专用低模量钛合金，经激光粉床熔融验证（Nat. Commun.）","id":"10.1038/s41467-026-71927-6","url":"https://doi.org/10.1038/s41467-026-71927-6","venue":"Nature Communications","date":"2026-04-14","why":{"zh":"AI 设计合金进入可制造的植入物工程验证","en":"AI-designed alloy validated for manufacturable implants"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"AI-accelerated scalable synthesis of nanocrystals for cost-effective photovoltaics","titleZh":"AI 加速钙钛矿纳米晶规模化合成：成本降至 2.16 美元/克，太阳能电池效率 19.37%（Nature Communications）","id":"10.1038/s41467-026-78268-4","url":"https://doi.org/10.1038/s41467-026-78268-4","venue":"Nature Communications","date":"2026-10-02","why":{"zh":"AI：机器学习与自动化流动合成闭环优化工艺；学科：把实验室材料推向低成本规模化制备","en":"AI: ML plus automated flow synthesis optimise the process; Field: pushes lab materials toward low-cost scale-up"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Bayesian\nConavigation of a Computational Physical\nModel and Atomic Force Microscopy Experiment to Autonomously Survey\na Combinatorial Materials Library","titleZh":"贝叶斯“共导航”：AFM 实验与动力学蒙特卡洛模型自主协同（ACS Nano）","id":"10.1021/acsnano.6c03268","url":"https://doi.org/10.1021/acsnano.6c03268","venue":"ACS Nano","date":"2026-08-17","why":{"zh":"实验与模拟自主协同，属方法改进","en":"Method improvement"},"fieldImpact":"incremental","aiImpact":"incremental"}],"latest_papers":[{"id":"arXiv:2610.29362","title":"Compositional neural networks for multimodal X-ray and neutron scattering inversion","titleZh":"用组合式神经网络做 X 射线与中子散射多模态反演","summaryZh":"把互补的 X 射线和中子散射测量整合进组合式神经网络，减轻缺少相位信息导致的结构反演不唯一问题。","summaryEn":"Compositional neural networks combine X-ray and neutron scattering to reduce non-uniqueness in structural inversion.","authors":["Valentin Munteanu","Alexander Hinderhofer","Dmitry Lapkin","Vladimir Starostin"],"date":"2026-10-21","url":"https://arxiv.org/abs/2610.29362","site":"arXiv","venue":null,"category":"cs.LG","fetchedAt":"2026-10-22T10:02:32"},{"id":"arXiv:2610.28629","title":"GBClaw: An agentic AI framework for atomistic grain boundary research and mechanistic discovery","titleZh":"GBClaw：用于原子尺度晶界研究与机理发现的智能体 AI 框架","summaryZh":"把大模型编排器、专用智能体、模拟接口和分层记忆组合起来，自动研究一般晶界的结构与变形；应用于镍中 14 个非对称倾转晶界，表征界面缺陷并识别多种变形模式。","summaryEn":"An LLM agent framework automating atomistic studies of general grain boundaries; applied to 14 Ni boundaries to identify deformation modes.","authors":["Yazhuo Liu","Kunqing Ding","Xiaoguo Gong","Zifeng Wang"],"date":"2026-10-20","url":"https://arxiv.org/abs/2610.28629","site":"arXiv","venue":null,"category":"cond-mat.mtrl-sci","fetchedAt":"2026-10-22T10:02:32"},{"id":"arXiv:2610.28529","title":"Symmetry-Guided Machine Learning of Spin-Dependent Nonlinear Hall Response","titleZh":"对称性引导的机器学习预测自旋相关非线性霍尔响应","summaryZh":"在蜂窝模型中分析贝里曲率偶极如何受结构对称性与磁矩取向调控，对称性约束的多项式模型对 256 个留出自旋构型的预测 R² 超过 0.9995。","summaryEn":"A symmetry-constrained model predicts the Berry-curvature dipole for held-out spin configurations with R² > 0.9995.","authors":["Jyotirmoy Sau"],"date":"2026-10-20","url":"https://arxiv.org/abs/2610.28529","site":"arXiv","venue":null,"category":"cond-mat.mtrl-sci","fetchedAt":"2026-10-22T10:02:32"},{"id":"arXiv:2610.28363","title":"TexFlow: learning crystallographic texture on SO(3) manifold via normalizing flows","titleZh":"TexFlow：用归一化流在 SO(3) 流形上学习晶体织构","summaryZh":"提出条件归一化流框架，直接学习晶粒取向在旋转群 SO(3) 上的概率分布，为“工艺–组织–性能”预测建模提供织构表示。","summaryEn":"A conditional normalising flow on SO(3) that models crystallographic texture for process–structure–property prediction.","authors":["Gowtham Nimmal Haribabu","Tarakram Ramgopal","Divakar Vetriveeran","Siddhant Kumar"],"date":"2026-10-20","url":"https://arxiv.org/abs/2610.28363","site":"arXiv","venue":null,"category":"cs.CE","fetchedAt":"2026-10-22T10:02:32"},{"id":"arXiv:2610.27762","title":"CrystalAR: Autoregressive Modeling for Symmetric Crystal Generation","titleZh":"CrystalAR：对称晶体生成的自回归建模","summaryZh":"把 Wyckoff 对称模板和完整晶体结构放进同一生成序列，用分层离散定位逐级细化原子位置，缓解对称模板与连续结构之间的矛盾。","summaryEn":"An autoregressive model jointly generating Wyckoff templates and full crystal structures via hierarchical discrete localisation.","authors":["Ziqiao Meng","Rongsheng Chen","Zhi Zheng","Nikita Kazeev"],"date":"2026-10-20","url":"https://arxiv.org/abs/2610.27762","site":"arXiv","venue":null,"category":"cs.AI","fetchedAt":"2026-10-22T10:02:32"},{"id":"arXiv:2610.27036","title":"MOFFieldBench: A Machine Learning Benchmark for Continuous Electronic Fields in Metal-Organic Frameworks","titleZh":"MOFFieldBench：金属有机框架连续电子场的机器学习基准","summaryZh":"提供 MOF 电子密度和静电势场数据与基准，推动把这些连续场而非仅原子结构作为机器学习表示。","summaryEn":"A benchmark of electron-density and electrostatic-potential fields for learning representations of metal–organic frameworks.","authors":["Zhaolin Hu","Hehe Fan","Chenhao Rao","Wangyihan Guo"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.27036","site":"arXiv","venue":null,"category":"cond-mat.mtrl-sci","fetchedAt":"2026-10-22T10:02:32"},{"id":"arXiv:2610.26301","title":"Machine Learning-Assisted Discovery of Carbon-Related Defect Formation Mechanisms in Diamond","titleZh":"机器学习辅助发现金刚石中碳相关缺陷的形成机制","summaryZh":"用机器学习辅助模拟辐照和退火过程，揭示金刚石中双间隙原子和双空位等缺陷的形成机制（与自旋量子比特相关）。","summaryEn":"ML-assisted simulations reveal formation mechanisms of di-interstitials and di-vacancies in diamond, relevant to spin qubits.","authors":["Grace Huang","Christopher Fivecoat","Xun Xu","Mengen Wang"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.26301","site":"arXiv","venue":null,"category":"cond-mat.mtrl-sci","fetchedAt":"2026-10-22T10:02:32"},{"id":"arXiv:2610.25775","title":"Crystal Structure Based Graph Neural Network And LLM Multi-Modal Approach for Material Magnetic Ordering Prediction","titleZh":"基于晶体结构图神经网络与大模型的多模态磁有序预测","summaryZh":"结合晶体结构图神经网络和大模型的多模态方法，从晶体结构预测材料的磁有序类型。","summaryEn":"Combines crystal-graph neural networks and an LLM to predict magnetic ordering from crystal structure.","authors":["Adam Zheng","April Zheng","Sheng Ran"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.25775","site":"arXiv","venue":null,"category":"cond-mat.mtrl-sci","fetchedAt":"2026-10-22T10:02:32"}]},{"id":"life-sciences","slug":"life-sciences","category":"life-medical","categoryName":{"zh":"生命与医学","en":"Life & medical sciences"},"name":{"zh":"生命科学（结构与分子生物学）","en":"Life sciences (structural & molecular biology)"},"classification":{"oecdFos":"1.6","moe":"0710 生物学"},"level":3,"levelKey":"transforming","levelLabel":{"zh":"系统改造","en":"Systemic transformation"},"trend":"up","reason":{"zh":"L3（接近 L4）：AlphaFold 让结构预测成为默认步骤并获 2024 诺贝尔化学奖；RFdiffusion 2025 年在 Nature 报告原子级精确的从头抗体设计，Evo 2（Nature 2026-03）可跨生命域建模与设计基因组。差距在于独立的从业者采用调查和课程层面的证据。","en":"L3 (close to L4): AlphaFold made structure prediction the default step and won the 2024 chemistry Nobel; RFdiffusion reported atomically accurate de novo antibodies (Nature 2025); Evo 2 (Nature, Mar 2026) models and designs genomes across life. Missing: independent adoption surveys and curriculum evidence."},"reasonSources":[{"url":"https://www.nature.com/articles/s41586-021-03819-2","linkCheck":"ok"},{"url":"https://doi.org/10.1038/s41586-025-09721-5","linkCheck":"ok"}],"directions":{"zh":["蛋白质结构预测","从头蛋白设计","蛋白质语言模型","生物分子复合物"],"en":["Protein structure prediction","De novo protein design","Protein language models","Biomolecular complexes"]},"stats":{"milestones":13,"milestonesSince2025":7,"milestones2026":5,"milestonesLast6Months":4,"news":12,"latestProgress":"2026-10-21","latestMilestone":"2026-10-21","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":0,"leaps":{"ai":3,"field":5},"latestPapers":8,"keyPapers":10},"milestones":[{"date":"2021-07-15","title":{"zh":"AlphaFold 2 论文发表（Nature）","en":"AlphaFold 2 paper published (Nature)"},"url":"https://www.nature.com/articles/s41586-021-03819-2","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"CASP14 中达到接近实验精度，解决困扰 50 年的蛋白质折叠预测问题","en":"Near-experimental accuracy at CASP14; the 50-year folding problem"},"aiImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"CASP14 中达到接近实验精度，解决困扰 50 年的蛋白质折叠预测问题","en":"Near-experimental accuracy at CASP14; the 50-year folding problem"}},"fieldImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"CASP14 中达到接近实验精度，解决困扰 50 年的蛋白质折叠预测问题","en":"Near-experimental accuracy at CASP14; the 50-year folding problem"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-021-03819-2","title":"Highly accurate protein structure prediction with AlphaFold","venue":"Nature","date":"2021-07-15"},"addedIn":"v1"},{"date":"2022-07-28","title":{"zh":"AlphaFold 数据库扩展到 2 亿多个蛋白质结构","en":"AlphaFold DB expands to 200M+ structures"},"url":"https://deepmind.google/discover/blog/alphafold-reveals-the-structure-of-the-protein-universe/","source":"Google DeepMind","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"dataset","typeLabel":{"zh":"数据集","en":"Dataset"},"why":{"zh":"几乎覆盖所有已知蛋白，2 亿多个结构免费开放","en":"200M+ structures, nearly all known proteins"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"几乎覆盖所有已知蛋白，2 亿多个结构免费开放","en":"200M+ structures, nearly all known proteins"}},"fieldImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"几乎覆盖所有已知蛋白，2 亿多个结构免费开放","en":"200M+ structures, nearly all known proteins"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2023-03-17","title":{"zh":"ESMFold：语言模型直接预测 6 亿多个宏基因组蛋白结构（Science）","en":"ESMFold: language-model structure prediction for 600M+ metagenomic proteins (Science)"},"url":"https://doi.org/10.1126/science.ade2574","source":"Science","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"把结构预测扩展到“暗物质”宏基因组蛋白","en":"Extends prediction to metagenomic “dark matter”"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"把结构预测扩展到“暗物质”宏基因组蛋白","en":"Extends prediction to metagenomic “dark matter”"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"把结构预测扩展到“暗物质”宏基因组蛋白","en":"Extends prediction to metagenomic “dark matter”"}},"impactNotes":[],"verified":{"id":"10.1126/science.ade2574","title":"Evolutionary-scale prediction of atomic-level protein structure with a language model","venue":"Science","date":"2023-03-17"},"addedIn":"v2"},{"date":"2023-07-11","title":{"zh":"RFdiffusion 从头设计蛋白质（Nature）","en":"RFdiffusion de novo protein design (Nature)"},"url":"https://www.nature.com/articles/s41586-023-06415-8","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"扩散模型从头设计结合蛋白，实验成功率大幅提高","en":"Diffusion-based de novo binder design with high success rates"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"扩散模型从头设计结合蛋白，实验成功率大幅提高","en":"Diffusion-based de novo binder design with high success rates"}},"fieldImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"扩散模型从头设计结合蛋白，实验成功率大幅提高","en":"Diffusion-based de novo binder design with high success rates"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-023-06415-8","title":"De novo design of protein structure and function with RFdiffusion","venue":"Nature","date":"2023-07-11"},"addedIn":"v1"},{"date":"2024-05-08","title":{"zh":"AlphaFold 3 预测蛋白质与 DNA/RNA/配体复合物（Nature）","en":"AlphaFold 3 predicts complexes with DNA/RNA/ligands (Nature)"},"url":"https://www.nature.com/articles/s41586-024-07487-w","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"预测扩展到蛋白-核酸-配体复合物","en":"Extends to complexes with nucleic acids and ligands"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"预测扩展到蛋白-核酸-配体复合物","en":"Extends to complexes with nucleic acids and ligands"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"预测扩展到蛋白-核酸-配体复合物","en":"Extends to complexes with nucleic acids and ligands"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-024-07487-w","title":"Accurate structure prediction of biomolecular interactions with AlphaFold 3","venue":"Nature","date":"2024-05-08"},"addedIn":"v1"},{"date":"2024-10-09","title":{"zh":"AlphaFold 与蛋白质设计获诺贝尔化学奖","en":"AlphaFold and protein design win the Chemistry Nobel"},"url":"https://www.nobelprize.org/prizes/chemistry/2024/summary/","source":"Nobel Prize","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"prize","typeLabel":{"zh":"奖项","en":"Prize"},"why":{"zh":"诺贝尔化学奖授予 AI 结构预测与蛋白质设计","en":"Nobel prize for AI structure prediction and design"},"aiImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"诺贝尔化学奖授予 AI 结构预测与蛋白质设计","en":"Nobel prize for AI structure prediction and design"}},"fieldImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"诺贝尔化学奖授予 AI 结构预测与蛋白质设计","en":"Nobel prize for AI structure prediction and design"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2025-01-16","title":{"zh":"ESM3 设计全新荧光蛋白（Science）","en":"ESM3 designs a new fluorescent protein (Science)"},"url":"https://www.science.org/doi/10.1126/science.ads0018","source":"Science","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"生成与已知荧光蛋白差异很大的新荧光蛋白（esmGFP）","en":"Generates a distant new fluorescent protein"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"生成与已知荧光蛋白差异很大的新荧光蛋白（esmGFP）","en":"Generates a distant new fluorescent protein"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"生成与已知荧光蛋白差异很大的新荧光蛋白（esmGFP）","en":"Generates a distant new fluorescent protein"}},"impactNotes":[],"verified":{"id":"10.1126/science.ads0018","title":"Simulating 500 million years of evolution with a language model","venue":"Science","date":"2025-02-21"},"addedIn":"v1"},{"date":"2025-11-05","title":{"zh":"RFdiffusion 实现原子级精确的从头抗体设计（Nature）","en":"Atomically accurate de novo design of antibodies with RFdiffusion (Nature)"},"url":"https://doi.org/10.1038/s41586-025-09721-5","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"抗体此前被认为最难从头设计；冷冻电镜结构证实设计精度","en":"Antibodies were the hardest de novo design target; cryo-EM confirms accuracy"},"aiImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"抗体此前被认为最难从头设计；冷冻电镜结构证实设计精度","en":"Antibodies were the hardest de novo design target; cryo-EM confirms accuracy"}},"fieldImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"抗体此前被认为最难从头设计；冷冻电镜结构证实设计精度","en":"Antibodies were the hardest de novo design target; cryo-EM confirms accuracy"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-025-09721-5","title":"Atomically accurate de novo design of antibodies with RFdiffusion","venue":"Nature","date":"2025-11-05"},"addedIn":"v2"},{"date":"2026-02-19","title":{"zh":"Isomorphic Labs 发布药物设计引擎 IsoDDE，被称“AlphaFold 4”（未公开）","en":"Isomorphic Labs unveils IsoDDE, dubbed “an AlphaFold 4” (not public)"},"url":"https://www.nature.com/articles/d41586-026-00365-7","source":"Nature News","flags":["vendor"],"flagLabels":[{"zh":"厂商自述","en":"Vendor claim"}],"linkCheck":"ok","datePrecision":"day","type":"system","typeLabel":{"zh":"系统发布","en":"System"},"why":{"zh":"厂商称性能大幅提升但未公开细节","en":"Vendor claim, undisclosed"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"厂商称性能大幅提升但未公开细节","en":"Vendor claim, undisclosed"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"厂商称性能大幅提升但未公开细节","en":"Vendor claim, undisclosed"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-08-06","title":{"zh":"Germinal：AI 设计抗体，针对 4 个靶点得到经实验验证的纳摩尔级结合物（Nature Methods）","en":"Germinal AI-designed antibodies with validated nanomolar binders for four targets (Nature Methods)"},"url":"https://doi.org/10.1038/s41592-026-03190-y","source":"Nature Methods","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"抗体 AI 设计从个案走向可复用流程","en":"Antibody design becomes a reusable pipeline"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"抗体 AI 设计从个案走向可复用流程","en":"Antibody design becomes a reusable pipeline"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"抗体 AI 设计从个案走向可复用流程","en":"Antibody design becomes a reusable pipeline"}},"impactNotes":[],"verified":{"id":"10.1038/s41592-026-03190-y","title":"AI-designed antibodies with Germinal","venue":"Nature Methods","date":"2026-08-06"},"addedIn":"v2"},{"date":"2026-10-14","title":{"zh":"计算设计金属蛋白酶并实验验证活性（Nature）","en":"Computational design of metalloproteases (Nature)"},"url":"https://www.nature.com/articles/s41586-026-11167-2","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"AI：AI 管线设计出可切割肽键的全新金属蛋白酶；学科：蛋白酶长期难以从头设计，同期另有半胱氨酸蛋白酶从头设计（Nature s41586-026-11143-w）","en":"AI: an AI pipeline designs new metalloproteases that cleave peptide bonds; Field: proteases were long out of reach for de novo design"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：AI 管线设计出可切割肽键的全新金属蛋白酶；学科：蛋白酶长期难以从头设计，同期另有半胱氨酸蛋白酶从头设计（Nature s41586-026-11143-w）","en":"AI: an AI pipeline designs new metalloproteases that cleave peptide bonds; Field: proteases were long out of reach for de novo design"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：AI 管线设计出可切割肽键的全新金属蛋白酶；学科：蛋白酶长期难以从头设计，同期另有半胱氨酸蛋白酶从头设计（Nature s41586-026-11143-w）","en":"AI: an AI pipeline designs new metalloproteases that cleave peptide bonds; Field: proteases were long out of reach for de novo design"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-026-11167-2","title":"Computational design of metalloproteases","venue":"Nature","date":"2026-10-14"},"addedIn":"v2"},{"date":"2026-10-15","title":{"zh":"RFdiffusion3：原子级功能位点搭建，从头设计磷酸三酯酶与锌蛋白酶（Science）","en":"RFdiffusion3: atomic functional-site scaffolding in biopolymers (Science)"},"url":"https://doi.org/10.1126/science.aeh2982","source":"Science","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"AI：扩散模型直接在全原子层面围绕催化位点生成蛋白；学科：同期 Nature 两篇（金属蛋白酶、半胱氨酸蛋白酶从头设计）实验验证了活性，酶设计从结合蛋白走向催化","en":"AI: diffusion generates proteins at all-atom level around catalytic sites; Field: with two Nature papers on de novo proteases, enzyme design moves from binders to catalysis"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：扩散模型直接在全原子层面围绕催化位点生成蛋白；学科：同期 Nature 两篇（金属蛋白酶、半胱氨酸蛋白酶从头设计）实验验证了活性，酶设计从结合蛋白走向催化","en":"AI: diffusion generates proteins at all-atom level around catalytic sites; Field: with two Nature papers on de novo proteases, enzyme design moves from binders to catalysis"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：扩散模型直接在全原子层面围绕催化位点生成蛋白；学科：同期 Nature 两篇（金属蛋白酶、半胱氨酸蛋白酶从头设计）实验验证了活性，酶设计从结合蛋白走向催化","en":"AI: diffusion generates proteins at all-atom level around catalytic sites; Field: with two Nature papers on de novo proteases, enzyme design moves from binders to catalysis"}},"impactNotes":[],"verified":{"id":"10.1126/science.aeh2982","title":"Atomic functional site scaffolding in biopolymers with RFdiffusion3","venue":"Science","date":"2026-10-15"},"addedIn":"v2"},{"date":"2026-10-21","title":{"zh":"西湖大学等发布 Pinal：160 亿参数模型直接从自然语言功能描述设计蛋白质，荧光蛋白、PET 水解酶、醇脱氢酶和 H 蛋白四类设计均经实验验证（Nature）","en":"Pinal: a 16B-parameter foundation model that designs functional proteins from natural-language descriptions, validated in four experimental tasks (Nature)"},"url":"https://www.nature.com/articles/s41586-026-11205-z","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"model","typeLabel":{"zh":"模型发布","en":"Model"},"why":{"zh":"AI：在 17 亿条蛋白–文本对上训练，首次把“用自然语言描述功能”作为蛋白设计的主要入口，模型与网页服务开源；学科：四类设计均观测到预期功能，未经迭代优化的 H 蛋白使 CO₂ 固定通路产物滴度提高 1.7 倍，但仍需计算筛选后再实验","en":"AI: trained on 1.7B protein–text pairs, makes natural language the main interface for protein design (open weights). Field: intended function observed in all four tasks; an unoptimised H-protein raised CO₂-fixation titre 1.7-fold"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：在 17 亿条蛋白–文本对上训练，首次把“用自然语言描述功能”作为蛋白设计的主要入口，模型与网页服务开源；学科：四类设计均观测到预期功能，未经迭代优化的 H 蛋白使 CO₂ 固定通路产物滴度提高 1.7 倍，但仍需计算筛选后再实验","en":"AI: trained on 1.7B protein–text pairs, makes natural language the main interface for protein design (open weights). Field: intended function observed in all four tasks; an unoptimised H-protein raised CO₂-fixation titre 1.7-fold"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：在 17 亿条蛋白–文本对上训练，首次把“用自然语言描述功能”作为蛋白设计的主要入口，模型与网页服务开源；学科：四类设计均观测到预期功能，未经迭代优化的 H 蛋白使 CO₂ 固定通路产物滴度提高 1.7 倍，但仍需计算筛选后再实验","en":"AI: trained on 1.7B protein–text pairs, makes natural language the main interface for protein design (open weights). Field: intended function observed in all four tasks; an unoptimised H-protein raised CO₂-fixation titre 1.7-fold"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-026-11205-z","title":"Programmable design of functional proteins from natural language","venue":"Nature","date":"2026-10-21"},"addedIn":"v2-20261022pm"}],"papersRepos":[{"title":"google-deepmind/alphafold3","url":"https://github.com/google-deepmind/alphafold3","date":null,"kind":"repo","linkCheck":"ok"},{"title":"RosettaCommons/RFdiffusion","url":"https://github.com/RosettaCommons/RFdiffusion","date":null,"kind":"repo","linkCheck":"ok"},{"title":"evolutionaryscale/esm","url":"https://github.com/evolutionaryscale/esm","date":null,"kind":"repo","linkCheck":"ok"},{"title":"jwohlwend/boltz","url":"https://github.com/jwohlwend/boltz","date":null,"kind":"repo","linkCheck":"ok"},{"title":"Highly accurate protein structure prediction with AlphaFold","url":"https://doi.org/10.1038/s41586-021-03819-2","date":"2021-07-15","kind":"paper","linkCheck":"ok-crossref"},{"title":"De novo design of protein structure and function with RFdiffusion","url":"https://doi.org/10.1038/s41586-023-06415-8","date":"2023-07-11","kind":"paper","linkCheck":"ok-crossref"},{"title":"Atomically accurate de novo design of antibodies with RFdiffusion","url":"https://doi.org/10.1038/s41586-025-09721-5","date":"2025-11-05","kind":"paper","linkCheck":"ok-crossref"},{"title":"Evolutionary-scale prediction of atomic-level protein structure with a language model","url":"https://doi.org/10.1126/science.ade2574","date":"2023-03-17","kind":"paper","linkCheck":"ok-crossref"},{"title":"Accurate structure prediction of biomolecular interactions with AlphaFold 3","url":"https://doi.org/10.1038/s41586-024-07487-w","date":"2024-05-08","kind":"paper","linkCheck":"ok-crossref"},{"title":"Simulating 500 million years of evolution with a language model","url":"https://doi.org/10.1126/science.ads0018","date":"2025-02-21","kind":"paper","linkCheck":"ok-crossref"},{"title":"AI-designed antibodies with Germinal","url":"https://doi.org/10.1038/s41592-026-03190-y","date":"2026-08-06","kind":"paper","linkCheck":"ok-crossref"},{"title":"Computational design of metalloproteases","url":"https://doi.org/10.1038/s41586-026-11167-2","date":"2026-10-14","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"Google DeepMind / Isomorphic Labs","url":"https://www.isomorphiclabs.com"},{"name":"UW Institute for Protein Design","url":"https://www.ipd.uw.edu"},{"name":"EvolutionaryScale","url":"https://www.evolutionaryscale.ai"},{"name":"EMBL-EBI AlphaFold DB","url":"https://alphafold.ebi.ac.uk"}],"news":[{"id":"1dd5c007c495","date":"2026-10-20","title":"Nanoscale tech tracks proteins inside living neurons to guide Alzheimer's and Parkinson's drug discovery","url":"https://phys.org/news/2026-10-nanoscale-tech-tracks-proteins-neurons.html","source":"Phys.org 化学·材料(纳米)","lang":"en","kind":"news","summaryZh":"Phys.org 化学·材料(纳米)于2026-10-20发布的文章，涉及AI 化学材料、AI 生物/医药、AI 教育，原题《Nanoscale tech tracks proteins inside living neurons to guide Alzheimer's and Parkinson's 》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"00a0631a5da4","date":"2026-10-20","title":"CADS: Measuring the Incremental Contribution of LLM Proposers in Sequential Biological Screens","url":"https://arxiv.org/abs/2610.24597","source":"arXiv q-bio","lang":"en","kind":"paper","summaryZh":"arXiv q-bio于2026-10-20发布的论文，涉及AI 生物/医药、AI 商业融资、AI 软件开发，原题《CADS: Measuring the Incremental Contribution of LLM Proposers in Sequential Biological Scr》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"cda3e4a91ff6","date":"2026-10-20","title":"Disentanglement with Biological Constraints: A Theory of Functional Cell Types","url":"https://arxiv.org/abs/2210.01768","source":"arXiv q-bio","lang":"en","kind":"paper","summaryZh":"arXiv q-bio于2026-10-20发布的论文，涉及AI 生物/医药、AI 气候能源，原题《Disentanglement with Biological Constraints: A Theory of Functional Cell Types》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"b80f714c6281","date":"2026-10-19","title":"ConforFlux: Particle-Guided Trunk Repulsion for Diverse Protein Conformations","url":"https://www.biorxiv.org/content/10.64898/2026.05.16.725138v2","source":"bioRxiv API · 生物信息","lang":"en","kind":"paper","summaryZh":"bioRxiv API · 生物信息于2026-10-19发布的论文，涉及AI 生物/医药、AI 软件开发，原题《ConforFlux: Particle-Guided Trunk Repulsion for Diverse Protein Conformations》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"081a8c34a658","date":"2026-10-16","title":"Grounding AI agents in biological reality: Building blocks for autonomous wet labs","url":"https://research.google/blog/grounding-ai-agents-in-biological-reality-building-blocks-for-autonomous-wet-labs/","source":"Google Research Blog","lang":"en","kind":"news","summaryZh":"Google Research Blog于2026-10-17发布的文章，涉及AI 科研与调研、AI 生物/医药，原题《Grounding AI agents in biological reality: Building blocks for autonomous wet labs》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"feaf5f659071","date":"2026-10-15","title":"Hybrid modeling for improved structure prediction of large protein complexes","url":"https://www.nature.com/articles/s43588-026-01067-w","source":"Nature Computational Science","lang":"en","kind":"news","summaryZh":"Nature Computational Science于2026-10-15发布的文章，涉及AI 科研与调研、AI 生物/医药、AI 创意媒体，原题《Hybrid modeling for improved structure prediction of large protein complexes》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"9e65ecccdd2a","date":"2026-10-14","title":"STAT+: The startup chasing ‘general biological intelligence’","url":"https://www.statnews.com/2026/10/14/the-startup-chasing-general-biological-intelligence-ai-prognosis/?utm_campaign=rss","source":"STAT 话题：Artificial intelligence","lang":"en","kind":"news","summaryZh":"STAT 话题：Artificial intelligence于2026-10-14发布的文章，涉及AI 生物/医药、AI 商业融资，原题《STAT+: The startup chasing ‘general biological intelligence’》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"972ce96df38a","date":"2026-10-14","title":"Researchers use AI to design proteins that cut up other proteins","url":"https://arstechnica.com/science/2026/10/researchers-use-ai-to-design-proteins-that-cut-up-other-proteins/","source":"Ars Technica AI","lang":"en","kind":"news","summaryZh":"Ars Technica AI于2026-10-15发布的文章，涉及AI 生物/医药，原题《Researchers use AI to design proteins that cut up other proteins》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"d2e5912a3db6","date":"2026-10-10","title":"Why AlphaFold Didn't Solve Protein Folding — Pushmeet Kohli, Google DeepMind & Sal Candido, Biohub","url":"https://www.latent.space/p/biohub-deepmind","source":"Latent Space","lang":"en","kind":"news","summaryZh":"Latent Space于2026-10-10发布的文章，涉及AI 生物/医药，原题《Why AlphaFold Didn't Solve Protein Folding — Pushmeet Kohli, Google DeepMind & Sal Candido》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"f91905c59acb","date":"2026-10-09","title":"Chai Discovery signs GSK deal to make biologics with AI","url":"https://endpoints.news/chai-discovery-signs-gsk-deal-to-make-biologics-with-ai/","source":"Endpoints News","lang":"en","kind":"news","summaryZh":"Chai Discovery 与 GSK 签约，用其 AI 模型设计生物药（抗体等），继礼来等之后再添一家大药企客户。","via":"research-samples-20261010","linkCheck":"from-feed"},{"id":"2f5a5f8b9182","date":"2026-10-05","title":"Designing enzymes for new-to-nature chemistry and non-natural substrates with AlphaProtein Novo","url":"https://www.biorxiv.org/content/10.64898/2026.10.01.756017v1","source":"bioRxiv (synthetic biology)","lang":"en","kind":"news","summaryZh":"bioRxiv：AlphaProtein Novo 从头设计能催化非天然反应、作用于非天然底物的酶。","via":"research-samples-20261010","linkCheck":"from-feed"},{"id":"1abd1ab6b8cd","date":"2026-09-28","title":"Opinion: AI is eroding the barriers that kept biological weapons rare","url":"https://www.statnews.com/2026/09/28/ai-bioweapons-pathogens-guardrails-policy-warning/?utm_campaign=rss","source":"STAT 话题：Artificial intelligence","lang":"en","kind":"news","summaryZh":"STAT 话题：Artificial intelligence于2026-09-28发布的文章，涉及AI 生物/医药，原题《Opinion: AI is eroding the barriers that kept biological weapons rare》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"}],"review":{"lastReviewed":"2026-10-22","cycleDays":60,"nextReview":"2026-12-20"},"urls":{"zh":"/zh-hans/disciplines/life-sciences/","en":"/en/disciplines/life-sciences/"},"dimensions":{"method":{"score":4,"label":{"zh":"研究方法","en":"Research method"},"evidence":"结构预测已是结构生物学默认第一步，AlphaFold 数据库覆盖 2 亿多个结构","source":"https://deepmind.google/discover/blog/alphafold-reveals-the-structure-of-the-protein-universe/","weight":0.3},"discovery":{"score":4,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"诺贝尔奖级成果；从头设计蛋白、抗体（Nature 2025）经实验结构验证","source":"https://doi.org/10.1038/s41586-025-09721-5","weight":0.35},"adoption":{"score":3,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"工具普遍使用，但缺少独立的使用率调查","source":null,"weight":0.25},"education":{"score":1,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到课程层面证据","source":null,"weight":0.1}},"compositeScore":3.45,"levelCapped":false,"levelChange":{"from":3,"to":3,"direction":"same","reason":null},"biggestBreakthrough":{"date":"2025-11-05","title":{"zh":"RFdiffusion 实现原子级精确的从头抗体设计（Nature）","en":"Atomically accurate de novo design of antibodies with RFdiffusion (Nature)"},"url":"https://doi.org/10.1038/s41586-025-09721-5","aiImpact":"leap","fieldImpact":"leap","why":{"zh":"抗体此前被认为最难从头设计；冷冻电镜结构证实设计精度","en":"Antibodies were the hardest de novo design target; cryo-EM confirms accuracy"}},"keyPapers":[{"title":"Highly accurate protein structure prediction with AlphaFold","titleZh":"AlphaFold 2 论文发表（Nature）","id":"10.1038/s41586-021-03819-2","url":"https://doi.org/10.1038/s41586-021-03819-2","venue":"Nature","date":"2021-07-15","why":{"zh":"CASP14 中达到接近实验精度，解决困扰 50 年的蛋白质折叠预测问题","en":"Near-experimental accuracy at CASP14; the 50-year folding problem"},"fieldImpact":"leap","aiImpact":"leap"},{"title":"De novo design of protein structure and function with RFdiffusion","titleZh":"RFdiffusion 从头设计蛋白质（Nature）","id":"10.1038/s41586-023-06415-8","url":"https://doi.org/10.1038/s41586-023-06415-8","venue":"Nature","date":"2023-07-11","why":{"zh":"扩散模型从头设计结合蛋白，实验成功率大幅提高","en":"Diffusion-based de novo binder design with high success rates"},"fieldImpact":"leap","aiImpact":"major"},{"title":"Atomically accurate de novo design of antibodies with RFdiffusion","titleZh":"RFdiffusion 实现原子级精确的从头抗体设计（Nature）","id":"10.1038/s41586-025-09721-5","url":"https://doi.org/10.1038/s41586-025-09721-5","venue":"Nature","date":"2025-11-05","why":{"zh":"抗体此前被认为最难从头设计；冷冻电镜结构证实设计精度","en":"Antibodies were the hardest de novo design target; cryo-EM confirms accuracy"},"fieldImpact":"leap","aiImpact":"leap"},{"title":"Evolutionary-scale prediction of atomic-level protein structure with a language model","titleZh":"ESMFold：语言模型直接预测 6 亿多个宏基因组蛋白结构（Science）","id":"10.1126/science.ade2574","url":"https://doi.org/10.1126/science.ade2574","venue":"Science","date":"2023-03-17","why":{"zh":"把结构预测扩展到“暗物质”宏基因组蛋白","en":"Extends prediction to metagenomic “dark matter”"},"fieldImpact":"major","aiImpact":"major"},{"title":"Accurate structure prediction of biomolecular interactions with AlphaFold 3","titleZh":"AlphaFold 3 预测蛋白质与 DNA/RNA/配体复合物（Nature）","id":"10.1038/s41586-024-07487-w","url":"https://doi.org/10.1038/s41586-024-07487-w","venue":"Nature","date":"2024-05-08","why":{"zh":"预测扩展到蛋白-核酸-配体复合物","en":"Extends to complexes with nucleic acids and ligands"},"fieldImpact":"major","aiImpact":"major"},{"title":"Simulating 500 million years of evolution with a language model","titleZh":"ESM3 设计全新荧光蛋白（Science）","id":"10.1126/science.ads0018","url":"https://doi.org/10.1126/science.ads0018","venue":"Science","date":"2025-02-21","why":{"zh":"生成与已知荧光蛋白差异很大的新荧光蛋白（esmGFP）","en":"Generates a distant new fluorescent protein"},"fieldImpact":"major","aiImpact":"major"},{"title":"AI-designed antibodies with Germinal","titleZh":"Germinal：AI 设计抗体，针对 4 个靶点得到经实验验证的纳摩尔级结合物（Nature Methods）","id":"10.1038/s41592-026-03190-y","url":"https://doi.org/10.1038/s41592-026-03190-y","venue":"Nature Methods","date":"2026-08-06","why":{"zh":"抗体 AI 设计从个案走向可复用流程","en":"Antibody design becomes a reusable pipeline"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Computational design of metalloproteases","titleZh":"计算设计金属蛋白酶并实验验证活性（Nature）","id":"10.1038/s41586-026-11167-2","url":"https://doi.org/10.1038/s41586-026-11167-2","venue":"Nature","date":"2026-10-14","why":{"zh":"AI：AI 管线设计出可切割肽键的全新金属蛋白酶；学科：蛋白酶长期难以从头设计，同期另有半胱氨酸蛋白酶从头设计（Nature s41586-026-11143-w）","en":"AI: an AI pipeline designs new metalloproteases that cleave peptide bonds; Field: proteases were long out of reach for de novo design"},"fieldImpact":"major","aiImpact":"major"},{"title":"Atomic functional site scaffolding in biopolymers with RFdiffusion3","titleZh":"RFdiffusion3：原子级功能位点搭建，从头设计磷酸三酯酶与锌蛋白酶（Science）","id":"10.1126/science.aeh2982","url":"https://doi.org/10.1126/science.aeh2982","venue":"Science","date":"2026-10-15","why":{"zh":"AI：扩散模型直接在全原子层面围绕催化位点生成蛋白；学科：同期 Nature 两篇（金属蛋白酶、半胱氨酸蛋白酶从头设计）实验验证了活性，酶设计从结合蛋白走向催化","en":"AI: diffusion generates proteins at all-atom level around catalytic sites; Field: with two Nature papers on de novo proteases, enzyme design moves from binders to catalysis"},"fieldImpact":"major","aiImpact":"major"},{"title":"Programmable design of functional proteins from natural language","titleZh":"西湖大学等发布 Pinal：160 亿参数模型直接从自然语言功能描述设计蛋白质，荧光蛋白、PET 水解酶、醇脱氢酶和 H 蛋白四类设计均经实验验证（Nature）","id":"10.1038/s41586-026-11205-z","url":"https://www.nature.com/articles/s41586-026-11205-z","venue":"Nature","date":"2026-10-21","why":{"zh":"AI：在 17 亿条蛋白–文本对上训练，首次把“用自然语言描述功能”作为蛋白设计的主要入口，模型与网页服务开源；学科：四类设计均观测到预期功能，未经迭代优化的 H 蛋白使 CO₂ 固定通路产物滴度提高 1.7 倍，但仍需计算筛选后再实验","en":"AI: trained on 1.7B protein–text pairs, makes natural language the main interface for protein design (open weights). Field: intended function observed in all four tasks; an unoptimised H-protein raised CO₂-fixation titre 1.7-fold"},"fieldImpact":"major","aiImpact":"major"}],"latest_papers":[{"id":"doi:10.64898/2026.10.14.759239","title":"Facilitating Valid Statistical Inference in Biomedical Image Synthesis","titleZh":"让生物医学合成图像支持有效的统计推断","summaryZh":"检验生成式 AI 合成的 MRI 图像能否支撑可靠的下游统计分析和生物学推断，而不只是看起来相似。","summaryEn":"Tests whether generative-AI synthetic MRI images support valid downstream statistical and biological inference, not just visual similarity.","authors":["Xiong, J.","Bao, S.","Liu, Q.","Lau, K."],"date":"2026-10-16","url":"https://www.biorxiv.org/content/10.64898/2026.10.14.759239v1","site":"bioRxiv","venue":null,"category":"bioinformatics","fetchedAt":"2026-10-22T10:13:27"},{"id":"doi:10.64898/2026.06.28.735057","title":"Homology-aware cross-validation strategies for generalization assessment in RNA secondary structure prediction","titleZh":"面向 RNA 二级结构预测泛化评估的同源感知交叉验证","summaryZh":"针对深度学习 RNA 二级结构预测中同源数据泄漏导致的过度乐观，比较并提出考虑同源关系的交叉验证划分策略。","summaryEn":"Proposes homology-aware cross-validation to avoid leakage-driven over-optimism in deep-learning RNA secondary structure prediction.","authors":["Bugnon, L.","Kulemeyer, G.","Gerard, M.","Di Persia, L."],"date":"2026-10-16","url":"https://www.biorxiv.org/content/10.64898/2026.06.28.735057v2","site":"bioRxiv","venue":null,"category":"bioinformatics","fetchedAt":"2026-10-22T10:13:27"},{"id":"arXiv:2610.22562","title":"Evolutionary Informed Translation from Protein to Nucleic Acid Sequence","titleZh":"考虑进化信息的蛋白质到核酸序列翻译","summaryZh":"提出 PINATA 模型，利用系统发育和序列比对信息，为给定蛋白生成更自然的物种特异核酸序列，避免过度偏好高频密码子。","summaryEn":"PINATA uses phylogenetic alignments to translate proteins into more natural, species-specific nucleic-acid sequences.","authors":["Anna Hart","Mohammed El-Kebir","Huimin Zhao","Heng Ji"],"date":"2026-10-15","url":"https://arxiv.org/abs/2610.22562","site":"arXiv","venue":null,"category":"q-bio.BM","fetchedAt":"2026-10-22T10:13:27"},{"id":"doi:10.64898/2026.10.06.757114","title":"ExpertTwins: Auditable Long Horizon Scientific Panels through Grounded Expertise","titleZh":"ExpertTwins：基于可审计证据策略的长程科学专家组","summaryZh":"让多智能体科学专家组中每个席位按可检查的证据策略检索和判断，避免角色标签式讨论中智能体互相趋同。","summaryEn":"Defines each seat of a multi-agent scientific panel by an inspectable evidence policy so agents do not simply converge.","authors":["Foroughi pour, A.","Shirinifard, A."],"date":"2026-10-15","url":"https://www.biorxiv.org/content/10.64898/2026.10.06.757114v1","site":"bioRxiv","venue":null,"category":"bioinformatics","fetchedAt":"2026-10-22T10:13:27"},{"id":"doi:10.64898/2026.06.04.730241","title":"scMTG reconstructs single-cell temporal dynamics with Markov transition generators","titleZh":"scMTG：用马尔可夫转移生成器重建单细胞时间动态","summaryZh":"针对细胞被破坏性测量、无法追踪的问题，提出 scMTG，从单细胞时间序列数据中重建细胞状态转变和动态调控程序。","summaryEn":"scMTG reconstructs cell-state transitions from destructive single-cell time-series data.","authors":["Cui, X.","Wang, H.","Gao, Z.","Jiang, R."],"date":"2026-10-14","url":"https://www.biorxiv.org/content/10.64898/2026.06.04.730241v2","site":"bioRxiv","venue":null,"category":"bioinformatics","fetchedAt":"2026-10-22T10:13:27"},{"id":"doi:10.64898/2026.04.30.721585","title":"Joint Variable Selection for Omic Biomarkers in Time-to-Event Data","titleZh":"生存时间数据中组学生物标志物的联合变量选择","summaryZh":"在大型生物样本库中把电子病历的诊断时间与多组学数据关联，联合选择与疾病发病相关的基因和通路。","summaryEn":"Joint variable selection linking time-to-diagnosis in biobank EHRs to multi-omic measurements.","authors":["Bajzik, J.","Depope, A.","Zolfimoselo, Y.","Sharipov, A."],"date":"2026-10-14","url":"https://www.biorxiv.org/content/10.64898/2026.04.30.721585v2","site":"bioRxiv","venue":null,"category":"bioinformatics","fetchedAt":"2026-10-22T10:13:27"},{"id":"doi:10.64898/2026.10.07.757259","title":"DNA-LLM2Vec: Architecture-Aware Task-Conditioned Adaptation of Decoder DNA Foundation Models","titleZh":"DNA-LLM2Vec：对解码器式 DNA 基础模型做架构感知的任务条件适配","summaryZh":"把单向的解码器 DNA 基础模型分阶段适配为能利用上下游双向语境的表示模型，用于基因组任务。","summaryEn":"Adapts directional decoder DNA foundation models into bidirectional, task-conditioned representations for genomic tasks.","authors":["Liang, Y.","Zhai, K.","Gan, Y.","Li, S."],"date":"2026-10-14","url":"https://www.biorxiv.org/content/10.64898/2026.10.07.757259v1","site":"bioRxiv","venue":null,"category":"bioinformatics","fetchedAt":"2026-10-22T10:13:27"},{"id":"doi:10.64898/2026.10.06.756984","title":"Do LLM Proposers Improve Budgeted Search for Ionizable Lipids?","titleZh":"大模型提案者能改善可电离脂质的有限预算搜索吗？","summaryZh":"提出 LEAF 协议，在 30044 个候选可电离脂质空间和相同实验预算下，检验大模型提出候选是否优于经典搜索方法。","summaryEn":"LEAF tests, at matched budget over 30,044 ionisable lipids, whether LLM proposers beat classical search.","authors":["Singh, H.","SINGH, R. K.","Krishna, C. S.","Srivastava, S. P."],"date":"2026-10-13","url":"https://www.biorxiv.org/content/10.64898/2026.10.06.756984v1","site":"bioRxiv","venue":null,"category":"bioinformatics","fetchedAt":"2026-10-22T10:13:27"}]},{"id":"genomics","slug":"genomics","category":"life-medical","categoryName":{"zh":"生命与医学","en":"Life & medical sciences"},"name":{"zh":"基因组学与遗传学","en":"Genomics & genetics"},"classification":{"oecdFos":"1.6","moe":"071007 遗传学"},"level":3,"levelKey":"transforming","levelLabel":{"zh":"系统改造","en":"Systemic transformation"},"trend":"up","reason":{"zh":"深度学习变异检测（DeepVariant）早已进入测序流程；AlphaMissense 给出 7100 万个错义变异的致病性预测（Science 2023），AlphaGenome 可读入 1Mb DNA 预测数千种功能轨迹（Nature 2026-01），基因组语言模型 Evo 已能生成功能序列。","en":"Deep-learning variant calling (DeepVariant) is in standard pipelines; AlphaMissense scores 71M missense variants (Science 2023); AlphaGenome reads 1 Mb of DNA to predict thousands of tracks (Nature, Jan 2026); genome LMs like Evo generate functional sequences."},"reasonSources":[{"url":"https://www.science.org/doi/10.1126/science.adg7492","linkCheck":"ok-crossref"},{"url":"https://www.nature.com/articles/s41586-025-10014-0","linkCheck":"ok"}],"directions":{"zh":["变异效应预测","调控基因组建模","DNA 语言模型","单细胞基础模型"],"en":["Variant effect prediction","Regulatory genomics","DNA language models","Single-cell foundation models"]},"stats":{"milestones":8,"milestonesSince2025":4,"milestones2026":4,"milestonesLast6Months":2,"news":12,"latestProgress":"2026-10-21","latestMilestone":"2026-10-21","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":0,"leaps":{"ai":1,"field":0},"latestPapers":7,"keyPapers":6},"milestones":[{"date":"2018-09-24","title":{"zh":"DeepVariant 深度学习变异检测（Nature Biotechnology）","en":"DeepVariant deep-learning variant caller (Nature Biotechnology)"},"url":"https://www.nature.com/articles/nbt.4235","source":"Nature Biotechnology","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"system","typeLabel":{"zh":"系统发布","en":"System"},"why":{"zh":"把变异检测变成图像分类问题，精度超过传统流程并进入标准工具链","en":"Variant calling as image classification; entered standard pipelines"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"把变异检测变成图像分类问题，精度超过传统流程并进入标准工具链","en":"Variant calling as image classification; entered standard pipelines"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"把变异检测变成图像分类问题，精度超过传统流程并进入标准工具链","en":"Variant calling as image classification; entered standard pipelines"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2021-10-04","title":{"zh":"Enformer：Transformer 结合长程相互作用预测基因表达（Nature Methods）","en":"Enformer predicts gene expression from sequence with long-range interactions (Nature Methods)"},"url":"https://doi.org/10.1038/s41592-021-01252-x","source":"Nature Methods","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"把序列到功能预测的上下文扩展到 10 万碱基","en":"Extends sequence-to-function context to 100 kb"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"把序列到功能预测的上下文扩展到 10 万碱基","en":"Extends sequence-to-function context to 100 kb"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"把序列到功能预测的上下文扩展到 10 万碱基","en":"Extends sequence-to-function context to 100 kb"}},"impactNotes":[],"verified":{"id":"10.1038/s41592-021-01252-x","title":"Effective gene expression prediction from sequence by integrating long-range interactions","venue":"Nature Methods","date":"2021-10-04"},"addedIn":"v2"},{"date":"2023-09-19","title":{"zh":"AlphaMissense 预测 7100 万个错义变异（Science）","en":"AlphaMissense predicts 71M missense variants (Science)"},"url":"https://www.science.org/doi/10.1126/science.adg7492","source":"Science","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"对 7100 万个错义变异给出致病性预测，覆盖大部分未知意义变异","en":"Pathogenicity for 71M missense variants"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"对 7100 万个错义变异给出致病性预测，覆盖大部分未知意义变异","en":"Pathogenicity for 71M missense variants"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"对 7100 万个错义变异给出致病性预测，覆盖大部分未知意义变异","en":"Pathogenicity for 71M missense variants"}},"impactNotes":[],"verified":{"id":"10.1126/science.adg7492","title":"Accurate proteome-wide missense variant effect prediction with AlphaMissense","venue":"Science","date":"2023-09-22"},"addedIn":"v1"},{"date":"2024-11-15","title":{"zh":"Evo：从分子到基因组尺度的序列建模（Science）","en":"Evo: sequence modelling from molecular to genome scale (Science)"},"url":"https://www.science.org/doi/10.1126/science.ado9336","source":"Science","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"首个在分子到基因组尺度生成功能序列的 DNA 语言模型","en":"First DNA LM generating functional sequences"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首个在分子到基因组尺度生成功能序列的 DNA 语言模型","en":"First DNA LM generating functional sequences"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首个在分子到基因组尺度生成功能序列的 DNA 语言模型","en":"First DNA LM generating functional sequences"}},"impactNotes":[],"verified":{"id":"10.1126/science.ado9336","title":"Sequence modeling and design from molecular to genome scale with Evo","venue":"Science","date":"2024-11-15"},"addedIn":"v1"},{"date":"2026-01-28","title":{"zh":"AlphaGenome 发表于 Nature，预测非编码变异效应","en":"AlphaGenome published in Nature"},"url":"https://www.nature.com/articles/s41586-025-10014-0","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"读入 1Mb DNA 统一预测数千种功能轨迹","en":"1 Mb input, thousands of functional tracks"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"读入 1Mb DNA 统一预测数千种功能轨迹","en":"1 Mb input, thousands of functional tracks"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"读入 1Mb DNA 统一预测数千种功能轨迹","en":"1 Mb input, thousands of functional tracks"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-025-10014-0","title":"Advancing regulatory variant effect prediction with AlphaGenome","venue":"Nature","date":"2026-01-28"},"addedIn":"v1"},{"date":"2026-03-04","title":{"zh":"Evo 2：400 亿参数、百万碱基上下文，跨所有生命域建模与设计基因组（Nature）","en":"Evo 2: 40B-parameter genome model with 1 Mb context across all domains of life (Nature)"},"url":"https://doi.org/10.1038/s41586-026-10176-5","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"训练 9.3 万亿碱基，可预测变异效应并生成经实验验证的序列","en":"Trained on 9.3T bases; predicts variant effects and designs validated sequences"},"aiImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"训练 9.3 万亿碱基，可预测变异效应并生成经实验验证的序列","en":"Trained on 9.3T bases; predicts variant effects and designs validated sequences"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"训练 9.3 万亿碱基，可预测变异效应并生成经实验验证的序列","en":"Trained on 9.3T bases; predicts variant effects and designs validated sequences"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-026-10176-5","title":"Genome modelling and design across all domains of life with Evo 2","venue":"Nature","date":"2026-03-04"},"addedIn":"v2"},{"date":"2026-09-09","title":{"zh":"GPN-Star：引入物种树与全基因组比对的基因组语言模型，变异效应预测达到最优（Nature）","en":"GPN-Star: phylogeny-informed genomic language model for genome-wide functional constraints (Nature)"},"url":"https://www.nature.com/articles/s41586-026-11005-5","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"model","typeLabel":{"zh":"模型发布","en":"Model"},"why":{"zh":"AI：以系统发育感知架构让较小的基因组语言模型超过经典进化模型；学科：在致病与 GWAS 精细定位变异排序、复杂性状遗传力富集和罕见变异关联检验上显著提升，并推广到小鼠、果蝇、拟南芥等 5 个物种","en":"AI: a phylogeny-aware design lets a compact gLM beat classical evolutionary models; Field: better pathogenic/GWAS variant prioritisation and rare-variant tests, extended to five model organisms"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：以系统发育感知架构让较小的基因组语言模型超过经典进化模型；学科：在致病与 GWAS 精细定位变异排序、复杂性状遗传力富集和罕见变异关联检验上显著提升，并推广到小鼠、果蝇、拟南芥等 5 个物种","en":"AI: a phylogeny-aware design lets a compact gLM beat classical evolutionary models; Field: better pathogenic/GWAS variant prioritisation and rare-variant tests, extended to five model organisms"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：以系统发育感知架构让较小的基因组语言模型超过经典进化模型；学科：在致病与 GWAS 精细定位变异排序、复杂性状遗传力富集和罕见变异关联检验上显著提升，并推广到小鼠、果蝇、拟南芥等 5 个物种","en":"AI: a phylogeny-aware design lets a compact gLM beat classical evolutionary models; Field: better pathogenic/GWAS variant prioritisation and rare-variant tests, extended to five model organisms"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-026-11005-5","title":"Predicting genome-wide functional constraints with GPN-Star","venue":"Nature","date":"2026-09-09"},"addedIn":"v2"},{"date":"2026-10-21","title":{"zh":"hEVO：在 6.38 万人、47.8 亿条单倍型序列上训练的人类基因组基础模型，19 项主要对比中 18 项排名第一（bioRxiv）","en":"hEVO: a human genomic foundation model trained on 4.784B haplotype sequences from 63,809 genomes, ranking first in 18 of 19 primary comparisons (bioRxiv)"},"url":"https://www.biorxiv.org/content/10.64898/2026.10.19.760276v1","source":"bioRxiv","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok","datePrecision":"day","type":"model","typeLabel":{"zh":"模型发布","en":"Model"},"why":{"zh":"AI：直接用全基因组测序重建的单倍型（而非参考基因组）预训练，把人群内天然等位差异纳入表征；学科：在功能注释、变异解读和人类演化任务上对比 Evo 2、NT、GPN 等模型 18/19 项领先（作者自评，预印本未审稿）","en":"AI: pretrained on WGS-reconstructed haplotypes rather than a reference genome. Field: leads Evo 2, NT, GPN etc. in 18/19 comparisons on annotation, variant interpretation and evolution (self-reported preprint)"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：直接用全基因组测序重建的单倍型（而非参考基因组）预训练，把人群内天然等位差异纳入表征；学科：在功能注释、变异解读和人类演化任务上对比 Evo 2、NT、GPN 等模型 18/19 项领先（作者自评，预印本未审稿）","en":"AI: pretrained on WGS-reconstructed haplotypes rather than a reference genome. Field: leads Evo 2, NT, GPN etc. in 18/19 comparisons on annotation, variant interpretation and evolution (self-reported preprint)"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：直接用全基因组测序重建的单倍型（而非参考基因组）预训练，把人群内天然等位差异纳入表征；学科：在功能注释、变异解读和人类演化任务上对比 Evo 2、NT、GPN 等模型 18/19 项领先（作者自评，预印本未审稿）","en":"AI: pretrained on WGS-reconstructed haplotypes rather than a reference genome. Field: leads Evo 2, NT, GPN etc. in 18/19 comparisons on annotation, variant interpretation and evolution (self-reported preprint)"}},"impactNotes":[],"verified":{"id":"10.64898/2026.10.19.760276","title":"A genomic foundation model for interpreting human genetic diversity","venue":"bioRxiv","date":"2026-10-21"},"addedIn":"v2-20261022pm"}],"papersRepos":[{"title":"google/deepvariant","url":"https://github.com/google/deepvariant","date":null,"kind":"repo","linkCheck":"ok"},{"title":"google-deepmind/alphagenome","url":"https://github.com/google-deepmind/alphagenome","date":null,"kind":"repo","linkCheck":"ok"},{"title":"ArcInstitute/evo2","url":"https://github.com/ArcInstitute/evo2","date":null,"kind":"repo","linkCheck":"ok"},{"title":"Genome modelling and design across all domains of life with Evo 2","url":"https://doi.org/10.1038/s41586-026-10176-5","date":"2026-03-04","kind":"paper","linkCheck":"ok-crossref"},{"title":"Effective gene expression prediction from sequence by integrating long-range interactions","url":"https://doi.org/10.1038/s41592-021-01252-x","date":"2021-10-04","kind":"paper","linkCheck":"ok-crossref"},{"title":"Accurate proteome-wide missense variant effect prediction with AlphaMissense","url":"https://doi.org/10.1126/science.adg7492","date":"2023-09-22","kind":"paper","linkCheck":"ok-crossref"},{"title":"Sequence modeling and design from molecular to genome scale with Evo","url":"https://doi.org/10.1126/science.ado9336","date":"2024-11-15","kind":"paper","linkCheck":"ok-crossref"},{"title":"Advancing regulatory variant effect prediction with AlphaGenome","url":"https://doi.org/10.1038/s41586-025-10014-0","date":"2026-01-28","kind":"paper","linkCheck":"ok-crossref"},{"title":"Predicting genome-wide functional constraints with GPN-Star","url":"https://doi.org/10.1038/s41586-026-11005-5","date":"2026-09-09","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"Google DeepMind","url":"https://deepmind.google"},{"name":"Arc Institute","url":"https://arcinstitute.org"},{"name":"Broad Institute","url":"https://www.broadinstitute.org"}],"news":[{"id":"266096f8043a","date":"2026-10-20","title":"High-risk antimicrobial resistance genomic traits in multidrug-resistant Salmonella across China’s food chain: projections to 2050","url":"https://www.nature.com/articles/s41467-026-78346-7","source":"Nature 学科·机器学习","lang":"en","kind":"news","summaryZh":"Nature 学科·机器学习于2026-10-20发布的文章，涉及AI 科研与调研、AI 生物/医药、AI 安全，原题《High-risk antimicrobial resistance genomic traits in multidrug-resistant Salmonella across》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"5113aa021d9b","date":"2026-10-20","title":"CPTCs Drive Somatic-Visceral Communication via the Wnt Axis in Somatic Mechanotherapy: A Single-Cell Deep Learning Study","url":"https://arxiv.org/abs/2602.05451","source":"arXiv q-bio.BM（生物分子）","lang":"en","kind":"paper","summaryZh":"arXiv q-bio.BM（生物分子）于2026-10-20发布的论文，涉及AI 生物/医药、AI 安全、AI 软件开发，原题《CPTCs Drive Somatic-Visceral Communication via the Wnt Axis in Somatic Mechanotherapy: A S》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"3616f057275c","date":"2026-10-20","title":"RNAGenScape: Property-Guided, Optimized Generation of mRNA Sequences with Manifold Langevin Dynamics","url":"https://arxiv.org/abs/2510.24736","source":"arXiv q-bio.BM（生物分子）","lang":"en","kind":"paper","summaryZh":"arXiv q-bio.BM（生物分子）于2026-10-20发布的论文，涉及AI 生物/医药、AI 软件开发，原题《RNAGenScape: Property-Guided, Optimized Generation of mRNA Sequences with Manifold Langevi》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"d4de2afff9b3","date":"2026-10-20","title":"gpuGEM: GPU-based linear optimisation of genome-scale metabolic models","url":"https://arxiv.org/abs/2610.24106","source":"arXiv q-bio","lang":"en","kind":"paper","summaryZh":"arXiv q-bio于2026-10-20发布的论文，涉及AI 生物/医药、AI 芯片硬件、AI 软件开发，原题《gpuGEM: GPU-based linear optimisation of genome-scale metabolic models》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"6e336da678f0","date":"2026-10-19","title":"ASHG: Long-Read Genomics Meets Clinical Data in PacBio, Verily Health AI Deal","url":"https://www.genengnews.com/topics/omics/ashg-long-read-genomics-meets-clinical-data-in-pacbio-verily-health-ai-deal/","source":"GEN 话题：Artificial Intelligence","lang":"en","kind":"news","summaryZh":"GEN 话题：Artificial Intelligence于2026-10-20发布的文章，涉及AI 生物/医药、AI 医疗、AI 商业融资，原题《ASHG: Long-Read Genomics Meets Clinical Data in PacBio, Verily Health AI Deal》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"8498e5ace732","date":"2026-10-17","title":"Why AI Isn’t Ready to Tackle Our Convoluted Genomes","url":"https://www.wired.com/story/why-ai-isnt-ready-to-tackle-our-convoluted-genomes/","source":"WIRED · AI","lang":"en","kind":"news","summaryZh":"WIRED · AI于2026-10-17发布的文章，涉及AI 生物/医药、AI 商业融资，原题《Why AI Isn’t Ready to Tackle Our Convoluted Genomes》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"d8c8bb65cc6a","date":"2026-10-16","title":"LitGene: an interpretable transformer model integrating text and ontology for gene representation learning","url":"https://www.nature.com/articles/s41467-026-78454-4","source":"Nature 学科·机器学习","lang":"en","kind":"news","summaryZh":"Nature 学科·机器学习于2026-10-16发布的文章，涉及AI 科研与调研、AI 安全，原题《LitGene: an interpretable transformer model integrating text and ontology for gene represe》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"f79fe6c9123a","date":"2026-10-15","title":"Newly Identified Microexons Offer Clues to Autism Spectrum Disorder’s Genetic Roots","url":"https://www.genengnews.com/topics/omics/newly-identified-microexons-offer-clues-to-autism-spectrum-disorders-genetic-roots/","source":"GEN 话题：Artificial Intelligence","lang":"en","kind":"news","summaryZh":"GEN 话题：Artificial Intelligence于2026-10-16发布的文章，涉及AI 生物/医药、AI 法律政策、AI 软件开发，原题《Newly Identified Microexons Offer Clues to Autism Spectrum Disorder’s Genetic Roots》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"996d714d9cdf","date":"2026-10-15","title":"Spatiotemporal AI-based modeling of gene regulation and variant effects in the developing human brain using Seal","url":"https://www.nature.com/articles/s41593-026-02470-2","source":"Nature 学科·机器学习","lang":"en","kind":"news","summaryZh":"Nature 学科·机器学习于2026-10-15发布的文章，涉及AI 科研与调研、AI 法律政策，原题《Spatiotemporal AI-based modeling of gene regulation and variant effects in the developing 》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"6bfedb269278","date":"2026-10-13","title":"India, UK Inaugurate Genomics Centre of Excellence in Delhi, Advancing Biotechnology & Healthcare Research","url":"https://www.digitalhealthnews.com/india-uk-inaugurate-genomics-centre-of-excellence-in-delhi-advancing-biotechnology-healthcare-research","source":"Digital Health News","lang":"en","kind":"news","summaryZh":"Digital Health News于2026-10-13发布的文章，涉及AI 医疗、AI 生物/医药、AI 安全，原题《India, UK Inaugurate Genomics Centre of Excellence in Delhi, Advancing Biotechnology & Hea》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"a287e408fa36","date":"2026-10-09","title":"Unlocking the Regulatory Genome by ARGUS: An Evidence-Constrained Agentic Framework for Interpreting Single Nucleotide Variants","url":"https://arxiv.org/abs/2610.12281","source":"arXiv q-bio.GN","lang":"en","kind":"news","summaryZh":"arXiv：ARGUS 是“证据约束”的 agent 框架，用于解释非编码区单核苷酸变异，减少大模型编造结合位点等幻觉。","via":"research-samples-20261010","linkCheck":"from-feed"},{"id":"d34c9257f044","date":"2026-10-08","title":"Evo2HiC: a multimodal foundation model for integrative analysis of genome sequence and architecture","url":"https://www.biorxiv.org/content/10.1101/2025.11.18.689171v2","source":"bioRxiv (bioinformatics)","lang":"en","kind":"news","summaryZh":"bioRxiv：Evo2HiC 多模态基础模型联合基因组序列与 Hi-C 数据，预测细胞类型特异的三维基因组结构。","via":"research-samples-20261010","linkCheck":"from-feed"}],"review":{"lastReviewed":"2026-10-22","cycleDays":60,"nextReview":"2026-12-20"},"urls":{"zh":"/zh-hans/disciplines/genomics/","en":"/en/disciplines/genomics/"},"dimensions":{"method":{"score":3,"label":{"zh":"研究方法","en":"Research method"},"evidence":"DeepVariant 等已进入测序标准流程；基因组语言模型快速普及","source":"https://doi.org/10.1038/nbt.4235","weight":0.3},"discovery":{"score":3,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"AlphaMissense、AlphaGenome、Evo 2 等多项独立顶刊成果","source":"https://doi.org/10.1038/s41586-026-10176-5","weight":0.35},"adoption":{"score":3,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"临床变异解读开始引用 AlphaMissense 等预测分数","source":null,"weight":0.25},"education":{"score":1,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到课程证据","source":null,"weight":0.1}},"compositeScore":2.8,"levelCapped":false,"levelChange":{"from":3,"to":3,"direction":"same","reason":null},"biggestBreakthrough":{"date":"2026-03-04","title":{"zh":"Evo 2：400 亿参数、百万碱基上下文，跨所有生命域建模与设计基因组（Nature）","en":"Evo 2: 40B-parameter genome model with 1 Mb context across all domains of life (Nature)"},"url":"https://doi.org/10.1038/s41586-026-10176-5","aiImpact":"leap","fieldImpact":"major","why":{"zh":"训练 9.3 万亿碱基，可预测变异效应并生成经实验验证的序列","en":"Trained on 9.3T bases; predicts variant effects and designs validated sequences"}},"keyPapers":[{"title":"Genome modelling and design across all domains of life with Evo 2","titleZh":"Evo 2：400 亿参数、百万碱基上下文，跨所有生命域建模与设计基因组（Nature）","id":"10.1038/s41586-026-10176-5","url":"https://doi.org/10.1038/s41586-026-10176-5","venue":"Nature","date":"2026-03-04","why":{"zh":"训练 9.3 万亿碱基，可预测变异效应并生成经实验验证的序列","en":"Trained on 9.3T bases; predicts variant effects and designs validated sequences"},"fieldImpact":"major","aiImpact":"leap"},{"title":"Effective gene expression prediction from sequence by integrating long-range interactions","titleZh":"Enformer：Transformer 结合长程相互作用预测基因表达（Nature Methods）","id":"10.1038/s41592-021-01252-x","url":"https://doi.org/10.1038/s41592-021-01252-x","venue":"Nature Methods","date":"2021-10-04","why":{"zh":"把序列到功能预测的上下文扩展到 10 万碱基","en":"Extends sequence-to-function context to 100 kb"},"fieldImpact":"major","aiImpact":"major"},{"title":"Accurate proteome-wide missense variant effect prediction with AlphaMissense","titleZh":"AlphaMissense 预测 7100 万个错义变异（Science）","id":"10.1126/science.adg7492","url":"https://doi.org/10.1126/science.adg7492","venue":"Science","date":"2023-09-22","why":{"zh":"对 7100 万个错义变异给出致病性预测，覆盖大部分未知意义变异","en":"Pathogenicity for 71M missense variants"},"fieldImpact":"major","aiImpact":"major"},{"title":"Sequence modeling and design from molecular to genome scale with Evo","titleZh":"Evo：从分子到基因组尺度的序列建模（Science）","id":"10.1126/science.ado9336","url":"https://doi.org/10.1126/science.ado9336","venue":"Science","date":"2024-11-15","why":{"zh":"首个在分子到基因组尺度生成功能序列的 DNA 语言模型","en":"First DNA LM generating functional sequences"},"fieldImpact":"major","aiImpact":"major"},{"title":"Advancing regulatory variant effect prediction with AlphaGenome","titleZh":"AlphaGenome 发表于 Nature，预测非编码变异效应","id":"10.1038/s41586-025-10014-0","url":"https://doi.org/10.1038/s41586-025-10014-0","venue":"Nature","date":"2026-01-28","why":{"zh":"读入 1Mb DNA 统一预测数千种功能轨迹","en":"1 Mb input, thousands of functional tracks"},"fieldImpact":"major","aiImpact":"major"},{"title":"Predicting genome-wide functional constraints with GPN-Star","titleZh":"GPN-Star：引入物种树与全基因组比对的基因组语言模型，变异效应预测达到最优（Nature）","id":"10.1038/s41586-026-11005-5","url":"https://doi.org/10.1038/s41586-026-11005-5","venue":"Nature","date":"2026-09-09","why":{"zh":"AI：以系统发育感知架构让较小的基因组语言模型超过经典进化模型；学科：在致病与 GWAS 精细定位变异排序、复杂性状遗传力富集和罕见变异关联检验上显著提升，并推广到小鼠、果蝇、拟南芥等 5 个物种","en":"AI: a phylogeny-aware design lets a compact gLM beat classical evolutionary models; Field: better pathogenic/GWAS variant prioritisation and rare-variant tests, extended to five model organisms"},"fieldImpact":"major","aiImpact":"major"}],"latest_papers":[{"id":"doi:10.64898/2026.10.19.760276","title":"A genomic foundation model for interpreting human genetic diversity","titleZh":"解读人类遗传多样性的基因组基础模型 hEVO","summaryZh":"用 6.38 万人全基因组测序重建的 47.8 亿条单倍型序列预训练，在功能注释、变异解读和人类演化的 19 项主要对比中 18 项排名第一。","summaryEn":"Pretrained on 4.784B haplotype sequences from 63,809 WGS individuals; ranks first in 18 of 19 primary comparisons.","authors":["Q. Xu","Z. Wu","M. J. Li"],"date":"2026-10-21","url":"https://www.biorxiv.org/content/10.64898/2026.10.19.760276v1","site":"bioRxiv","venue":null,"category":"genomics","fetchedAt":"2026-10-22T15:20:00"},{"id":"arXiv:2610.22617","title":"Harness-Driven Agentic Operation of Optical Pooled Screen Analysis","titleZh":"框架驱动的光学混池筛选分析智能体操作","summaryZh":"提出 Brieflow-auto，用 Claude Code 实现的专用框架让智能体自动运行光学混池筛选的多日分析流程，包括选参数、检查输出和处理故障。","summaryEn":"Brieflow-auto, a Claude Code harness, lets an agent run multi-day optical pooled screen analyses end to end.","authors":["Matteo Di Bernardo","Claire Peterson","Ege Topkoc","Matt Might"],"date":"2026-10-15","url":"https://arxiv.org/abs/2610.22617","site":"arXiv","venue":null,"category":"q-bio.GN","fetchedAt":"2026-10-22T10:06:35"},{"id":"arXiv:2610.21943","title":"Panoramic modelling of single-cell transcriptomes at scale with CellNet","titleZh":"用 CellNet 大规模全景建模单细胞转录组","summaryZh":"提出 CellNet 基础模型，覆盖更多细胞特异转录本和分子关联，以全面刻画细胞多样性。","summaryEn":"CellNet models a broader range of single-cell transcriptomic information to characterise cellular diversity.","authors":["Xiaoqing Lian","Pei Liu","Jiahui Tan","Luoda Tan"],"date":"2026-10-15","url":"https://arxiv.org/abs/2610.21943","site":"arXiv","venue":null,"category":"q-bio.GN","fetchedAt":"2026-10-22T10:06:35"},{"id":"doi:10.64898/2026.10.11.756618","title":"Decoding super-enhancer regulatory grammar and pathogenic gain-of-function variants directly from DNA sequence","titleZh":"直接从 DNA 序列解码超级增强子调控语法与致病功能获得变异","summaryZh":"基于基因组 Transformer Borzoi 构建 Borzoi-SE，不依赖表观组测量即可从序列识别细胞类型特异的超级增强子并解读其语法。","summaryEn":"Borzoi-SE, built on the Borzoi genomic transformer, identifies cell-type super-enhancers and their grammar from sequence alone.","authors":["Liang, Y.","Xu, X.","Mao, Y.","Zheng, Z."],"date":"2026-10-14","url":"https://www.biorxiv.org/content/10.64898/2026.10.11.756618v1","site":"bioRxiv","venue":null,"category":"genomics","fetchedAt":"2026-10-22T10:06:35"},{"id":"arXiv:2610.19029","title":"Deep learning-based virtual perturbation reveals increasing heterogeneity of gene regulatory responses with aging","titleZh":"基于深度学习的虚拟扰动揭示基因调控响应随衰老增加的异质性","summaryZh":"用深度学习框架 DKO 做单细胞虚拟敲除，发现衰老过程中单个细胞的基因调控响应异质性增加。","summaryEn":"Deep-learning virtual knockouts (DKO) show gene-regulatory responses become more heterogeneous with ageing.","authors":["Yongjian He","Julia Hanna Haddad Guez","Xu-Wen Wang","Orr Levy"],"date":"2026-10-13","url":"https://arxiv.org/abs/2610.19029","site":"arXiv","venue":null,"category":"q-bio.MN","fetchedAt":"2026-10-22T10:06:35"},{"id":"doi:10.64898/2026.10.12.758490","title":"BioMod: Biological Module Tokenization for Single-Cell Foundation Modeling","titleZh":"BioMod：面向单细胞基础模型的生物模块分词","summaryZh":"把基因组织成生物学上有意义的模块作为“词元”，代替逐基因表示，以更高效地表示细胞状态。","summaryEn":"Tokenises cells by biological gene modules rather than single genes for single-cell foundation models.","authors":["Meng, L.","Zhu, J.","Han, B.","Zang, Y."],"date":"2026-10-13","url":"https://www.biorxiv.org/content/10.64898/2026.10.12.758490v1","site":"bioRxiv","venue":null,"category":"bioinformatics","fetchedAt":"2026-10-22T10:06:35"},{"id":"arXiv:2610.16346","title":"CellChain: Benchmarking Cross-Scale Evidence Use for Perturbation Reasoning","titleZh":"CellChain：跨尺度证据推理的扰动推理基准","summaryZh":"构建评估大模型在细胞扰动推理中能否把分子机制、细胞响应和功能结果跨尺度联系起来的基准。","summaryEn":"A benchmark for whether LLMs link molecular mechanisms, cellular responses and outcomes when reasoning about perturbations.","authors":["Xue Xia","Yue Wang","Shuizhou Chen","Lang Yu"],"date":"2026-10-12","url":"https://arxiv.org/abs/2610.16346","site":"arXiv","venue":null,"category":"cs.DB","fetchedAt":"2026-10-22T10:06:35"}]},{"id":"neuroscience","slug":"neuroscience","category":"life-medical","categoryName":{"zh":"生命与医学","en":"Life & medical sciences"},"name":{"zh":"神经科学","en":"Neuroscience"},"classification":{"oecdFos":"3.1","moe":"071006 神经生物学"},"level":2,"levelKey":"breakthrough","levelLabel":{"zh":"局部突破","en":"Local breakthroughs"},"trend":"up","reason":{"zh":"AI 帮助完成成年果蝇全脑连接组（FlyWire，Nature 2024）和脑机接口语音解码（Nature 2023）；2026 年连接组基础模型大量出现，但多为预印本或基准提升，尚未改写主要研究范式。","en":"AI enabled the adult-fly whole-brain connectome (FlyWire, Nature 2024) and speech-decoding BCIs (Nature 2023); 2026 saw many connectome foundation models, mostly preprints/benchmarks."},"reasonSources":[{"url":"https://www.nature.com/articles/s41586-024-07558-y","linkCheck":"ok"},{"url":"https://www.nature.com/articles/s41586-023-06377-x","linkCheck":"ok"}],"directions":{"zh":["连接组重建","脑机接口解码","神经数据基础模型"],"en":["Connectomics","BCI decoding","Neural-data foundation models"]},"stats":{"milestones":10,"milestonesSince2025":7,"milestones2026":5,"milestonesLast6Months":4,"news":12,"latestProgress":"2026-10-20","latestMilestone":"2026-10-16","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":0,"leaps":{"ai":0,"field":1},"latestPapers":6,"keyPapers":10},"milestones":[{"date":"2023-08-23","title":{"zh":"高性能语音神经假体，解码 62 词/分（Nature）","en":"High-performance speech neuroprosthesis, 62 words/min (Nature)"},"url":"https://www.nature.com/articles/s41586-023-06377-x","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"语音脑机接口解码速度达 62 词/分，较此前提升约 3 倍","en":"62 words/min speech BCI, ~3× prior"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"语音脑机接口解码速度达 62 词/分，较此前提升约 3 倍","en":"62 words/min speech BCI, ~3× prior"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"语音脑机接口解码速度达 62 词/分，较此前提升约 3 倍","en":"62 words/min speech BCI, ~3× prior"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-023-06377-x","title":"A high-performance speech neuroprosthesis","venue":"Nature","date":"2023-08-23"},"addedIn":"v1"},{"date":"2023-08-23","title":{"zh":"高性能脑机接口：语音神经假体以 78 词/分解码并驱动虚拟化身（Nature，Metzger 等）","en":"High-performance neuroprosthesis for speech decoding and avatar control (Nature)"},"url":"https://doi.org/10.1038/s41586-023-06443-4","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"瘫痪患者通过深度学习解码实现接近自然速度的交流","en":"Near-natural-rate communication for a paralysed patient"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"瘫痪患者通过深度学习解码实现接近自然速度的交流","en":"Near-natural-rate communication for a paralysed patient"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"瘫痪患者通过深度学习解码实现接近自然速度的交流","en":"Near-natural-rate communication for a paralysed patient"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-023-06443-4","title":"A high-performance neuroprosthesis for speech decoding and avatar control","venue":"Nature","date":"2023-08-23"},"addedIn":"v2"},{"date":"2024-10-02","title":{"zh":"FlyWire：成年果蝇全脑连接组（Nature）","en":"FlyWire: whole-brain connectome of the adult fly (Nature)"},"url":"https://www.nature.com/articles/s41586-024-07558-y","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"dataset","typeLabel":{"zh":"数据集","en":"Dataset"},"why":{"zh":"首个成年果蝇全脑连接组（约 14 万神经元），AI 分割加人工校对完成","en":"First whole adult fly brain connectome (~140k neurons)"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首个成年果蝇全脑连接组（约 14 万神经元），AI 分割加人工校对完成","en":"First whole adult fly brain connectome (~140k neurons)"}},"fieldImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"首个成年果蝇全脑连接组（约 14 万神经元），AI 分割加人工校对完成","en":"First whole adult fly brain connectome (~140k neurons)"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-024-07558-y","title":"Neuronal wiring diagram of an adult brain","venue":"Nature","date":"2024-10-02"},"addedIn":"v1"},{"date":"2025-04-09","title":{"zh":"神经活动基础模型可预测新刺激类型下的反应，构建小鼠视觉皮层数字孪生（Nature）","en":"Foundation model of neural activity predicts responses to new stimulus types (Nature)"},"url":"https://doi.org/10.1038/s41586-025-08829-y","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"首个跨小鼠、跨刺激泛化的视觉皮层“数字孪生”","en":"First visual-cortex digital twin generalising across mice and stimuli"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首个跨小鼠、跨刺激泛化的视觉皮层“数字孪生”","en":"First visual-cortex digital twin generalising across mice and stimuli"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首个跨小鼠、跨刺激泛化的视觉皮层“数字孪生”","en":"First visual-cortex digital twin generalising across mice and stimuli"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-025-08829-y","title":"Foundation model of neural activity predicts response to new stimulus types","venue":"Nature","date":"2025-04-09"},"addedIn":"v2"},{"date":"2025-04-09","title":{"zh":"MICrONS：跨多个小鼠视觉皮层区的功能连接组（Nature）","en":"MICrONS functional connectomics across mouse visual cortex (Nature)"},"url":"https://doi.org/10.1038/s41586-025-08790-w","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"AI 自动重建立方毫米级脑组织中的神经元与突触并对齐功能数据","en":"AI reconstructs a cubic millimetre of cortex aligned with function"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI 自动重建立方毫米级脑组织中的神经元与突触并对齐功能数据","en":"AI reconstructs a cubic millimetre of cortex aligned with function"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI 自动重建立方毫米级脑组织中的神经元与突触并对齐功能数据","en":"AI reconstructs a cubic millimetre of cortex aligned with function"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-025-08790-w","title":"Functional connectomics spanning multiple areas of mouse visual cortex","venue":"Nature","date":"2025-04-09"},"addedIn":"v2"},{"date":"2026-02-25","title":{"zh":"用数字孪生分析 V1 感受野的功能二部不变性（Nature Neuroscience）","en":"Functional bipartite invariance in mouse V1 receptive fields via digital twin (Nat. Neurosci.)"},"url":"https://doi.org/10.1038/s41593-026-02213-3","source":"Nature Neuroscience","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"数字孪生开始被用来产生可检验的神经科学假设","en":"Digital twins used to generate testable hypotheses"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"数字孪生开始被用来产生可检验的神经科学假设","en":"Digital twins used to generate testable hypotheses"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"数字孪生开始被用来产生可检验的神经科学假设","en":"Digital twins used to generate testable hypotheses"}},"impactNotes":[],"verified":{"id":"10.1038/s41593-026-02213-3","title":"Functional bipartite invariance in mouse primary visual cortex receptive fields","venue":"Nature Neuroscience","date":"2026-02-25"},"addedIn":"v2"},{"date":"2026-06-30","title":{"zh":"FedFound：联邦学习脑形态连接组基础模型（npj Digital Medicine）","en":"FedFound federated brain-connectome foundation model (npj Digit. Med.)"},"url":"https://doi.org/10.1038/s41746-026-02925-7","source":"npj Digital Medicine","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"联邦学习脑影像基础模型，属方法改进","en":"Federated foundation model; method improvement"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"联邦学习脑影像基础模型，属方法改进","en":"Federated foundation model; method improvement"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"联邦学习脑影像基础模型，属方法改进","en":"Federated foundation model; method improvement"}},"impactNotes":[],"verified":{"id":"10.1038/s41746-026-02925-7","title":"FedFound: a federated foundation model for lifespan brain morphological connectome analysis","venue":"npj Digital Medicine","date":"2026-06-30"},"addedIn":"v1"},{"date":"2026-10-13","title":{"zh":"NiCLIP：在 2.3 万篇文献上训练的对比式语言–脑图模型，从脑激活图预测认知任务与概念（Nature Communications）","en":"NiCLIP: contrastive language–brain-image model predicting cognitive tasks from activation maps (Nat. Commun.)"},"url":"https://www.nature.com/articles/s41467-026-78451-7","source":"Nature Communications","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"model","typeLabel":{"zh":"模型发布","en":"Model"},"why":{"zh":"AI：把 CLIP 式对比学习用于神经影像文献；学科：在群体水平激活图上准确解码认知任务，但个体噪声图上表现有限","en":"AI: CLIP-style learning on the neuroimaging literature; Field: accurate group-level functional decoding, weaker on noisy individual maps"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：把 CLIP 式对比学习用于神经影像文献；学科：在群体水平激活图上准确解码认知任务，但个体噪声图上表现有限","en":"AI: CLIP-style learning on the neuroimaging literature; Field: accurate group-level functional decoding, weaker on noisy individual maps"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：把 CLIP 式对比学习用于神经影像文献；学科：在群体水平激活图上准确解码认知任务，但个体噪声图上表现有限","en":"AI: CLIP-style learning on the neuroimaging literature; Field: accurate group-level functional decoding, weaker on noisy individual maps"}},"impactNotes":[],"verified":{"id":"10.1038/s41467-026-78451-7","title":"NiCLIP: neuroimaging contrastive language-image pretraining model for predicting text from brain activation images","venue":"Nature Communications","date":"2026-10-13"},"addedIn":"v2"},{"date":"2026-10-15","title":{"zh":"Seal：可解释 AI 建模发育中人脑 26 个脑区、30 种细胞、7 个阶段的基因调控与变异效应（Nature Neuroscience）","en":"Seal: spatiotemporal AI modelling of gene regulation and variant effects in the developing human brain (Nat. Neurosci.)"},"url":"https://www.nature.com/articles/s41593-026-02470-2","source":"Nature Neuroscience","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"AI：迁移学习在数据稀缺的瞬时细胞类型上预测表达与变异效应；学科：在自闭症队列中发现胎儿兴奋性神经元的新生调控变异负荷，连接遗传变异与脑发育程序","en":"AI: transfer learning predicts expression/variant effects in data-poor transient cell types; Field: finds de novo regulatory burden in fetal excitatory neurons in autism"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：迁移学习在数据稀缺的瞬时细胞类型上预测表达与变异效应；学科：在自闭症队列中发现胎儿兴奋性神经元的新生调控变异负荷，连接遗传变异与脑发育程序","en":"AI: transfer learning predicts expression/variant effects in data-poor transient cell types; Field: finds de novo regulatory burden in fetal excitatory neurons in autism"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：迁移学习在数据稀缺的瞬时细胞类型上预测表达与变异效应；学科：在自闭症队列中发现胎儿兴奋性神经元的新生调控变异负荷，连接遗传变异与脑发育程序","en":"AI: transfer learning predicts expression/variant effects in data-poor transient cell types; Field: finds de novo regulatory burden in fetal excitatory neurons in autism"}},"impactNotes":[],"verified":{"id":"10.1038/s41593-026-02470-2","title":"Spatiotemporal AI-based modeling of gene regulation and variant effects in the developing human brain using Seal","venue":"Nature Neuroscience","date":"2026-10-15"},"addedIn":"v2"},{"date":"2026-10-16","title":{"zh":"OpenAI 与 PerturbAI、Scripps 用 1946 个 GPT-Rosalind 智能体并行分析小鼠全脑 CRISPR 扰动图谱，32 分钟生成可检验的机制假说（bioRxiv）","en":"OpenAI, PerturbAI and Scripps run 1,946 parallel GPT-Rosalind agents on a whole-brain in vivo CRISPR atlas, producing testable hypotheses in 32 minutes (bioRxiv)"},"url":"https://www.biorxiv.org/content/10.64898/2026.10.15.759436v1","source":"bioRxiv / OpenAI","flags":["preprint","vendor"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"},{"zh":"厂商自述","en":"Vendor claim"}],"linkCheck":"site-blocks-bots","datePrecision":"day","type":"discovery","typeLabel":{"zh":"科学发现","en":"Discovery"},"why":{"zh":"AI：每个智能体负责一个基因扰动，解读细胞反应、检索文献并提出机制，汇总为“推理图谱”，专家评审更偏好多智能体报告（OpenAI 10-21 公布）；学科：提出 DCTN1 与神经元胆固醇代谢等待验证线索，尚无实验验证","en":"AI: one agent per perturbation interprets responses, searches literature and proposes mechanisms; experts preferred the multi-agent reports (announced by OpenAI on 10-21). Field: yields untested leads such as DCTN1 and neuronal cholesterol"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：每个智能体负责一个基因扰动，解读细胞反应、检索文献并提出机制，汇总为“推理图谱”，专家评审更偏好多智能体报告（OpenAI 10-21 公布）；学科：提出 DCTN1 与神经元胆固醇代谢等待验证线索，尚无实验验证","en":"AI: one agent per perturbation interprets responses, searches literature and proposes mechanisms; experts preferred the multi-agent reports (announced by OpenAI on 10-21). Field: yields untested leads such as DCTN1 and neuronal cholesterol"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：每个智能体负责一个基因扰动，解读细胞反应、检索文献并提出机制，汇总为“推理图谱”，专家评审更偏好多智能体报告（OpenAI 10-21 公布）；学科：提出 DCTN1 与神经元胆固醇代谢等待验证线索，尚无实验验证","en":"AI: one agent per perturbation interprets responses, searches literature and proposes mechanisms; experts preferred the multi-agent reports (announced by OpenAI on 10-21). Field: yields untested leads such as DCTN1 and neuronal cholesterol"}},"impactNotes":[],"verified":{"id":"10.64898/2026.10.15.759436","title":"Distributed AI agents enable comprehensive discovery of biological insights at scale","venue":"bioRxiv","date":"2026-10-16"},"addedIn":"v2-20261022pm"}],"papersRepos":[{"title":"ConnectoFM (bioRxiv 2026)","url":"https://doi.org/10.64898/2026.06.06.730367","date":"2026-06-10","kind":"paper","linkCheck":"ok-crossref"},{"title":"Neuronal wiring diagram of an adult brain","url":"https://doi.org/10.1038/s41586-024-07558-y","date":"2024-10-02","kind":"paper","linkCheck":"ok-crossref"},{"title":"A high-performance speech neuroprosthesis","url":"https://doi.org/10.1038/s41586-023-06377-x","date":"2023-08-23","kind":"paper","linkCheck":"ok-crossref"},{"title":"A high-performance neuroprosthesis for speech decoding and avatar control","url":"https://doi.org/10.1038/s41586-023-06443-4","date":"2023-08-23","kind":"paper","linkCheck":"ok-crossref"},{"title":"Foundation model of neural activity predicts response to new stimulus types","url":"https://doi.org/10.1038/s41586-025-08829-y","date":"2025-04-09","kind":"paper","linkCheck":"ok-crossref"},{"title":"Functional connectomics spanning multiple areas of mouse visual cortex","url":"https://doi.org/10.1038/s41586-025-08790-w","date":"2025-04-09","kind":"paper","linkCheck":"ok-crossref"},{"title":"Spatiotemporal AI-based modeling of gene regulation and variant effects in the developing human brain using Seal","url":"https://doi.org/10.1038/s41593-026-02470-2","date":"2026-10-15","kind":"paper","linkCheck":"ok-crossref"},{"title":"Functional bipartite invariance in mouse primary visual cortex receptive fields","url":"https://doi.org/10.1038/s41593-026-02213-3","date":"2026-02-25","kind":"paper","linkCheck":"ok-crossref"},{"title":"FedFound: a federated foundation model for lifespan brain morphological connectome analysis","url":"https://doi.org/10.1038/s41746-026-02925-7","date":"2026-06-30","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"FlyWire / Princeton","url":"https://flywire.ai"},{"name":"Allen Institute","url":"https://alleninstitute.org"}],"news":[{"id":"24b854638327","date":"2026-10-20","title":"EEG-based Alzheimer’s disease detection using two-branch multi-scale deep learning with integrated gradients-guided channel selection","url":"https://www.nature.com/articles/s41598-026-76382-3","source":"Nature 学科·计算生物","lang":"en","kind":"news","summaryZh":"Nature 学科·计算生物于2026-10-20发布的文章，涉及AI 生物/医药，原题《EEG-based Alzheimer’s disease detection using two-branch multi-scale deep learning with in》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"1dd5c007c495","date":"2026-10-20","title":"Nanoscale tech tracks proteins inside living neurons to guide Alzheimer's and Parkinson's drug discovery","url":"https://phys.org/news/2026-10-nanoscale-tech-tracks-proteins-neurons.html","source":"Phys.org 化学·材料(纳米)","lang":"en","kind":"news","summaryZh":"Phys.org 化学·材料(纳米)于2026-10-20发布的文章，涉及AI 化学材料、AI 生物/医药、AI 教育，原题《Nanoscale tech tracks proteins inside living neurons to guide Alzheimer's and Parkinson's 》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"f79af707eb63","date":"2026-10-20","title":"顺为投了一位北大PI，想用AI“解密”人脑","url":"https://36kr.com/p/4032248929686151?f=rss","source":"36氪","lang":"zh","kind":"news","summaryZh":"36氪于2026-10-20发布的文章，涉及AI 商业融资、AI 生物/医药、AI 科研与调研，原题《顺为投了一位北大PI，想用AI“解密”人脑》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"19854dbc34ce","date":"2026-10-20","title":"Spontaneous brain-wide rotational dynamics constrain cue-evoked behavior","url":"https://arxiv.org/abs/2610.25061","source":"arXiv q-bio","lang":"en","kind":"paper","summaryZh":"arXiv q-bio于2026-10-20发布的论文，涉及AI 生物/医药，原题《Spontaneous brain-wide rotational dynamics constrain cue-evoked behavior》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"a4f43a5a2b21","date":"2026-10-20","title":"Brain-OF: An Omnifunctional Foundation Model for fMRI, EEG and MEG","url":"https://arxiv.org/abs/2602.23410","source":"arXiv q-bio","lang":"en","kind":"paper","summaryZh":"arXiv q-bio于2026-10-20发布的论文，涉及AI 生物/医药、AI 软件开发，原题《Brain-OF: An Omnifunctional Foundation Model for fMRI, EEG and MEG》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"65d8539fcd23","date":"2026-10-16","title":"Mini brain organoids 'read' braille in key step toward biohybrid computing","url":"https://techxplore.com/news/2026-10-mini-brain-organoids-braille-key.html","source":"Phys.org 机器学习与AI","lang":"en","kind":"news","summaryZh":"Phys.org 机器学习与AI于2026-10-16发布的文章，涉及AI 科研与调研、AI 气候能源，原题《Mini brain organoids 'read' braille in key step toward biohybrid computing》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"996d714d9cdf","date":"2026-10-15","title":"Spatiotemporal AI-based modeling of gene regulation and variant effects in the developing human brain using Seal","url":"https://www.nature.com/articles/s41593-026-02470-2","source":"Nature 学科·机器学习","lang":"en","kind":"news","summaryZh":"Nature 学科·机器学习于2026-10-15发布的文章，涉及AI 科研与调研、AI 法律政策，原题《Spatiotemporal AI-based modeling of gene regulation and variant effects in the developing 》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"bdb5f9d6d099","date":"2026-10-15","title":"Q&A: The “collision” between AI, your brain, and society","url":"https://news.mit.edu/2026/collision-between-ai-your-brain-and-society-eric-so-book-1015","source":"MIT News：Artificial intelligence","lang":"en","kind":"news","summaryZh":"MIT News：Artificial intelligence于2026-10-15发布的文章，涉及AI 科研与调研，原题《Q&A: The “collision” between AI, your brain, and society》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"bf9d41780fd4","date":"2026-10-13","title":"NiCLIP: neuroimaging contrastive language-image pretraining model for predicting text from brain activation images","url":"https://www.nature.com/articles/s41467-026-78451-7","source":"Nature 学科·机器学习","lang":"en","kind":"news","summaryZh":"Nature 学科·机器学习于2026-10-13发布的文章，涉及AI 科研与调研，原题《NiCLIP: neuroimaging contrastive language-image pretraining model for predicting text from》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"536019a6ba61","date":"2026-10-07","title":"Bridge Neurotech Partners with Butterfly Network to Develop Non-Invasive Brain-Computer Interface","url":"https://www.digitalhealthnews.com/bridge-neurotech-partners-with-butterfly-network-to-develop-non-invasive-brain-computer-interface","source":"Digital Health News","lang":"en","kind":"news","summaryZh":"Digital Health News于2026-10-07发布的文章，涉及AI 医疗、AI 金融、AI 芯片硬件，原题《Bridge Neurotech Partners with Butterfly Network to Develop Non-Invasive Brain-Computer In》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"2e78c3d42d00","date":"2026-09-28","title":"Could soil process information without a brain? New engineering perspective makes the case","url":"https://techxplore.com/news/2026-09-soil-brain-perspective-case.html","source":"TechXplore 机器人","lang":"en","kind":"news","summaryZh":"TechXplore 机器人于2026-09-28发布的文章，涉及AI 机器人、AI 生物/医药、AI 化学材料，原题《Could soil process information without a brain? New engineering perspective makes the case》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"c056c2e41db1","date":"2026-09-21","title":"Import AI 473: The US's superintelligence strategy; human brain in a mouse skull; and machine hermeneutics","url":"https://importai.substack.com/p/import-ai-473-the-uss-superintelligence","source":"Import AI (Jack Clark)","lang":"en","kind":"news","summaryZh":"Import AI (Jack Clark)于2026-09-21发布的文章，涉及AI 综合，原题《Import AI 473: The US's superintelligence strategy; human brain in a mouse skull; and mach》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"}],"review":{"lastReviewed":"2026-10-22","cycleDays":90,"nextReview":"2027-01-19"},"urls":{"zh":"/zh-hans/disciplines/neuroscience/","en":"/en/disciplines/neuroscience/"},"dimensions":{"method":{"score":3,"label":{"zh":"研究方法","en":"Research method"},"evidence":"连接组重建、行为追踪（DeepLabCut）、神经解码普遍依赖深度学习","source":"https://doi.org/10.1038/s41586-024-07558-y","weight":0.3},"discovery":{"score":2,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"脑机接口语音解码、小鼠视觉皮层数字孪生（Nature 2025）属子领域突破","source":"https://doi.org/10.1038/s41586-025-08829-y","weight":0.35},"adoption":{"score":2,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"实验室工具层面普遍，临床 BCI 仍是试验","source":null,"weight":0.25},"education":{"score":1,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到课程证据","source":null,"weight":0.1}},"compositeScore":2.2,"levelCapped":false,"levelChange":{"from":2,"to":2,"direction":"same","reason":null},"biggestBreakthrough":{"date":"2024-10-02","title":{"zh":"FlyWire：成年果蝇全脑连接组（Nature）","en":"FlyWire: whole-brain connectome of the adult fly (Nature)"},"url":"https://www.nature.com/articles/s41586-024-07558-y","aiImpact":"major","fieldImpact":"leap","why":{"zh":"首个成年果蝇全脑连接组（约 14 万神经元），AI 分割加人工校对完成","en":"First whole adult fly brain connectome (~140k neurons)"}},"keyPapers":[{"title":"Neuronal wiring diagram of an adult brain","titleZh":"FlyWire：成年果蝇全脑连接组（Nature）","id":"10.1038/s41586-024-07558-y","url":"https://doi.org/10.1038/s41586-024-07558-y","venue":"Nature","date":"2024-10-02","why":{"zh":"首个成年果蝇全脑连接组（约 14 万神经元），AI 分割加人工校对完成","en":"First whole adult fly brain connectome (~140k neurons)"},"fieldImpact":"leap","aiImpact":"major"},{"title":"A high-performance speech neuroprosthesis","titleZh":"高性能语音神经假体，解码 62 词/分（Nature）","id":"10.1038/s41586-023-06377-x","url":"https://doi.org/10.1038/s41586-023-06377-x","venue":"Nature","date":"2023-08-23","why":{"zh":"语音脑机接口解码速度达 62 词/分，较此前提升约 3 倍","en":"62 words/min speech BCI, ~3× prior"},"fieldImpact":"major","aiImpact":"major"},{"title":"A high-performance neuroprosthesis for speech decoding and avatar control","titleZh":"高性能脑机接口：语音神经假体以 78 词/分解码并驱动虚拟化身（Nature，Metzger 等）","id":"10.1038/s41586-023-06443-4","url":"https://doi.org/10.1038/s41586-023-06443-4","venue":"Nature","date":"2023-08-23","why":{"zh":"瘫痪患者通过深度学习解码实现接近自然速度的交流","en":"Near-natural-rate communication for a paralysed patient"},"fieldImpact":"major","aiImpact":"major"},{"title":"Foundation model of neural activity predicts response to new stimulus types","titleZh":"神经活动基础模型可预测新刺激类型下的反应，构建小鼠视觉皮层数字孪生（Nature）","id":"10.1038/s41586-025-08829-y","url":"https://doi.org/10.1038/s41586-025-08829-y","venue":"Nature","date":"2025-04-09","why":{"zh":"首个跨小鼠、跨刺激泛化的视觉皮层“数字孪生”","en":"First visual-cortex digital twin generalising across mice and stimuli"},"fieldImpact":"major","aiImpact":"major"},{"title":"Functional connectomics spanning multiple areas of mouse visual cortex","titleZh":"MICrONS：跨多个小鼠视觉皮层区的功能连接组（Nature）","id":"10.1038/s41586-025-08790-w","url":"https://doi.org/10.1038/s41586-025-08790-w","venue":"Nature","date":"2025-04-09","why":{"zh":"AI 自动重建立方毫米级脑组织中的神经元与突触并对齐功能数据","en":"AI reconstructs a cubic millimetre of cortex aligned with function"},"fieldImpact":"major","aiImpact":"major"},{"title":"Spatiotemporal AI-based modeling of gene regulation and variant effects in the developing human brain using Seal","titleZh":"Seal：可解释 AI 建模发育中人脑 26 个脑区、30 种细胞、7 个阶段的基因调控与变异效应（Nature Neuroscience）","id":"10.1038/s41593-026-02470-2","url":"https://doi.org/10.1038/s41593-026-02470-2","venue":"Nature Neuroscience","date":"2026-10-15","why":{"zh":"AI：迁移学习在数据稀缺的瞬时细胞类型上预测表达与变异效应；学科：在自闭症队列中发现胎儿兴奋性神经元的新生调控变异负荷，连接遗传变异与脑发育程序","en":"AI: transfer learning predicts expression/variant effects in data-poor transient cell types; Field: finds de novo regulatory burden in fetal excitatory neurons in autism"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Functional bipartite invariance in mouse primary visual cortex receptive fields","titleZh":"用数字孪生分析 V1 感受野的功能二部不变性（Nature Neuroscience）","id":"10.1038/s41593-026-02213-3","url":"https://doi.org/10.1038/s41593-026-02213-3","venue":"Nature Neuroscience","date":"2026-02-25","why":{"zh":"数字孪生开始被用来产生可检验的神经科学假设","en":"Digital twins used to generate testable hypotheses"},"fieldImpact":"incremental","aiImpact":"incremental"},{"title":"ConnectoFM: A Foundation Model for Learning the Language of the Connectome","titleZh":"ConnectoFM (bioRxiv 2026)","id":"10.64898/2026.06.06.730367","url":"https://doi.org/10.64898/2026.06.06.730367","venue":"","date":"2026-06-10","why":null,"fieldImpact":null,"aiImpact":null},{"title":"FedFound: a federated foundation model for lifespan brain morphological connectome analysis","titleZh":"FedFound：联邦学习脑形态连接组基础模型（npj Digital Medicine）","id":"10.1038/s41746-026-02925-7","url":"https://doi.org/10.1038/s41746-026-02925-7","venue":"npj Digital Medicine","date":"2026-06-30","why":{"zh":"联邦学习脑影像基础模型，属方法改进","en":"Federated foundation model; method improvement"},"fieldImpact":"incremental","aiImpact":"incremental"},{"title":"NiCLIP: neuroimaging contrastive language-image pretraining model for predicting text from brain activation images","titleZh":"NiCLIP：在 2.3 万篇文献上训练的对比式语言–脑图模型，从脑激活图预测认知任务与概念（Nature Communications）","id":"10.1038/s41467-026-78451-7","url":"https://doi.org/10.1038/s41467-026-78451-7","venue":"Nature Communications","date":"2026-10-13","why":{"zh":"AI：把 CLIP 式对比学习用于神经影像文献；学科：在群体水平激活图上准确解码认知任务，但个体噪声图上表现有限","en":"AI: CLIP-style learning on the neuroimaging literature; Field: accurate group-level functional decoding, weaker on noisy individual maps"},"fieldImpact":"incremental","aiImpact":"incremental"}],"latest_papers":[{"id":"arXiv:2610.23638","title":"Deep LLM layers expose passage-level context in EEG during silent reading","titleZh":"大模型深层揭示默读时脑电中的段落级语境","summaryZh":"通过把 19 通道脑电与大模型各层隐藏状态对齐，发现深层表示能从默读脑电中解码出段落级语境信息。","summaryEn":"Aligning 19-channel EEG with LLM layers shows deep layers expose passage-level context during silent reading.","authors":["Ingo Marquardt","Anthilia Alchanat","Priyanka Jain"],"date":"2026-10-16","url":"https://arxiv.org/abs/2610.23638","site":"arXiv","venue":null,"category":"cs.LG","fetchedAt":"2026-10-22T10:06:50"},{"id":"arXiv:2610.22961","title":"What Shapes Brain Alignment of Large Language Models during Dialogue? Insights from Language Production and Comprehension","titleZh":"对话中大模型与大脑对齐由什么决定？来自语言产出与理解的证据","summaryZh":"利用对话时采集的 fMRI 数据，研究大模型在语言产出和理解两种情形下与人脑表征的相似性。","summaryEn":"Uses dialogue fMRI to study LLM–brain alignment in both language production and comprehension.","authors":["Yuko Nakagi","Yuya Chiba","Yasue Kishino"],"date":"2026-10-16","url":"https://arxiv.org/abs/2610.22961","site":"arXiv","venue":null,"category":"cs.CL","fetchedAt":"2026-10-22T10:06:50"},{"id":"doi:10.64898/2025.12.10.693611","title":"Systematic comparison of color representations between humans and deep neural networks: towards predicting human color perception in a vast color space","titleZh":"系统比较人类与深度神经网络的颜色表征","summaryZh":"在大范围颜色空间中对比人类颜色知觉与深度网络的表征结构，探索用网络预测人类颜色感知。","summaryEn":"Compares colour representations of humans and DNNs across a vast colour space toward predicting human colour perception.","authors":["Wickramanayaka, N. R.","Oizumi, M."],"date":"2026-10-16","url":"https://www.biorxiv.org/content/10.64898/2025.12.10.693611v5","site":"bioRxiv","venue":null,"category":"neuroscience","fetchedAt":"2026-10-22T10:06:50"},{"id":"arXiv:2610.22706","title":"Brain-Aligned Multimodal LLMs for Social and Emotional Understanding of Cartoon Movies","titleZh":"与大脑对齐的多模态大模型理解卡通电影中的社会与情绪","summaryZh":"比较扩散式与自回归多模态大模型对卡通电影心理理论和情绪效价的标注，与人类评分和脑活动预测的一致性。","summaryEn":"Compares multimodal LLM annotations of theory of mind and affect in cartoons against human ratings and neural data.","authors":["Yiran Huang","Zeyong Zhang","Bharat Biswal","Xin Di"],"date":"2026-10-15","url":"https://arxiv.org/abs/2610.22706","site":"arXiv","venue":null,"category":"cs.AI","fetchedAt":"2026-10-22T10:06:50"},{"id":"arXiv:2610.21176","title":"Language Models Share a Readout Bottleneck in Working Memory","titleZh":"语言模型在工作记忆中共享读出瓶颈","summaryZh":"大模型在 N-back 任务上重现人类工作记忆特征，研究探讨这种行为相似是否源自相同机制。","summaryEn":"LLMs reproduce human N-back working-memory signatures; the study tests whether a shared readout mechanism explains this.","authors":["Dongyu Gong"],"date":"2026-10-15","url":"https://arxiv.org/abs/2610.21176","site":"arXiv","venue":null,"category":"cs.LG","fetchedAt":"2026-10-22T10:06:50"},{"id":"doi:10.64898/2026.10.09.757963","title":"Predictive Grid Cell Coding Does Not Require Explicit Planning","titleZh":"预测性网格细胞编码不需要显式规划","summaryZh":"对内嗅皮层中“预测性网格细胞”建模，表明其编码不需要显式的目标导向规划也能出现。","summaryEn":"Models show predictive grid-cell coding in entorhinal cortex can arise without explicit planning.","authors":["Sinha, A.","Fasehun, O. C.","Redman, W. T."],"date":"2026-10-15","url":"https://www.biorxiv.org/content/10.64898/2026.10.09.757963v1","site":"bioRxiv","venue":null,"category":"neuroscience","fetchedAt":"2026-10-22T10:06:50"}]},{"id":"pharmacy","slug":"pharmacy","category":"life-medical","categoryName":{"zh":"生命与医学","en":"Life & medical sciences"},"name":{"zh":"药学与药物研发","en":"Pharmacy & drug discovery"},"classification":{"oecdFos":"3.1","moe":"1007 药学"},"level":2,"levelKey":"breakthrough","levelLabel":{"zh":"局部突破","en":"Local breakthroughs"},"trend":"up","reason":{"zh":"AI 发现/设计的药物已进入后期临床：Insilico 的 rentosertib 2a 期结果发表于 Nature Medicine（2025）并已开展 3 期；武田 zasocitinib 3 期头对头胜过 Sotyktu（媒体称 AI 设计）。截至 2026-10 尚无“AI 设计药”获批上市。","en":"AI-discovered drugs are in late-stage trials: Insilico's rentosertib Phase 2a in Nature Medicine (2025), now in Phase 3; Takeda's zasocitinib beat Sotyktu head-to-head in Phase 3. No “AI-designed” drug approved as of Oct 2026."},"reasonSources":[{"url":"https://www.nature.com/articles/s41591-025-03743-2","linkCheck":"ok"},{"url":"https://www.biospace.com/drug-development/takedas-ai-designed-pill-bests-bms-sotyktu-in-head-to-head-psoriasis-trial","linkCheck":"ok"},{"url":"https://www.stateof.ai/2026-explore/answers/ai-drugs-phase-3","linkCheck":"ok"}],"directions":{"zh":["靶点发现","生成式分子设计","结构与亲和力预测","临床试验优化"],"en":["Target discovery","Generative molecule design","Structure & affinity prediction","Trial optimisation"]},"stats":{"milestones":10,"milestonesSince2025":6,"milestones2026":5,"milestonesLast6Months":4,"news":12,"latestProgress":"2026-10-20","latestMilestone":"2026-10-20","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":1,"leaps":{"ai":0,"field":0},"latestPapers":7,"keyPapers":4},"milestones":[{"date":"2020-02-20","title":{"zh":"深度学习发现新抗生素 halicin（Cell）","en":"Deep learning discovers antibiotic halicin (Cell)"},"url":"https://www.cell.com/cell/fulltext/S0092-8674(20)30102-1","source":"Cell","flags":[],"flagLabels":[],"linkCheck":"site-blocks-bots","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"深度学习筛选 1 亿分子发现结构全新的抗生素 halicin","en":"Deep learning screens 100M molecules and finds halicin"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"深度学习筛选 1 亿分子发现结构全新的抗生素 halicin","en":"Deep learning screens 100M molecules and finds halicin"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"深度学习筛选 1 亿分子发现结构全新的抗生素 halicin","en":"Deep learning screens 100M molecules and finds halicin"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2023-05-25","title":{"zh":"深度学习发现窄谱抗生素 abaucin，靶向鲍曼不动杆菌（Nat. Chem. Biol.）","en":"Deep learning discovers abaucin, a narrow-spectrum antibiotic against A. baumannii"},"url":"https://doi.org/10.1038/s41589-023-01349-8","source":"Nature Chemical Biology","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"AI 针对 WHO 头号耐药病原体找到新化合物","en":"AI finds a compound for a WHO priority pathogen"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI 针对 WHO 头号耐药病原体找到新化合物","en":"AI finds a compound for a WHO priority pathogen"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI 针对 WHO 头号耐药病原体找到新化合物","en":"AI finds a compound for a WHO priority pathogen"}},"impactNotes":[],"verified":{"id":"10.1038/s41589-023-01349-8","title":"Deep learning-guided discovery of an antibiotic targeting Acinetobacter baumannii","venue":"Nature Chemical Biology","date":"2023-05-25"},"addedIn":"v2"},{"date":"2023-12-20","title":{"zh":"可解释深度学习发现一类新的结构类别抗生素（Nature）","en":"Explainable deep learning discovers a new structural class of antibiotics (Nature)"},"url":"https://doi.org/10.1038/s41586-023-06887-8","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"发现对 MRSA 有效的新结构类别，并解释模型依据的子结构","en":"New structural class active against MRSA, with explained substructures"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"发现对 MRSA 有效的新结构类别，并解释模型依据的子结构","en":"New structural class active against MRSA, with explained substructures"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"发现对 MRSA 有效的新结构类别，并解释模型依据的子结构","en":"New structural class active against MRSA, with explained substructures"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-023-06887-8","title":"Discovery of a structural class of antibiotics with explainable deep learning","venue":"Nature","date":"2023-12-20"},"addedIn":"v2"},{"date":"2024-03-08","title":{"zh":"Insilico：AI 发现靶点 + 生成式设计的 TNIK 抑制剂从发现到临床全过程（Nature Biotechnology）","en":"Insilico TNIK inhibitor from AI target discovery to clinic (Nature Biotechnology)"},"url":"https://doi.org/10.1038/s41587-024-02143-0","source":"Nature Biotechnology","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"首个公开发表全流程的“AI 靶点 + AI 分子”药物","en":"First published end-to-end AI-target + AI-molecule drug"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首个公开发表全流程的“AI 靶点 + AI 分子”药物","en":"First published end-to-end AI-target + AI-molecule drug"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首个公开发表全流程的“AI 靶点 + AI 分子”药物","en":"First published end-to-end AI-target + AI-molecule drug"}},"impactNotes":[],"verified":{"id":"10.1038/s41587-024-02143-0","title":"A small-molecule TNIK inhibitor targets fibrosis in preclinical and clinical models","venue":"Nature Biotechnology","date":"2024-03-08"},"addedIn":"v2"},{"date":"2025-06-03","title":{"zh":"rentosertib（TNIK 抑制剂）2a 期结果发表（Nature Medicine）","en":"Rentosertib (TNIK inhibitor) Phase 2a published (Nature Medicine)"},"url":"https://www.nature.com/articles/s41591-025-03743-2","source":"Nature Medicine","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"clinical","typeLabel":{"zh":"临床","en":"Clinical"},"why":{"zh":"AI 设计药物首个随机对照 2a 期结果在顶刊发表","en":"First randomised Phase 2a of an AI-designed drug in a top journal"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI 设计药物首个随机对照 2a 期结果在顶刊发表","en":"First randomised Phase 2a of an AI-designed drug in a top journal"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI 设计药物首个随机对照 2a 期结果在顶刊发表","en":"First randomised Phase 2a of an AI-designed drug in a top journal"}},"impactNotes":[],"verified":{"id":"10.1038/s41591-025-03743-2","title":"A generative AI-discovered TNIK inhibitor for idiopathic pulmonary fibrosis: a randomized phase 2a trial","venue":"Nature Medicine","date":"2025-06-03"},"addedIn":"v1"},{"date":"2026","title":{"zh":"State of AI 2026：盘点进入 3 期的 AI 辅助药物","en":"State of AI 2026: AI-assisted medicines in Phase 3"},"url":"https://www.stateof.ai/2026-explore/answers/ai-drugs-phase-3","source":"State of AI Report","flags":["tbv"],"flagLabels":[{"zh":"日期或细节待核实","en":"Date/details to be verified"}],"linkCheck":"ok","datePrecision":"year","type":"industry","typeLabel":{"zh":"产业/采用","en":"Industry"},"why":{"zh":"行业盘点，非一手结果","en":"Industry roundup"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"行业盘点，非一手结果","en":"Industry roundup"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"行业盘点，非一手结果","en":"Industry roundup"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-06-11","title":{"zh":"武田 zasocitinib 3 期头对头优于 BMS Sotyktu","en":"Takeda zasocitinib beats BMS Sotyktu in Phase 3 head-to-head"},"url":"https://www.biospace.com/drug-development/takedas-ai-designed-pill-bests-bms-sotyktu-in-head-to-head-psoriasis-trial","source":"BioSpace","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"clinical","typeLabel":{"zh":"临床","en":"Clinical"},"why":{"zh":"3 期头对头胜过已上市药物；“AI 设计”是媒体说法","en":"Phase 3 head-to-head win; “AI-designed” is a media label"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"3 期头对头胜过已上市药物；“AI 设计”是媒体说法","en":"Phase 3 head-to-head win; “AI-designed” is a media label"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"3 期头对头胜过已上市药物；“AI 设计”是媒体说法","en":"Phase 3 head-to-head win; “AI-designed” is a media label"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-07-07","title":{"zh":"Insilico 启动 rentosertib 3 期临床（IPF，320 例，52 周）","en":"Insilico starts Phase 3 of rentosertib (IPF, 320 patients, 52 weeks)"},"url":"https://insilico.com/news/xmjsn4l091-insilico-initiates-phase-iii-clinical-tr","source":"Insilico Medicine","flags":["vendor"],"flagLabels":[{"zh":"厂商自述","en":"Vendor claim"}],"linkCheck":"ok","datePrecision":"day","type":"clinical","typeLabel":{"zh":"临床","en":"Clinical"},"why":{"zh":"首个“AI 发现靶点 + 生成式 AI 设计分子”的药物进入 3 期","en":"First AI-target, AI-designed molecule in Phase 3"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首个“AI 发现靶点 + 生成式 AI 设计分子”的药物进入 3 期","en":"First AI-target, AI-designed molecule in Phase 3"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首个“AI 发现靶点 + 生成式 AI 设计分子”的药物进入 3 期","en":"First AI-target, AI-designed molecule in Phase 3"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-09-14","title":{"zh":"FDA 优先审评受理武田 zasocitinib 新药申请，PDUFA 目标日 2027 年一季度","en":"FDA accepts zasocitinib NDA under priority review (PDUFA Q1 2027)"},"url":"https://www.takeda.com/newsroom/newsreleases/2026/fda-priority-review-zasocitinib-psoriasis/","source":"Takeda","flags":["contested"],"flagLabels":[{"zh":"有争议","en":"Contested"}],"linkCheck":"ok","datePrecision":"day","type":"regulatory","typeLabel":{"zh":"监管","en":"Regulatory"},"why":{"zh":"可能成为首个获批的计算设计药物；但其设计主要靠物理模拟，“AI 设计”说法有争议","en":"Could be the first approved computationally designed drug; “AI-designed” label disputed"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"可能成为首个获批的计算设计药物；但其设计主要靠物理模拟，“AI 设计”说法有争议","en":"Could be the first approved computationally designed drug; “AI-designed” label disputed"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"可能成为首个获批的计算设计药物；但其设计主要靠物理模拟，“AI 设计”说法有争议","en":"Could be the first approved computationally designed drug; “AI-designed” label disputed"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-10-20","title":{"zh":"Radical Numerics：Omnii-1 设计的噬菌体在同情用药途径下首次用于耐药大肠杆菌感染患者","en":"Radical Numerics: Omnii-1-designed phages administered to a patient under compassionate use"},"url":"https://www.radicalnumerics.ai/blog/omnii-phage","source":"Radical Numerics","flags":["vendor"],"flagLabels":[{"zh":"厂商自述","en":"Vendor claim"}],"linkCheck":"ok","datePrecision":"day","type":"clinical","typeLabel":{"zh":"临床","en":"Clinical"},"why":{"zh":"AI：基因组语言模型设计的 T7 噬菌体变体首次进入人体（实验室中清除了患者分离株）；学科：个性化噬菌体疗法的新路径，但单例、未披露疗效，尚无同行评审","en":"AI: genome-LM-designed T7 phage variants reach a human for the first time; Field: a new route to personalised phage therapy, single case, outcome undisclosed"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：基因组语言模型设计的 T7 噬菌体变体首次进入人体（实验室中清除了患者分离株）；学科：个性化噬菌体疗法的新路径，但单例、未披露疗效，尚无同行评审","en":"AI: genome-LM-designed T7 phage variants reach a human for the first time; Field: a new route to personalised phage therapy, single case, outcome undisclosed"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：基因组语言模型设计的 T7 噬菌体变体首次进入人体（实验室中清除了患者分离株）；学科：个性化噬菌体疗法的新路径，但单例、未披露疗效，尚无同行评审","en":"AI: genome-LM-designed T7 phage variants reach a human for the first time; Field: a new route to personalised phage therapy, single case, outcome undisclosed"}},"impactNotes":[],"verified":null,"addedIn":"v2"}],"papersRepos":[{"title":"Deep learning-guided discovery of an antibiotic targeting Acinetobacter baumannii","url":"https://doi.org/10.1038/s41589-023-01349-8","date":"2023-05-25","kind":"paper","linkCheck":"ok-crossref"},{"title":"Discovery of a structural class of antibiotics with explainable deep learning","url":"https://doi.org/10.1038/s41586-023-06887-8","date":"2023-12-20","kind":"paper","linkCheck":"ok-crossref"},{"title":"A small-molecule TNIK inhibitor targets fibrosis in preclinical and clinical models","url":"https://doi.org/10.1038/s41587-024-02143-0","date":"2024-03-08","kind":"paper","linkCheck":"ok-crossref"},{"title":"A generative AI-discovered TNIK inhibitor for idiopathic pulmonary fibrosis: a randomized phase 2a trial","url":"https://doi.org/10.1038/s41591-025-03743-2","date":"2025-06-03","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"Insilico Medicine","url":"https://insilico.com"},{"name":"Isomorphic Labs","url":"https://www.isomorphiclabs.com"},{"name":"Recursion","url":"https://www.recursion.com"},{"name":"Takeda","url":"https://www.takeda.com"}],"news":[{"id":"16d59cdddbc5","date":"2026-10-20","title":"AZ and Oracle team up on AI platform for clinical trials","url":"https://pharmaphorum.com/news/az-and-oracle-team-ai-platform-clinical-trials","source":"pharmaphorum","lang":"en","kind":"news","summaryZh":"pharmaphorum于2026-10-20发布的文章，涉及AI 生物/医药、AI 医疗，原题《AZ and Oracle team up on AI platform for clinical trials》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"2e9ce6273cd6","date":"2026-10-20","title":"STAT+: Pharmalittle: We’re reading about drug launch prices, an AstraZeneca AI deal, and more","url":"https://www.statnews.com/pharmalot/2026/10/20/drug-launch-prices-rose-astrazeneca-ai-deal/?utm_campaign=rss","source":"STAT News","lang":"en","kind":"news","summaryZh":"STAT News于2026-10-20发布的文章，涉及AI 生物/医药、AI 医疗、AI 商业融资，原题《STAT+: Pharmalittle: We’re reading about drug launch prices, an AstraZeneca AI deal, and m》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"4825cf433d8f","date":"2026-10-20","title":"Citrini Bets on Life Sciences After Claude AI Drug Study","url":"https://www.bloomberg.com/news/articles/2026-10-20/citrini-bets-on-life-sciences-stocks-after-claude-ai-drug-study","source":"Bloomberg Technology","lang":"en","kind":"news","summaryZh":"Bloomberg Technology于2026-10-20发布的文章，涉及AI 金融、AI 生物/医药，原题《Citrini Bets on Life Sciences After Claude AI Drug Study》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"1dd5c007c495","date":"2026-10-20","title":"Nanoscale tech tracks proteins inside living neurons to guide Alzheimer's and Parkinson's drug discovery","url":"https://phys.org/news/2026-10-nanoscale-tech-tracks-proteins-neurons.html","source":"Phys.org 化学·材料(纳米)","lang":"en","kind":"news","summaryZh":"Phys.org 化学·材料(纳米)于2026-10-20发布的文章，涉及AI 化学材料、AI 生物/医药、AI 教育，原题《Nanoscale tech tracks proteins inside living neurons to guide Alzheimer's and Parkinson's 》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"4f2bfcfd9737","date":"2026-10-19","title":"'24/7 unmanned synthesis platform' uses robots and light to produce drug candidates","url":"https://phys.org/news/2026-10-unmanned-synthesis-platform-robots-drug.html","source":"Phys.org 化学","lang":"en","kind":"news","summaryZh":"Phys.org 化学于2026-10-20发布的文章，涉及AI 化学材料、AI 医疗、AI 机器人，原题《'24/7 unmanned synthesis platform' uses robots and light to produce drug candidates》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"bded5b119b7f","date":"2026-10-19","title":"Proteome scale drug prioritization from AlphaFold 3 interaction confidence","url":"https://www.biorxiv.org/content/10.64898/2026.10.19.757941v1","source":"bioRxiv API · 生物信息","lang":"en","kind":"paper","summaryZh":"bioRxiv API · 生物信息于2026-10-19发布的论文，涉及AI 生物/医药、AI 气候能源、AI 商业融资，原题《Proteome scale drug prioritization from AlphaFold 3 interaction confidence》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"68a6cd6d9d4c","date":"2026-10-16","title":"Network-enabled drug repurposing identifies lithium to promote cardiomyocyte proliferation","url":"https://www.biorxiv.org/content/10.64898/2026.10.15.759417v1","source":"bioRxiv API · 系统生物","lang":"en","kind":"paper","summaryZh":"bioRxiv API · 系统生物于2026-10-16发布的论文，涉及AI 生物/医药、AI 医疗、AI 软件开发，原题《Network-enabled drug repurposing identifies lithium to promote cardiomyocyte proliferation》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"05290bca769b","date":"2026-10-15","title":"AI Biotech Iambic Gets $150M in IPO Cash for ‘Molecular Superintelligence’ Drug R&D","url":"https://medcitynews.com/2026/10/iambic-ipo-techbio-ai-drug-discovery-cancer-her2-kif18a-cdk2-cdk4-iam/","source":"MedCity News","lang":"en","kind":"news","summaryZh":"MedCity News于2026-10-16发布的文章，涉及AI 医疗、AI 生物/医药、AI 商业融资，原题《AI Biotech Iambic Gets $150M in IPO Cash for ‘Molecular Superintelligence’ Drug R&D》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"1a0605610a6b","date":"2026-10-15","title":"inClinico Agent for clinical trial endpoint prediction and prospective evaluation","url":"https://www.medrxiv.org/content/10.64898/2026.10.13.26365373v1","source":"medRxiv API · 健康信息学","lang":"en","kind":"paper","summaryZh":"medRxiv API · 健康信息学于2026-10-15发布的论文，涉及AI 医疗、AI 金融、AI 安全，原题《inClinico Agent for clinical trial endpoint prediction and prospective evaluation》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"550b47acf648","date":"2026-10-13","title":"Weekly Tech+Bio #87: TechBio 3.0 Is Closing the Loop. Is \"Prompt to Drug\" Next?","url":"https://www.techlifesci.com/p/weekly-techbio-87-techbio-30-is-closing","source":"Tech+Bio Highlights","lang":"en","kind":"news","summaryZh":"Tech+Bio Highlights于2026-10-14发布的文章，涉及AI 生物/医药、AI 软件开发，原题《Weekly Tech+Bio #87: TechBio 3.0 Is Closing the Loop. Is \"Prompt to Drug\" Next?》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"208102b750b5","date":"2026-09-14","title":"FDA grants priority review to zasocitinib","url":"https://www.takeda.com/newsroom/newsreleases/2026/fda-priority-review-zasocitinib-psoriasis/","source":"Takeda","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"ok"},{"id":"5d25d9f9db3f","date":"2026-07-08","title":"Insilico initiates Phase III trial of rentosertib","url":"https://insilico.com/news/xmjsn4l091-insilico-initiates-phase-iii-clinical-tr","source":"Insilico Medicine","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"ok"}],"review":{"lastReviewed":"2026-10-21","cycleDays":90,"nextReview":"2027-01-19"},"urls":{"zh":"/zh-hans/disciplines/pharmacy/","en":"/en/disciplines/pharmacy/"},"dimensions":{"method":{"score":3,"label":{"zh":"研究方法","en":"Research method"},"evidence":"AI 靶点发现、分子生成和 ADMET 预测已是主要药企的常规流程","source":null,"weight":0.3},"discovery":{"score":2,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"AI 发现的抗生素（halicin、abaucin）和 AI 设计药物进入 3 期；但截至 2026-10 无 AI 设计药获批","source":"https://insilico.com/news/xmjsn4l091-insilico-initiates-phase-iii-clinical-tr","weight":0.35},"adoption":{"score":2,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"缺少独立的行业采用率调查","source":null,"weight":0.25},"education":{"score":1,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到课程证据","source":null,"weight":0.1}},"compositeScore":2.2,"levelCapped":false,"levelChange":{"from":2,"to":2,"direction":"same","reason":null},"biggestBreakthrough":{"date":"2026-10-20","title":{"zh":"Radical Numerics：Omnii-1 设计的噬菌体在同情用药途径下首次用于耐药大肠杆菌感染患者","en":"Radical Numerics: Omnii-1-designed phages administered to a patient under compassionate use"},"url":"https://www.radicalnumerics.ai/blog/omnii-phage","aiImpact":"major","fieldImpact":"major","why":{"zh":"AI：基因组语言模型设计的 T7 噬菌体变体首次进入人体（实验室中清除了患者分离株）；学科：个性化噬菌体疗法的新路径，但单例、未披露疗效，尚无同行评审","en":"AI: genome-LM-designed T7 phage variants reach a human for the first time; Field: a new route to personalised phage therapy, single case, outcome undisclosed"}},"keyPapers":[{"title":"Deep learning-guided discovery of an antibiotic targeting Acinetobacter baumannii","titleZh":"深度学习发现窄谱抗生素 abaucin，靶向鲍曼不动杆菌（Nat. Chem. Biol.）","id":"10.1038/s41589-023-01349-8","url":"https://doi.org/10.1038/s41589-023-01349-8","venue":"Nature Chemical Biology","date":"2023-05-25","why":{"zh":"AI 针对 WHO 头号耐药病原体找到新化合物","en":"AI finds a compound for a WHO priority pathogen"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Discovery of a structural class of antibiotics with explainable deep learning","titleZh":"可解释深度学习发现一类新的结构类别抗生素（Nature）","id":"10.1038/s41586-023-06887-8","url":"https://doi.org/10.1038/s41586-023-06887-8","venue":"Nature","date":"2023-12-20","why":{"zh":"发现对 MRSA 有效的新结构类别，并解释模型依据的子结构","en":"New structural class active against MRSA, with explained substructures"},"fieldImpact":"major","aiImpact":"major"},{"title":"A small-molecule TNIK inhibitor targets fibrosis in preclinical and clinical models","titleZh":"Insilico：AI 发现靶点 + 生成式设计的 TNIK 抑制剂从发现到临床全过程（Nature Biotechnology）","id":"10.1038/s41587-024-02143-0","url":"https://doi.org/10.1038/s41587-024-02143-0","venue":"Nature Biotechnology","date":"2024-03-08","why":{"zh":"首个公开发表全流程的“AI 靶点 + AI 分子”药物","en":"First published end-to-end AI-target + AI-molecule drug"},"fieldImpact":"major","aiImpact":"major"},{"title":"A generative AI-discovered TNIK inhibitor for idiopathic pulmonary fibrosis: a randomized phase 2a trial","titleZh":"rentosertib（TNIK 抑制剂）2a 期结果发表（Nature Medicine）","id":"10.1038/s41591-025-03743-2","url":"https://doi.org/10.1038/s41591-025-03743-2","venue":"Nature Medicine","date":"2025-06-03","why":{"zh":"AI 设计药物首个随机对照 2a 期结果在顶刊发表","en":"First randomised Phase 2a of an AI-designed drug in a top journal"},"fieldImpact":"major","aiImpact":"major"}],"latest_papers":[{"id":"arXiv:2610.25605","title":"PRISM: Prior Relational Information for Self-supervised Modeling to Enhance Solubility OOD Generalization","titleZh":"PRISM：引入先验关系信息的自监督建模提升溶解度分布外泛化","summaryZh":"让自监督模型显式学习溶质与溶剂间的化学关系，改善对未见溶质、溶剂及混合溶剂的溶解度预测。","summaryEn":"Self-supervised learning with prior solute–solvent relational information improves out-of-distribution solubility prediction.","authors":["Xinyi Chen","Jiahuan Pang","Titus Chima","Paul Weng"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.25605","site":"arXiv","venue":null,"category":"cs.LG","fetchedAt":"2026-10-22T10:07:23"},{"id":"arXiv:2610.24458","title":"Generative Retrieval for Reverse Target Fishing with a Proteome-Scale Benchmark","titleZh":"用生成式检索做反向靶点垂钓及蛋白质组规模基准","summaryZh":"把为分子寻找蛋白靶点的问题改写为生成式检索，提出 GenFish 和蛋白质组规模的基准。","summaryEn":"Reformulates reverse target fishing as generative retrieval (GenFish) with a proteome-scale benchmark.","authors":["Zhichao Deng","Haichuan Tan","Xin Hong","Huaizhi Wang"],"date":"2026-10-17","url":"https://arxiv.org/abs/2610.24458","site":"arXiv","venue":null,"category":"cs.LG","fetchedAt":"2026-10-22T10:07:23"},{"id":"arXiv:2610.24124","title":"On the Selectivity of Generative Models in Structure-Based Drug Design","titleZh":"结构药物设计中生成模型的选择性问题","summaryZh":"指出多数深度生成模型只优化与单一靶点的结合而忽略选择性，研究如何让设计的小分子避免脱靶。","summaryEn":"Examines selectivity in deep generative structure-based drug design, which usually optimises single-target binding.","authors":["Ella Miray Rajaonson","Jungyoon Lee","William Chau","Alan Aspuru-Guzik"],"date":"2026-10-16","url":"https://arxiv.org/abs/2610.24124","site":"arXiv","venue":null,"category":"cs.LG","fetchedAt":"2026-10-22T10:07:23"},{"id":"arXiv:2610.19123","title":"PairIT: Autoregressive Transformers for Low-Data Molecule Optimization","titleZh":"PairIT：低数据条件下的自回归 Transformer 分子优化","summaryZh":"提出 PairIT 训练框架，在数据稀少时以种子分子为条件、同时优化多个性质生成新分子。","summaryEn":"PairIT trains SMILES transformers for seed-conditioned, multi-property molecule optimisation with little data.","authors":["Hannah Lawrence","Bodhi P. Vani","Ji Won Park","Natasa Tagasovska"],"date":"2026-10-13","url":"https://arxiv.org/abs/2610.19123","site":"arXiv","venue":null,"category":"cs.LG","fetchedAt":"2026-10-22T10:07:23"},{"id":"arXiv:2610.19011","title":"PMO-Dock: Benchmarking Docking, Specificity, and Generalization in Molecular Optimization","titleZh":"PMO-Dock：分子优化中对接、特异性与泛化的基准","summaryZh":"在 PMO 基准基础上加入基于对接的目标，为结构药物设计中的分子优化提供统一公平的评测。","summaryEn":"Extends the PMO benchmark with docking-based objectives for fair evaluation of molecular optimisation.","authors":["Gor Simonyan","Tatevik Abrahamyan","Narek Abelyan","Tigran Fahradyan"],"date":"2026-10-13","url":"https://arxiv.org/abs/2610.19011","site":"arXiv","venue":null,"category":"cs.LG","fetchedAt":"2026-10-22T10:07:23"},{"id":"arXiv:2610.18851","title":"MFG: Multi-turn Feedback Guided Molecular Optimization","titleZh":"MFG：多轮反馈引导的分子优化","summaryZh":"用多轮计算或实验反馈引导大模型逐步修改分子，同时改进多个性质并保持结构相似。","summaryEn":"Multi-turn feedback guides LLM-based multi-objective molecule optimisation.","authors":["Trieu Nguyen","Frazier N. Baker","Ruoxi Gao","Daniel Adu-Ampratwum"],"date":"2026-10-13","url":"https://arxiv.org/abs/2610.18851","site":"arXiv","venue":null,"category":"cs.AI","fetchedAt":"2026-10-22T10:07:23"},{"id":"arXiv:2610.18456","title":"Learning Perturbation-Specific Representations through Conditional Energy Prior for Cellular Response Prediction","titleZh":"通过条件能量先验学习扰动特异表示以预测细胞响应","summaryZh":"提出 PerturbEnergy，用条件能量先验分离扰动特异信号，提高 AI 预测药物等扰动下细胞响应的能力。","summaryEn":"PerturbEnergy uses a conditional energy prior to isolate perturbation-specific signals for cellular response prediction.","authors":["Yuchen Yan","Yatao Bian"],"date":"2026-10-13","url":"https://arxiv.org/abs/2610.18456","site":"arXiv","venue":null,"category":"cs.LG","fetchedAt":"2026-10-22T10:07:23"}]},{"id":"clinical-medicine","slug":"clinical-medicine","category":"life-medical","categoryName":{"zh":"生命与医学","en":"Life & medical sciences"},"name":{"zh":"临床医学","en":"Clinical medicine"},"classification":{"oecdFos":"3.2","moe":"1002 临床医学"},"level":2,"levelKey":"breakthrough","levelLabel":{"zh":"局部突破","en":"Local breakthroughs"},"trend":"up","reason":{"zh":"FDA 已列出 1614 个 AI 医疗器械（2026-09-04 更新）；AMIE 在模拟问诊中诊断准确率超过对照医生（Nature 2025）。但大模型直接参与诊疗的前瞻性证据仍少。","en":"FDA lists 1,614 AI-enabled devices (4 Sep 2026 update); AMIE out-diagnosed clinicians in simulated consults (Nature 2025). Prospective evidence for LLMs in care is still thin."},"reasonSources":[{"url":"https://www.fda.gov/medical-devices/digital-health-center-excellence/artificial-intelligence-enabled-medical-devices","linkCheck":"ok"},{"url":"https://www.nature.com/articles/s41586-025-08866-7","linkCheck":"ok"}],"directions":{"zh":["诊断对话智能体","临床文书自动化","风险预测","医疗大模型评测"],"en":["Diagnostic dialogue agents","Clinical documentation","Risk prediction","Medical LLM evaluation"]},"stats":{"milestones":10,"milestonesSince2025":6,"milestones2026":4,"milestonesLast6Months":4,"news":12,"latestProgress":"2026-10-20","latestMilestone":"2026-10-13","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":0,"leaps":{"ai":0,"field":0},"latestPapers":8,"keyPapers":8},"milestones":[{"date":"2018-04-11","title":{"zh":"FDA 批准首个自主 AI 诊断设备 IDx-DR（糖网筛查）","en":"FDA authorises first autonomous AI diagnostic, IDx-DR"},"url":"https://www.accessdata.fda.gov/scripts/cdrh/cfdocs/cfPMN/denovo.cfm?knumber=DEN180001&start_search=1","source":"FDA De Novo DEN180001","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"regulatory","typeLabel":{"zh":"监管","en":"Regulatory"},"why":{"zh":"FDA 首次批准无需医生解读的自主 AI 诊断设备","en":"First FDA-authorised autonomous AI diagnostic"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"FDA 首次批准无需医生解读的自主 AI 诊断设备","en":"First FDA-authorised autonomous AI diagnostic"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"FDA 首次批准无需医生解读的自主 AI 诊断设备","en":"First FDA-authorised autonomous AI diagnostic"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2018-08-28","title":{"zh":"IDx-DR 关键性前瞻试验：自主 AI 在初级诊疗中筛查糖尿病视网膜病变（npj Digit. Med.）","en":"Pivotal trial of autonomous AI for diabetic retinopathy in primary care (npj Digit. Med.)"},"url":"https://doi.org/10.1038/s41746-018-0040-6","source":"npj Digital Medicine","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"clinical","typeLabel":{"zh":"临床","en":"Clinical"},"why":{"zh":"支撑首个无需医生解读即可出诊断结果的 AI 获批","en":"Evidence behind the first autonomous AI diagnostic clearance"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"支撑首个无需医生解读即可出诊断结果的 AI 获批","en":"Evidence behind the first autonomous AI diagnostic clearance"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"支撑首个无需医生解读即可出诊断结果的 AI 获批","en":"Evidence behind the first autonomous AI diagnostic clearance"}},"impactNotes":[],"verified":{"id":"10.1038/s41746-018-0040-6","title":"Pivotal trial of an autonomous AI-based diagnostic system for detection of diabetic retinopathy in primary care offices","venue":"npj Digital Medicine","date":"2018-08-28"},"addedIn":"v2"},{"date":"2019-01-07","title":{"zh":"Topol：高性能医学——人与人工智能的融合（Nature Medicine 综述）","en":"Topol: High-performance medicine, the convergence of human and AI (Nature Medicine)"},"url":"https://doi.org/10.1038/s41591-018-0300-7","source":"Nature Medicine","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"系统提出 AI 改造临床医学的路线图，被大量引用","en":"Influential roadmap for AI in clinical medicine"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"系统提出 AI 改造临床医学的路线图，被大量引用","en":"Influential roadmap for AI in clinical medicine"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"系统提出 AI 改造临床医学的路线图，被大量引用","en":"Influential roadmap for AI in clinical medicine"}},"impactNotes":[],"verified":{"id":"10.1038/s41591-018-0300-7","title":"High-performance medicine: the convergence of human and artificial intelligence","venue":"Nature Medicine","date":"2019-01-07"},"addedIn":"v2"},{"date":"2023-07-12","title":{"zh":"Med-PaLM：大模型编码临床知识（Nature）","en":"Med-PaLM: LLMs encode clinical knowledge (Nature)"},"url":"https://www.nature.com/articles/s41586-023-06291-2","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"大模型首次在美国执业医师考试题上达到及格水平","en":"First LLM to pass USMLE-style questions"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"大模型首次在美国执业医师考试题上达到及格水平","en":"First LLM to pass USMLE-style questions"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"大模型首次在美国执业医师考试题上达到及格水平","en":"First LLM to pass USMLE-style questions"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-023-06291-2","title":"Large language models encode clinical knowledge","venue":"Nature","date":"2023-07-12"},"addedIn":"v1"},{"date":"2025-03-27","title":{"zh":"Therabot：生成式 AI 心理治疗聊天机器人首个随机对照试验（NEJM AI）","en":"Therabot: first RCT of a generative-AI therapy chatbot (NEJM AI)"},"url":"https://doi.org/10.1056/aioa2400802","source":"NEJM AI","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"clinical","typeLabel":{"zh":"临床","en":"Clinical"},"why":{"zh":"210 名成人中抑郁、焦虑、进食障碍症状均较对照组显著下降","en":"Significant symptom reductions vs waitlist in 210 adults"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"210 名成人中抑郁、焦虑、进食障碍症状均较对照组显著下降","en":"Significant symptom reductions vs waitlist in 210 adults"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"210 名成人中抑郁、焦虑、进食障碍症状均较对照组显著下降","en":"Significant symptom reductions vs waitlist in 210 adults"}},"impactNotes":[],"verified":{"id":"10.1056/aioa2400802","title":"Randomized Trial of a Generative AI Chatbot for Mental Health Treatment","venue":"NEJM AI","date":"2025-03-27"},"addedIn":"v2"},{"date":"2025-04-09","title":{"zh":"AMIE 诊断对话 AI（Nature）","en":"AMIE conversational diagnostic AI (Nature)"},"url":"https://www.nature.com/articles/s41586-025-08866-7","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"模拟问诊中诊断准确率和多项沟通指标超过初级保健医生","en":"Beats PCPs on diagnostic accuracy in simulated consults"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"模拟问诊中诊断准确率和多项沟通指标超过初级保健医生","en":"Beats PCPs on diagnostic accuracy in simulated consults"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"模拟问诊中诊断准确率和多项沟通指标超过初级保健医生","en":"Beats PCPs on diagnostic accuracy in simulated consults"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-025-08866-7","title":"Towards conversational diagnostic artificial intelligence","venue":"Nature","date":"2025-04-09"},"addedIn":"v1"},{"date":"2026-04-23","title":{"zh":"受过 AI 素养培训的医生在大模型辅助下仍出现自动化偏差（NEJM AI 随机试验）","en":"Automation bias in LLM-assisted diagnosis among AI-literate physicians (NEJM AI RCT)"},"url":"https://doi.org/10.1056/aioa2501001","source":"NEJM AI","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"clinical","typeLabel":{"zh":"临床","en":"Clinical"},"why":{"zh":"提示人机协作诊断的风险","en":"Shows risks in human–AI diagnostic teaming"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"提示人机协作诊断的风险","en":"Shows risks in human–AI diagnostic teaming"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"提示人机协作诊断的风险","en":"Shows risks in human–AI diagnostic teaming"}},"impactNotes":[],"verified":{"id":"10.1056/aioa2501001","title":"Automation Bias in Large Language Model–Assisted Diagnostic Reasoning among Physicians Trained in AI Literacy — A Randomized Clinical Trial","venue":"NEJM AI","date":"2026-04-23"},"addedIn":"v2"},{"date":"2026-06-26","title":{"zh":"肯尼亚初级诊疗整群随机试验：大模型临床决策支持未降低 14 天治疗失败率（Nature Medicine）","en":"Kenya cluster-RCT: LLM decision support did not reduce 14-day treatment failure (Nature Medicine)"},"url":"https://doi.org/10.1038/s41591-026-04503-6","source":"Nature Medicine","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"clinical","typeLabel":{"zh":"临床","en":"Clinical"},"why":{"zh":"9691 名患者的大型试验给出“安全但无显著获益”的结论，是评估大模型临床价值的重要反证","en":"9,691-patient trial: safe but no significant benefit"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"9691 名患者的大型试验给出“安全但无显著获益”的结论，是评估大模型临床价值的重要反证","en":"9,691-patient trial: safe but no significant benefit"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"9691 名患者的大型试验给出“安全但无显著获益”的结论，是评估大模型临床价值的重要反证","en":"9,691-patient trial: safe but no significant benefit"}},"impactNotes":[],"verified":{"id":"10.1038/s41591-026-04503-6","title":"Generative AI-enabled clinical decision support system in primary care: a pragmatic, cluster-randomized trial","venue":"Nature Medicine","date":"2026-06-26"},"addedIn":"v2"},{"date":"2026-09-04","title":{"zh":"FDA AI 医疗器械清单更新至 1614 个","en":"FDA AI-enabled device list updated to 1,614"},"url":"https://www.fda.gov/medical-devices/digital-health-center-excellence/artificial-intelligence-enabled-medical-devices","source":"FDA","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"regulatory","typeLabel":{"zh":"监管","en":"Regulatory"},"why":{"zh":"FDA AI 器械清单累计 1614 个，反映临床部署规模","en":"1,614 AI devices on the FDA list"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"FDA AI 器械清单累计 1614 个，反映临床部署规模","en":"1,614 AI devices on the FDA list"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"FDA AI 器械清单累计 1614 个，反映临床部署规模","en":"1,614 AI devices on the FDA list"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-10-13","title":{"zh":"宾大：大模型 AI 连接头皮与颅内脑电，经手机与癫痫患者实时对话、共同管理病情（Nature Medicine，13 例）","en":"Human–AI collaboration for comanaging epilepsy in real time (Nature Medicine)"},"url":"https://www.nature.com/articles/s41591-026-04680-4","source":"Nature Medicine","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"clinical","typeLabel":{"zh":"临床","en":"Clinical"},"why":{"zh":"AI：LLM 根据睡眠、用药、认知和发作概率等生物标志物与患者双向沟通，并用反馈重训发作检测器；学科：医疗器械从“只记录”走向与患者实时协作的新模式（住院原型）","en":"AI: an LLM talks with patients about EEG biomarkers and retrains its detector from feedback; Field: devices move from passive recording to real-time co-management (inpatient prototype)"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：LLM 根据睡眠、用药、认知和发作概率等生物标志物与患者双向沟通，并用反馈重训发作检测器；学科：医疗器械从“只记录”走向与患者实时协作的新模式（住院原型）","en":"AI: an LLM talks with patients about EEG biomarkers and retrains its detector from feedback; Field: devices move from passive recording to real-time co-management (inpatient prototype)"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：LLM 根据睡眠、用药、认知和发作概率等生物标志物与患者双向沟通，并用反馈重训发作检测器；学科：医疗器械从“只记录”走向与患者实时协作的新模式（住院原型）","en":"AI: an LLM talks with patients about EEG biomarkers and retrains its detector from feedback; Field: devices move from passive recording to real-time co-management (inpatient prototype)"}},"impactNotes":[],"verified":{"id":"10.1038/s41591-026-04680-4","title":"Human–artificial intelligence collaboration for comanaging epilepsy in real time","venue":"Nature Medicine","date":"2026-10-13"},"addedIn":"v2"}],"papersRepos":[{"title":"Pivotal trial of an autonomous AI-based diagnostic system for detection of diabetic retinopathy in primary care offices","url":"https://doi.org/10.1038/s41746-018-0040-6","date":"2018-08-28","kind":"paper","linkCheck":"ok-crossref"},{"title":"High-performance medicine: the convergence of human and artificial intelligence","url":"https://doi.org/10.1038/s41591-018-0300-7","date":"2019-01-07","kind":"paper","linkCheck":"ok-crossref"},{"title":"Large language models encode clinical knowledge","url":"https://doi.org/10.1038/s41586-023-06291-2","date":"2023-07-12","kind":"paper","linkCheck":"ok-crossref"},{"title":"Randomized Trial of a Generative AI Chatbot for Mental Health Treatment","url":"https://doi.org/10.1056/aioa2400802","date":"2025-03-27","kind":"paper","linkCheck":"ok-crossref"},{"title":"Towards conversational diagnostic artificial intelligence","url":"https://doi.org/10.1038/s41586-025-08866-7","date":"2025-04-09","kind":"paper","linkCheck":"ok-crossref"},{"title":"Generative AI-enabled clinical decision support system in primary care: a pragmatic, cluster-randomized trial","url":"https://doi.org/10.1038/s41591-026-04503-6","date":"2026-06-26","kind":"paper","linkCheck":"ok-crossref"},{"title":"Human–artificial intelligence collaboration for comanaging epilepsy in real time","url":"https://doi.org/10.1038/s41591-026-04680-4","date":"2026-10-13","kind":"paper","linkCheck":"ok-crossref"},{"title":"Automation Bias in Large Language Model–Assisted Diagnostic Reasoning among Physicians Trained in AI Literacy — A Randomized Clinical Trial","url":"https://doi.org/10.1056/aioa2501001","date":"2026-04-23","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"FDA Digital Health Center of Excellence","url":"https://www.fda.gov/medical-devices/digital-health-center-excellence"},{"name":"Google Health","url":"https://health.google"}],"news":[{"id":"20f1d6773062","date":"2026-10-20","title":"Liquid neural network-driven intelligent topology perception and path anomaly diagnosis for Ipv6 smart grids","url":"https://www.nature.com/articles/s41598-026-76378-z","source":"Nature 学科·数学","lang":"en","kind":"news","summaryZh":"Nature 学科·数学于2026-10-20发布的文章，涉及AI 数学、AI 医疗，原题《Liquid neural network-driven intelligent topology perception and path anomaly diagnosis fo》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"2f91e4fb8506","date":"2026-10-20","title":"Expertise modulates automation bias and sentinel behavior in human–AI collaborative diagnosis of neonatal pneumoperitoneum","url":"https://www.nature.com/articles/s41746-026-03399-3","source":"Nature 学科·计算生物","lang":"en","kind":"news","summaryZh":"Nature 学科·计算生物于2026-10-20发布的文章，涉及AI 生物/医药、AI 医疗，原题《Expertise modulates automation bias and sentinel behavior in human–AI collaborative diagno》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"16d59cdddbc5","date":"2026-10-20","title":"AZ and Oracle team up on AI platform for clinical trials","url":"https://pharmaphorum.com/news/az-and-oracle-team-ai-platform-clinical-trials","source":"pharmaphorum","lang":"en","kind":"news","summaryZh":"pharmaphorum于2026-10-20发布的文章，涉及AI 生物/医药、AI 医疗，原题《AZ and Oracle team up on AI platform for clinical trials》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"939922e31753","date":"2026-10-20","title":"GliomaINK: A Tumor and Anatomy Conditioned Flow Matching Framework for 3D MRI Sequence Synthesis in Glioma Patients","url":"https://arxiv.org/abs/2610.24044","source":"arXiv q-bio","lang":"en","kind":"paper","summaryZh":"arXiv q-bio于2026-10-20发布的论文，涉及AI 生物/医药、AI 医疗、AI 软件开发，原题《GliomaINK: A Tumor and Anatomy Conditioned Flow Matching Framework for 3D MRI Sequence Syn》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"7b8383d5ab45","date":"2026-10-20","title":"WILSON - Multi-magnification composites as a foundation for patient-level pathology analysis and diagnostic text generation","url":"https://arxiv.org/abs/2609.25123","source":"arXiv q-bio","lang":"en","kind":"paper","summaryZh":"arXiv q-bio于2026-10-20发布的论文，涉及AI 生物/医药、AI 医疗，原题《WILSON - Multi-magnification composites as a foundation for patient-level pathology analys》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"23a8a613afad","date":"2026-10-20","title":"A Controlled LLM Workflow for Evidence-Grounded Diagnosis of Risk Score Differences","url":"https://arxiv.org/abs/2610.23962","source":"arXiv cs.AI","lang":"en","kind":"paper","summaryZh":"arXiv cs.AI于2026-10-20发布的论文，涉及AI 科研与调研、AI 医疗、AI 安全，原题《A Controlled LLM Workflow for Evidence-Grounded Diagnosis of Risk Score Differences》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"6e336da678f0","date":"2026-10-19","title":"ASHG: Long-Read Genomics Meets Clinical Data in PacBio, Verily Health AI Deal","url":"https://www.genengnews.com/topics/omics/ashg-long-read-genomics-meets-clinical-data-in-pacbio-verily-health-ai-deal/","source":"GEN 话题：Artificial Intelligence","lang":"en","kind":"news","summaryZh":"GEN 话题：Artificial Intelligence于2026-10-20发布的文章，涉及AI 生物/医药、AI 医疗、AI 商业融资，原题《ASHG: Long-Read Genomics Meets Clinical Data in PacBio, Verily Health AI Deal》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"0dbde1ea1153","date":"2026-10-19","title":"Why teaching AI about symmetry could make prenatal diagnosis more reliable","url":"https://news.mit.edu/2026/teaching-ai-about-symmetry-could-make-prenatal-diagnosis-more-reliable-1019","source":"MIT News：Artificial intelligence","lang":"en","kind":"news","summaryZh":"MIT News：Artificial intelligence于2026-10-20发布的文章，涉及AI 科研与调研、AI 医疗，原题《Why teaching AI about symmetry could make prenatal diagnosis more reliable》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"0e6cbafd58a2","date":"2026-10-19","title":"IIT Kanpur, Gates Foundation Partner to Advance Affordable Diagnostics for Underserved Communities","url":"https://www.digitalhealthnews.com/iit-kanpur-gates-foundation-partner-to-advance-affordable-diagnostics-for-underserved-communities","source":"Digital Health News","lang":"en","kind":"news","summaryZh":"Digital Health News于2026-10-19发布的文章，涉及AI 医疗、AI 生物/医药，原题《IIT Kanpur, Gates Foundation Partner to Advance Affordable Diagnostics for Underserved Com》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"e67a858e355a","date":"2026-10-19","title":"ICMR Selects Seven Indian Institutions to Develop Medical Devices & Diagnostic Technologies","url":"https://www.digitalhealthnews.com/icmr-selects-seven-indian-institutions-to-develop-medical-devices-diagnostic-technologies","source":"Digital Health News","lang":"en","kind":"news","summaryZh":"Digital Health News于2026-10-19发布的文章，涉及AI 医疗、AI 商业融资，原题《ICMR Selects Seven Indian Institutions to Develop Medical Devices & Diagnostic Technologie》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"0c3049e37992","date":"2026-10-19","title":"AMA Introduces Governance Framework for Agentic AI in Clinical Care","url":"https://www.digitalhealthnews.com/ama-introduces-governance-framework-for-agentic-ai-in-clinical-care","source":"Digital Health News","lang":"en","kind":"news","summaryZh":"Digital Health News于2026-10-19发布的文章，涉及AI 医疗、AI 法律政策、AI 软件开发，原题《AMA Introduces Governance Framework for Agentic AI in Clinical Care》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"644354619d86","date":"2026-10-19","title":"Diagnosis Is Beside the Point","url":"https://medcitynews.com/2026/10/diagnosis-is-beside-the-point/","source":"MedCity News","lang":"en","kind":"news","summaryZh":"MedCity News于2026-10-19发布的文章，涉及AI 医疗，原题《Diagnosis Is Beside the Point》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"}],"review":{"lastReviewed":"2026-10-21","cycleDays":90,"nextReview":"2027-01-19"},"urls":{"zh":"/zh-hans/disciplines/clinical-medicine/","en":"/en/disciplines/clinical-medicine/"},"dimensions":{"method":{"score":2,"label":{"zh":"研究方法","en":"Research method"},"evidence":"临床研究中 AI 多为辅助分析；大模型前瞻性随机试验 2025–2026 才开始出现且结果不一","source":"https://doi.org/10.1038/s41591-026-04503-6","weight":0.3},"discovery":{"score":2,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"AMIE 在模拟问诊中诊断准确率超过医生（Nature 2025），属受控环境突破","source":"https://www.nature.com/articles/s41586-025-08866-7","weight":0.35},"adoption":{"score":3,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"FDA 已列 1614 个 AI 器械（2026-09-04），临床环境已大量部署","source":"https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-enabled-medical-devices","weight":0.25},"education":{"score":1,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到课程体系改革的公开证据","source":null,"weight":0.1}},"compositeScore":2.15,"levelCapped":false,"levelChange":{"from":2,"to":2,"direction":"same","reason":null},"biggestBreakthrough":{"date":"2025-04-09","title":{"zh":"AMIE 诊断对话 AI（Nature）","en":"AMIE conversational diagnostic AI (Nature)"},"url":"https://www.nature.com/articles/s41586-025-08866-7","aiImpact":"major","fieldImpact":"major","why":{"zh":"模拟问诊中诊断准确率和多项沟通指标超过初级保健医生","en":"Beats PCPs on diagnostic accuracy in simulated consults"}},"keyPapers":[{"title":"Pivotal trial of an autonomous AI-based diagnostic system for detection of diabetic retinopathy in primary care offices","titleZh":"IDx-DR 关键性前瞻试验：自主 AI 在初级诊疗中筛查糖尿病视网膜病变（npj Digit. Med.）","id":"10.1038/s41746-018-0040-6","url":"https://doi.org/10.1038/s41746-018-0040-6","venue":"npj Digital Medicine","date":"2018-08-28","why":{"zh":"支撑首个无需医生解读即可出诊断结果的 AI 获批","en":"Evidence behind the first autonomous AI diagnostic clearance"},"fieldImpact":"major","aiImpact":"major"},{"title":"High-performance medicine: the convergence of human and artificial intelligence","titleZh":"Topol：高性能医学——人与人工智能的融合（Nature Medicine 综述）","id":"10.1038/s41591-018-0300-7","url":"https://doi.org/10.1038/s41591-018-0300-7","venue":"Nature Medicine","date":"2019-01-07","why":{"zh":"系统提出 AI 改造临床医学的路线图，被大量引用","en":"Influential roadmap for AI in clinical medicine"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Large language models encode clinical knowledge","titleZh":"Med-PaLM：大模型编码临床知识（Nature）","id":"10.1038/s41586-023-06291-2","url":"https://doi.org/10.1038/s41586-023-06291-2","venue":"Nature","date":"2023-07-12","why":{"zh":"大模型首次在美国执业医师考试题上达到及格水平","en":"First LLM to pass USMLE-style questions"},"fieldImpact":"incremental","aiImpact":"major"},{"title":"Randomized Trial of a Generative AI Chatbot for Mental Health Treatment","titleZh":"Therabot：生成式 AI 心理治疗聊天机器人首个随机对照试验（NEJM AI）","id":"10.1056/aioa2400802","url":"https://doi.org/10.1056/aioa2400802","venue":"NEJM AI","date":"2025-03-27","why":{"zh":"210 名成人中抑郁、焦虑、进食障碍症状均较对照组显著下降","en":"Significant symptom reductions vs waitlist in 210 adults"},"fieldImpact":"major","aiImpact":"major"},{"title":"Towards conversational diagnostic artificial intelligence","titleZh":"AMIE 诊断对话 AI（Nature）","id":"10.1038/s41586-025-08866-7","url":"https://doi.org/10.1038/s41586-025-08866-7","venue":"Nature","date":"2025-04-09","why":{"zh":"模拟问诊中诊断准确率和多项沟通指标超过初级保健医生","en":"Beats PCPs on diagnostic accuracy in simulated consults"},"fieldImpact":"major","aiImpact":"major"},{"title":"Generative AI-enabled clinical decision support system in primary care: a pragmatic, cluster-randomized trial","titleZh":"肯尼亚初级诊疗整群随机试验：大模型临床决策支持未降低 14 天治疗失败率（Nature Medicine）","id":"10.1038/s41591-026-04503-6","url":"https://doi.org/10.1038/s41591-026-04503-6","venue":"Nature Medicine","date":"2026-06-26","why":{"zh":"9691 名患者的大型试验给出“安全但无显著获益”的结论，是评估大模型临床价值的重要反证","en":"9,691-patient trial: safe but no significant benefit"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Human–artificial intelligence collaboration for comanaging epilepsy in real time","titleZh":"宾大：大模型 AI 连接头皮与颅内脑电，经手机与癫痫患者实时对话、共同管理病情（Nature Medicine，13 例）","id":"10.1038/s41591-026-04680-4","url":"https://doi.org/10.1038/s41591-026-04680-4","venue":"Nature Medicine","date":"2026-10-13","why":{"zh":"AI：LLM 根据睡眠、用药、认知和发作概率等生物标志物与患者双向沟通，并用反馈重训发作检测器；学科：医疗器械从“只记录”走向与患者实时协作的新模式（住院原型）","en":"AI: an LLM talks with patients about EEG biomarkers and retrains its detector from feedback; Field: devices move from passive recording to real-time co-management (inpatient prototype)"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Automation Bias in Large Language Model–Assisted Diagnostic Reasoning among Physicians Trained in AI Literacy — A Randomized Clinical Trial","titleZh":"受过 AI 素养培训的医生在大模型辅助下仍出现自动化偏差（NEJM AI 随机试验）","id":"10.1056/aioa2501001","url":"https://doi.org/10.1056/aioa2501001","venue":"NEJM AI","date":"2026-04-23","why":{"zh":"提示人机协作诊断的风险","en":"Shows risks in human–AI diagnostic teaming"},"fieldImpact":"incremental","aiImpact":"incremental"}],"latest_papers":[{"id":"arXiv:2610.29606","title":"Overview of CHIP 2026 Shared Task 1: Patient-Specific Phenotype Concept Recognition in Full-Text Biomedical Literature","titleZh":"CHIP 2026 共享任务 1 综述：全文生物医学文献中的患者特异表型概念识别","summaryZh":"介绍面向罕见病诊断的评测任务：在含多名患者的全文文献中识别表型概念，并判断其属于哪位患者。","summaryEn":"Overview of a shared task on recognising phenotype concepts and linking them to specific patients in full-text biomedical papers.","authors":["Xuanlyu Zhu","Feng Tian","Yang Zhong","Pengqi Zheng"],"date":"2026-10-21","url":"https://arxiv.org/abs/2610.29606","site":"arXiv","venue":null,"category":"cs.CL","fetchedAt":"2026-10-22T10:07:39"},{"id":"arXiv:2610.29358","title":"Reading the Mask: Descriptor-Grounded BI-RADS Rationale Generation for Breast MRI Segmentation","titleZh":"解读掩膜：基于 BI-RADS 描述符为乳腺 MRI 分割生成判读理由","summaryZh":"从深度学习分割掩膜和动态增强序列中推导临床描述符，生成符合 BI-RADS 的判读理由，提升可解释性。","summaryEn":"Derives BI-RADS descriptors from breast DCE-MRI segmentation masks to generate radiologist-style rationales.","authors":["Hrishikesh Singh","Aarti Sharma","Divyanshu Kumar","Millie Pant"],"date":"2026-10-21","url":"https://arxiv.org/abs/2610.29358","site":"arXiv","venue":null,"category":"cs.CV","fetchedAt":"2026-10-22T10:07:39"},{"id":"doi:10.64898/2026.10.19.26365895","title":"Towards Early Warning Systems for Shigellosis in Africa: Multi-Horizon Forecasting of Incidence in Rural Western Kenya Using Statistical State-Space and Machine Learning Models","titleZh":"面向非洲志贺菌病早期预警：肯尼亚西部农村发病率多步预测","summaryZh":"用 16 年纵向数据比较统计状态空间模型与机器学习模型，评估 5 岁以下儿童志贺菌病发病率多步预测的准确度与成本。","summaryEn":"Compares state-space and ML models for multi-horizon forecasting of childhood shigellosis in rural western Kenya using 16 years of data.","authors":["Ogwel, B.","Onyando, B. O.","Nyawanda, B. O.","Awuor, A. O."],"date":"2026-10-21","url":"https://www.medrxiv.org/content/10.64898/2026.10.19.26365895v1","site":"medRxiv","venue":null,"category":"infectious diseases","fetchedAt":"2026-10-22T10:07:39"},{"id":"doi:10.64898/2026.10.19.26365885","title":"From Omics Data to Biological Context: The CURIE AI En-gine. Illustrated with the Transcriptomic Response to Bacte-rial Pneumonia","titleZh":"CURIE AI 引擎：从组学数据到生物学语境（以细菌性肺炎转录组为例）","summaryZh":"以细菌性肺炎患者外周血 RNA-seq 为例，演示用 AI 引擎把差异表达与功能富集结果整合成生物学解释。","summaryEn":"Demonstrates an AI engine that contextualises transcriptomic results, using blood RNA-seq from bacterial pneumonia patients.","authors":["Castellani, G.","Rodriguez-Mena, A.","Tarrago-Claramunt, X.","Madarnas-Gomez, V."],"date":"2026-10-21","url":"https://www.medrxiv.org/content/10.64898/2026.10.19.26365885v1","site":"medRxiv","venue":null,"category":"infectious diseases","fetchedAt":"2026-10-22T10:07:39"},{"id":"arXiv:2610.25819","title":"When to Stop Asking: Factor Coverage Bias and LLM Determinability Judgments with Underspecified Information in Medicine","titleZh":"何时停止提问：信息不完整时的医学大模型可判定性判断与因素覆盖偏差","summaryZh":"研究大模型在临床信息不完整时能否判断“现有信息是否足以决策”，发现模型过度依赖已有信息量而非缺失信息能否改变结论。","summaryEn":"Finds LLMs judge clinical determinability by amount of information rather than whether missing facts could change the decision.","authors":["Yusuke Watanabe","Yohei Kobashi","Taisei Tosaki","Takeshi Kojima"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.25819","site":"arXiv","venue":null,"category":"cs.AI","fetchedAt":"2026-10-22T10:07:39"},{"id":"doi:10.64898/2026.10.15.26365116","title":"Deep learning-based electrocardiographic differentiation of Takotsubo syndrome from acute anterior myocardial infarction: explainability, human-AI interaction, and a utilization bias hypothesis","titleZh":"深度学习心电图区分应激性心肌病与急性前壁心梗：可解释性、人机交互与使用偏差","summaryZh":"用两家医院的 12 导联心电图训练深度学习模型区分应激性心肌病与 STEMI，并评估其对不同背景医生诊断表现的影响。","summaryEn":"Deep-learning ECG model distinguishes Takotsubo from anterior STEMI; study measures its effect on readers of different backgrounds.","authors":["Kagawa, E.","Kato, M.","Hidaka, T.","Kunita, E."],"date":"2026-10-19","url":"https://www.medrxiv.org/content/10.64898/2026.10.15.26365116v1","site":"medRxiv","venue":null,"category":"cardiovascular medicine","fetchedAt":"2026-10-22T10:07:39"},{"id":"doi:10.64898/2026.10.15.26365634","title":"From meal text to carbohydrate to postprandial glucose rise: QLoRA fine-tuning of the Qwen3-4B language model and gradient-boosting prediction in type 2 diabetes","titleZh":"从餐食文字到碳水再到餐后血糖：微调 Qwen3-4B 与梯度提升预测 2 型糖尿病餐后血糖","summaryZh":"评估用公开数据微调的小型开源大模型能否从文字描述估算餐食碳水，以及这一估计对预测 2 型糖尿病患者餐后血糖的帮助。","summaryEn":"Tests whether a fine-tuned small open LLM estimates meal carbohydrates from text and improves post-meal glucose prediction in T2D.","authors":["Bou Ali, M. N.","Lassoued, N.","Ben Amor, B.","Nasr, S."],"date":"2026-10-19","url":"https://www.medrxiv.org/content/10.64898/2026.10.15.26365634v1","site":"medRxiv","venue":null,"category":"endocrinology","fetchedAt":"2026-10-22T10:07:39"},{"id":"doi:10.64898/2026.10.16.26365710","title":"Deep Learning-Based Anatomical Level Classification of Craniofacial CT Slices: Comparison of CNN Architectures and Ensemble Methods","titleZh":"基于深度学习的颅面 CT 层面解剖分级：CNN 架构与集成方法比较","summaryZh":"用日本 10 所大学医院的 11769 张图像，训练深度学习模型把颅面 CT 层面分为 6 个解剖类别。","summaryEn":"Classifies craniofacial CT slices into six anatomical levels using 11,769 images from 10 Japanese university hospitals.","authors":["Nishimoto, S.","Nakajima, K.","Yoshiike, H.","Ishise, H."],"date":"2026-10-19","url":"https://www.medrxiv.org/content/10.64898/2026.10.16.26365710v1","site":"medRxiv","venue":null,"category":"radiology and imaging","fetchedAt":"2026-10-22T10:07:39"}]},{"id":"medical-imaging","slug":"medical-imaging","category":"life-medical","categoryName":{"zh":"生命与医学","en":"Life & medical sciences"},"name":{"zh":"医学影像与病理","en":"Medical imaging & pathology"},"classification":{"oecdFos":"3.2","moe":"100207 影像医学与核医学"},"level":3,"levelKey":"transforming","levelLabel":{"zh":"系统改造","en":"Systemic transformation"},"trend":"flat","reason":{"zh":"影像是 AI 落地最深的医学分支：FDA AI 器械清单中放射科占大多数；瑞典 MASAI 随机试验显示 AI 辅助乳腺筛查可安全减少约 44% 阅片量（Lancet Oncology 2023）。","en":"Imaging is medicine's most AI-penetrated branch: radiology dominates FDA's AI device list; the MASAI RCT showed AI-supported mammography safely cut reading workload by ~44% (Lancet Oncology 2023)."},"reasonSources":[{"url":"https://www.fda.gov/medical-devices/artificial-intelligence-enabled-medical-devices/list-artificial-intelligence-enabled-medical-devices","linkCheck":"ok"},{"url":"https://www.thelancet.com/journals/lanonc/article/PIIS1470-2045(23)00298-X/fulltext","linkCheck":"site-blocks-bots"}],"directions":{"zh":["影像筛查与分诊","数字病理基础模型","报告自动生成"],"en":["Screening & triage","Pathology foundation models","Report generation"]},"stats":{"milestones":9,"milestonesSince2025":4,"milestones2026":4,"milestonesLast6Months":3,"news":12,"latestProgress":"2026-10-20","latestMilestone":"2026-10-20","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":0,"leaps":{"ai":0,"field":1},"latestPapers":8,"keyPapers":6},"milestones":[{"date":"2016-12-13","title":{"zh":"深度学习检测糖尿病视网膜病变达到眼科医生水平（JAMA）","en":"Deep learning for diabetic retinopathy detection at ophthalmologist level (JAMA)"},"url":"https://doi.org/10.1001/jama.2016.17216","source":"JAMA","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"首批在顶级医学期刊证明 AI 读片达专科医生水平的研究之一","en":"Among the first to show specialist-level AI reading in a top medical journal"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首批在顶级医学期刊证明 AI 读片达专科医生水平的研究之一","en":"Among the first to show specialist-level AI reading in a top medical journal"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首批在顶级医学期刊证明 AI 读片达专科医生水平的研究之一","en":"Among the first to show specialist-level AI reading in a top medical journal"}},"impactNotes":[],"verified":{"id":"10.1001/jama.2016.17216","title":"Development and Validation of a Deep Learning Algorithm for Detection of Diabetic Retinopathy in Retinal Fundus Photographs","venue":"JAMA","date":"2016-12-13"},"addedIn":"v2"},{"date":"2017-01-25","title":{"zh":"皮肤癌分类达到皮肤科医生水平（Nature，Esteva 等）","en":"Dermatologist-level classification of skin cancer (Nature)"},"url":"https://doi.org/10.1038/nature21056","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"引发医学影像 AI 研究热潮的标志性论文","en":"Landmark paper that launched the medical-imaging AI wave"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"引发医学影像 AI 研究热潮的标志性论文","en":"Landmark paper that launched the medical-imaging AI wave"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"引发医学影像 AI 研究热潮的标志性论文","en":"Landmark paper that launched the medical-imaging AI wave"}},"impactNotes":[],"verified":{"id":"10.1038/nature21056","title":"Dermatologist-level classification of skin cancer with deep neural networks","venue":"Nature","date":"2017-01-25"},"addedIn":"v2"},{"date":"2020-01-01","title":{"zh":"国际评估：AI 乳腺癌筛查系统在英美数据上超过放射科医生（Nature）","en":"International evaluation of an AI system for breast cancer screening (Nature)"},"url":"https://doi.org/10.1038/s41586-019-1799-6","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"跨国数据上假阳性和假阴性均低于放射科医生","en":"Fewer false positives and negatives than radiologists across two countries"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"跨国数据上假阳性和假阴性均低于放射科医生","en":"Fewer false positives and negatives than radiologists across two countries"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"跨国数据上假阳性和假阴性均低于放射科医生","en":"Fewer false positives and negatives than radiologists across two countries"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-019-1799-6","title":"International evaluation of an AI system for breast cancer screening","venue":"Nature","date":"2020-01-01"},"addedIn":"v2"},{"date":"2023-08-02","title":{"zh":"MASAI 随机试验：AI 辅助乳腺 X 线筛查（Lancet Oncology）","en":"MASAI RCT: AI-supported mammography screening (Lancet Oncology)"},"url":"https://www.thelancet.com/journals/lanonc/article/PIIS1470-2045(23)00298-X/fulltext","source":"Lancet Oncology","flags":[],"flagLabels":[],"linkCheck":"site-blocks-bots","datePrecision":"day","type":"clinical","typeLabel":{"zh":"临床","en":"Clinical"},"why":{"zh":"随机试验中期结果：AI 辅助筛查减少约 44% 阅片量且检出率不降","en":"Interim RCT: 44% less reading workload, no drop in detection"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"随机试验中期结果：AI 辅助筛查减少约 44% 阅片量且检出率不降","en":"Interim RCT: 44% less reading workload, no drop in detection"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"随机试验中期结果：AI 辅助筛查减少约 44% 阅片量且检出率不降","en":"Interim RCT: 44% less reading workload, no drop in detection"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2024-03-19","title":{"zh":"UNI：计算病理通用自监督基础模型（Nature Medicine）","en":"UNI: a general-purpose self-supervised model for computational pathology (Nature Medicine)"},"url":"https://doi.org/10.1038/s41591-024-02857-3","source":"Nature Medicine","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"病理基础模型在 30 多项任务上超过此前方法","en":"Pathology foundation model beats prior methods on 30+ tasks"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"病理基础模型在 30 多项任务上超过此前方法","en":"Pathology foundation model beats prior methods on 30+ tasks"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"病理基础模型在 30 多项任务上超过此前方法","en":"Pathology foundation model beats prior methods on 30+ tasks"}},"impactNotes":[],"verified":{"id":"10.1038/s41591-024-02857-3","title":"Towards a general-purpose foundation model for computational pathology","venue":"Nature Medicine","date":"2024-03-19"},"addedIn":"v2"},{"date":"2026-01","title":{"zh":"MASAI 终期结果：AI 辅助乳腺筛查间期癌非劣效，敏感度 80.5% vs 73.8%（The Lancet）","en":"MASAI final results: AI-supported screening non-inferior on interval cancers, sensitivity 80.5% vs 73.8% (The Lancet)"},"url":"https://doi.org/10.1016/s0140-6736(25)02464-x","source":"The Lancet","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"month","type":"clinical","typeLabel":{"zh":"临床","en":"Clinical"},"why":{"zh":"10.6 万名女性的随机试验首次以间期癌这一硬终点证明 AI 筛查可安全替代双人阅片，作者建议可考虑临床实施","en":"RCT of 105,934 women shows non-inferiority on the hard endpoint of interval cancers"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"10.6 万名女性的随机试验首次以间期癌这一硬终点证明 AI 筛查可安全替代双人阅片，作者建议可考虑临床实施","en":"RCT of 105,934 women shows non-inferiority on the hard endpoint of interval cancers"}},"fieldImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"10.6 万名女性的随机试验首次以间期癌这一硬终点证明 AI 筛查可安全替代双人阅片，作者建议可考虑临床实施","en":"RCT of 105,934 women shows non-inferiority on the hard endpoint of interval cancers"}},"impactNotes":[],"verified":{"id":"10.1016/s0140-6736(25)02464-x","title":"Interval cancer, sensitivity, and specificity comparing AI-supported mammography screening with standard double reading without AI in the MASAI study: a randomised, controlled, non-inferiority, single-blinded, population-based, screening-accuracy trial","venue":"The Lancet","date":"2026-01"},"addedIn":"v2"},{"date":"2026-09","title":{"zh":"APHINITY 随机试验二次分析：比较人工、数字与 AI 肿瘤浸润淋巴细胞评分（Lancet Oncology）","en":"Manual, digital and AI TIL scoring: secondary analysis of the APHINITY RCT (Lancet Oncology)"},"url":"https://doi.org/10.1016/s1470-2045(26)00247-0","source":"The Lancet Oncology","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"month","type":"clinical","typeLabel":{"zh":"临床","en":"Clinical"},"why":{"zh":"AI：AI 病理评分在大型随机试验样本中与人工评分对照；学科：为乳腺癌 TIL 生物标志物的 AI 标准化提供试验级证据","en":"AI: AI pathology scoring benchmarked against manual scoring in a large RCT cohort; Field: trial-level evidence for standardising TIL biomarkers"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：AI 病理评分在大型随机试验样本中与人工评分对照；学科：为乳腺癌 TIL 生物标志物的 AI 标准化提供试验级证据","en":"AI: AI pathology scoring benchmarked against manual scoring in a large RCT cohort; Field: trial-level evidence for standardising TIL biomarkers"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：AI 病理评分在大型随机试验样本中与人工评分对照；学科：为乳腺癌 TIL 生物标志物的 AI 标准化提供试验级证据","en":"AI: AI pathology scoring benchmarked against manual scoring in a large RCT cohort; Field: trial-level evidence for standardising TIL biomarkers"}},"impactNotes":[],"verified":{"id":"10.1016/s1470-2045(26)00247-0","title":"Manual, digital, and AI tumour-infiltrating lymphocyte scoring: a secondary analysis of the APHINITY randomised trial","venue":"The Lancet Oncology","date":"2026-09"},"addedIn":"v2"},{"date":"2026-09-04","title":{"zh":"FDA 清单累计 1614 个 AI 器械，放射科占多数","en":"FDA list reaches 1,614 AI devices, mostly radiology"},"url":"https://www.fda.gov/medical-devices/artificial-intelligence-enabled-medical-devices/list-artificial-intelligence-enabled-medical-devices","source":"FDA","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"regulatory","typeLabel":{"zh":"监管","en":"Regulatory"},"why":{"zh":"FDA AI 器械清单累计 1614 个，反映临床部署规模","en":"1,614 AI devices on the FDA list"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"FDA AI 器械清单累计 1614 个，反映临床部署规模","en":"1,614 AI devices on the FDA list"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"FDA AI 器械清单累计 1614 个，反映临床部署规模","en":"1,614 AI devices on the FDA list"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-10-20","title":{"zh":"Sophont 发布 Holon-S1：从常规 H&E 切片预测空间基因表达与通路活性的泛癌病理基础模型（仅限科研）","en":"Sophont launches Holon-S1, a pan-cancer pathology model predicting spatial gene expression from H&E (research use)"},"url":"https://www.prnewswire.com/news-releases/sophont-launches-holon-s1-to-reveal-tumor-biology-from-routine-pathology-slides-302914452.html","source":"PR Newswire","flags":["vendor"],"flagLabels":[{"zh":"厂商自述","en":"Vendor claim"}],"linkCheck":"ok","datePrecision":"day","type":"model","typeLabel":{"zh":"模型发布","en":"Model"},"why":{"zh":"AI：厂商称在 16 种器官上超过主流模型；学科：尚无临床验证，计划开展回顾性与前瞻性研究","en":"AI: vendor reports gains across 16 organs; Field: no clinical validation yet"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：厂商称在 16 种器官上超过主流模型；学科：尚无临床验证，计划开展回顾性与前瞻性研究","en":"AI: vendor reports gains across 16 organs; Field: no clinical validation yet"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：厂商称在 16 种器官上超过主流模型；学科：尚无临床验证，计划开展回顾性与前瞻性研究","en":"AI: vendor reports gains across 16 organs; Field: no clinical validation yet"}},"impactNotes":[],"verified":null,"addedIn":"v2"}],"papersRepos":[{"title":"Interval cancer, sensitivity, and specificity comparing AI-supported mammography screening with standard double reading without AI in the MASAI study: a randomised, controlled, non-inferiority, single-blinded, population-based, screening-accuracy trial","url":"https://doi.org/10.1016/s0140-6736(25)02464-x","date":"2026-01","kind":"paper","linkCheck":"ok-crossref"},{"title":"Development and Validation of a Deep Learning Algorithm for Detection of Diabetic Retinopathy in Retinal Fundus Photographs","url":"https://doi.org/10.1001/jama.2016.17216","date":"2016-12-13","kind":"paper","linkCheck":"ok-crossref"},{"title":"Dermatologist-level classification of skin cancer with deep neural networks","url":"https://doi.org/10.1038/nature21056","date":"2017-01-25","kind":"paper","linkCheck":"ok-crossref"},{"title":"International evaluation of an AI system for breast cancer screening","url":"https://doi.org/10.1038/s41586-019-1799-6","date":"2020-01-01","kind":"paper","linkCheck":"ok-crossref"},{"title":"Towards a general-purpose foundation model for computational pathology","url":"https://doi.org/10.1038/s41591-024-02857-3","date":"2024-03-19","kind":"paper","linkCheck":"ok-crossref"},{"title":"Manual, digital, and AI tumour-infiltrating lymphocyte scoring: a secondary analysis of the APHINITY randomised trial","url":"https://doi.org/10.1016/s1470-2045(26)00247-0","date":"2026-09","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"FDA","url":"https://www.fda.gov"},{"name":"Paige","url":"https://paige.ai"}],"news":[{"id":"939922e31753","date":"2026-10-20","title":"GliomaINK: A Tumor and Anatomy Conditioned Flow Matching Framework for 3D MRI Sequence Synthesis in Glioma Patients","url":"https://arxiv.org/abs/2610.24044","source":"arXiv q-bio","lang":"en","kind":"paper","summaryZh":"arXiv q-bio于2026-10-20发布的论文，涉及AI 生物/医药、AI 医疗、AI 软件开发，原题《GliomaINK: A Tumor and Anatomy Conditioned Flow Matching Framework for 3D MRI Sequence Syn》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"86a9549a04fd","date":"2026-10-20","title":"What Language Adds, and at What Cost: Four Controlled Probes of Pathology Vision-Language Models","url":"https://arxiv.org/abs/2610.26805","source":"arXiv q-bio","lang":"en","kind":"paper","summaryZh":"arXiv q-bio于2026-10-20发布的论文，涉及AI 生物/医药、AI 医疗、AI 软件开发，原题《What Language Adds, and at What Cost: Four Controlled Probes of Pathology Vision-Language 》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"7b8383d5ab45","date":"2026-10-20","title":"WILSON - Multi-magnification composites as a foundation for patient-level pathology analysis and diagnostic text generation","url":"https://arxiv.org/abs/2609.25123","source":"arXiv q-bio","lang":"en","kind":"paper","summaryZh":"arXiv q-bio于2026-10-20发布的论文，涉及AI 生物/医药、AI 医疗，原题《WILSON - Multi-magnification composites as a foundation for patient-level pathology analys》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"df25bb5da50f","date":"2026-10-18","title":"Multi-task deep learning framework for brain tumor classification and segmentation using ensemble feature extraction","url":"https://www.nature.com/articles/s41598-026-75592-z","source":"Nature 学科·计算生物","lang":"en","kind":"news","summaryZh":"Nature 学科·计算生物于2026-10-18发布的文章，涉及AI 生物/医药、AI 软件开发，原题《Multi-task deep learning framework for brain tumor classification and segmentation using e》，详见原文链接。","via":"collector-run-20261020","linkCheck":"from-feed"},{"id":"c2b6a51cc051","date":"2026-10-16","title":"Aarthi Scans & Koita Centre for Digital Health Sign MoU for R&D Collaboration in AI-based Medical Imaging","url":"https://www.digitalhealthnews.com/aarthi-scans-koita-centre-for-digital-health-sign-mou-for-r-d-collaboration-in-ai-based-medical-imaging","source":"Digital Health News","lang":"en","kind":"news","summaryZh":"Digital Health News于2026-10-16发布的文章，涉及AI 医疗、AI 教育，原题《Aarthi Scans & Koita Centre for Digital Health Sign MoU for R&D Collaboration in AI-based 》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"dddee25f8efe","date":"2026-10-16","title":"Radiology department uses AI to identify lung cancer screening patients in EHR","url":"https://radiologybusiness.com/topics/artificial-intelligence/radiology-department-uses-ai-identify-lung-cancer-screening-patients-ehr","source":"Radiology Business","lang":"en","kind":"news","summaryZh":"Radiology Business于2026-10-16发布的文章，涉及AI 医疗、AI 教育，原题《Radiology department uses AI to identify lung cancer screening patients in EHR》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"4ceee8b5fb24","date":"2026-10-15","title":"A 3-step buyer’s guide for evaluating radiology AI solutions","url":"https://radiologybusiness.com/topics/artificial-intelligence/3-step-buyers-guide-evaluating-radiology-ai-solutions","source":"Radiology Business","lang":"en","kind":"news","summaryZh":"Radiology Business于2026-10-15发布的文章，涉及AI 医疗，原题《A 3-step buyer’s guide for evaluating radiology AI solutions》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"6147badce51f","date":"2026-10-15","title":"Sirona Medical Secures $40M to Scale Cloud-Native Radiology and Cortex AI","url":"https://hitconsultant.net/2026/10/15/sirona-medical-raises-40m-series-d-avidity-partners-rados-cloud-radiology-cortex-ai/","source":"HIT Consultant","lang":"en","kind":"news","summaryZh":"HIT Consultant于2026-10-16发布的文章，涉及AI 医疗、AI 金融、AI 商业融资，原题《Sirona Medical Secures $40M to Scale Cloud-Native Radiology and Cortex AI》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"276dee532a65","date":"2026-10-14","title":"Researchers develop way of adapting adult radiology AI models to pediatric population","url":"https://radiologybusiness.com/topics/artificial-intelligence/researchers-develop-way-adapting-adult-radiology-ai-models-pediatric-population","source":"Radiology Business","lang":"en","kind":"news","summaryZh":"Radiology Business于2026-10-14发布的文章，涉及AI 医疗，原题《Researchers develop way of adapting adult radiology AI models to pediatric population》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"4eaf149e5826","date":"2026-10-14","title":"Agilent to Acquire Proscia for Up to $270 Mn to Expand AI-Enabled Digital Pathology","url":"https://www.digitalhealthnews.com/agilent-to-acquire-proscia-for-up-to-270-mn-to-expand-ai-enabled-digital-pathology","source":"Digital Health News","lang":"en","kind":"news","summaryZh":"Digital Health News于2026-10-14发布的文章，涉及AI 医疗、AI 商业融资，原题《Agilent to Acquire Proscia for Up to $270 Mn to Expand AI-Enabled Digital Pathology》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"9cbea024022a","date":"2026-10-13","title":"AI matches radiologists at prostate cancer detection via MRI","url":"https://radiologybusiness.com/topics/artificial-intelligence/ai-matches-radiologists-prostate-cancer-detection-mri","source":"Radiology Business","lang":"en","kind":"news","summaryZh":"Radiology Business于2026-10-14发布的文章，涉及AI 医疗，原题《AI matches radiologists at prostate cancer detection via MRI》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"9afe0d61f96c","date":"2026-10-12","title":"Digital Pathology in India: Infrastructure Readiness & the Next Wave of Adoption","url":"https://www.digitalhealthnews.com/digital-pathology-in-india-infrastructure-readiness-the-next-wave-of-adoption","source":"Digital Health News","lang":"en","kind":"news","summaryZh":"Digital Health News于2026-10-12发布的文章，涉及AI 医疗，原题《Digital Pathology in India: Infrastructure Readiness & the Next Wave of Adoption》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"}],"review":{"lastReviewed":"2026-10-21","cycleDays":60,"nextReview":"2026-12-20"},"urls":{"zh":"/zh-hans/disciplines/medical-imaging/","en":"/en/disciplines/medical-imaging/"},"dimensions":{"method":{"score":3,"label":{"zh":"研究方法","en":"Research method"},"evidence":"深度学习已是影像和病理研究的主流方法","source":"https://doi.org/10.1038/s41591-024-02857-3","weight":0.3},"discovery":{"score":3,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"多项大规模随机试验证实 AI 辅助筛查可行，MASAI 终期结果发表于 Lancet（2026）","source":"https://doi.org/10.1016/s0140-6736(25)02464-x","weight":0.35},"adoption":{"score":3,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"FDA AI 器械中放射科占大多数，欧洲多国筛查项目引入 AI","source":"https://www.fda.gov/medical-devices/software-medical-device-samd/artificial-intelligence-enabled-medical-devices","weight":0.25},"education":{"score":1,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到住院医师培训改革的公开证据","source":null,"weight":0.1}},"compositeScore":2.8,"levelCapped":false,"levelChange":{"from":3,"to":3,"direction":"same","reason":null},"biggestBreakthrough":{"date":"2026-01","title":{"zh":"MASAI 终期结果：AI 辅助乳腺筛查间期癌非劣效，敏感度 80.5% vs 73.8%（The Lancet）","en":"MASAI final results: AI-supported screening non-inferior on interval cancers, sensitivity 80.5% vs 73.8% (The Lancet)"},"url":"https://doi.org/10.1016/s0140-6736(25)02464-x","aiImpact":"major","fieldImpact":"leap","why":{"zh":"10.6 万名女性的随机试验首次以间期癌这一硬终点证明 AI 筛查可安全替代双人阅片，作者建议可考虑临床实施","en":"RCT of 105,934 women shows non-inferiority on the hard endpoint of interval cancers"}},"keyPapers":[{"title":"Interval cancer, sensitivity, and specificity comparing AI-supported mammography screening with standard double reading without AI in the MASAI study: a randomised, controlled, non-inferiority, single-blinded, population-based, screening-accuracy trial","titleZh":"MASAI 终期结果：AI 辅助乳腺筛查间期癌非劣效，敏感度 80.5% vs 73.8%（The Lancet）","id":"10.1016/s0140-6736(25)02464-x","url":"https://doi.org/10.1016/s0140-6736(25)02464-x","venue":"The Lancet","date":"2026-01","why":{"zh":"10.6 万名女性的随机试验首次以间期癌这一硬终点证明 AI 筛查可安全替代双人阅片，作者建议可考虑临床实施","en":"RCT of 105,934 women shows non-inferiority on the hard endpoint of interval cancers"},"fieldImpact":"leap","aiImpact":"major"},{"title":"Development and Validation of a Deep Learning Algorithm for Detection of Diabetic Retinopathy in Retinal Fundus Photographs","titleZh":"深度学习检测糖尿病视网膜病变达到眼科医生水平（JAMA）","id":"10.1001/jama.2016.17216","url":"https://doi.org/10.1001/jama.2016.17216","venue":"JAMA","date":"2016-12-13","why":{"zh":"首批在顶级医学期刊证明 AI 读片达专科医生水平的研究之一","en":"Among the first to show specialist-level AI reading in a top medical journal"},"fieldImpact":"major","aiImpact":"major"},{"title":"Dermatologist-level classification of skin cancer with deep neural networks","titleZh":"皮肤癌分类达到皮肤科医生水平（Nature，Esteva 等）","id":"10.1038/nature21056","url":"https://doi.org/10.1038/nature21056","venue":"Nature","date":"2017-01-25","why":{"zh":"引发医学影像 AI 研究热潮的标志性论文","en":"Landmark paper that launched the medical-imaging AI wave"},"fieldImpact":"major","aiImpact":"major"},{"title":"International evaluation of an AI system for breast cancer screening","titleZh":"国际评估：AI 乳腺癌筛查系统在英美数据上超过放射科医生（Nature）","id":"10.1038/s41586-019-1799-6","url":"https://doi.org/10.1038/s41586-019-1799-6","venue":"Nature","date":"2020-01-01","why":{"zh":"跨国数据上假阳性和假阴性均低于放射科医生","en":"Fewer false positives and negatives than radiologists across two countries"},"fieldImpact":"major","aiImpact":"major"},{"title":"Towards a general-purpose foundation model for computational pathology","titleZh":"UNI：计算病理通用自监督基础模型（Nature Medicine）","id":"10.1038/s41591-024-02857-3","url":"https://doi.org/10.1038/s41591-024-02857-3","venue":"Nature Medicine","date":"2024-03-19","why":{"zh":"病理基础模型在 30 多项任务上超过此前方法","en":"Pathology foundation model beats prior methods on 30+ tasks"},"fieldImpact":"major","aiImpact":"major"},{"title":"Manual, digital, and AI tumour-infiltrating lymphocyte scoring: a secondary analysis of the APHINITY randomised trial","titleZh":"APHINITY 随机试验二次分析：比较人工、数字与 AI 肿瘤浸润淋巴细胞评分（Lancet Oncology）","id":"10.1016/s1470-2045(26)00247-0","url":"https://doi.org/10.1016/s1470-2045(26)00247-0","venue":"The Lancet Oncology","date":"2026-09","why":{"zh":"AI：AI 病理评分在大型随机试验样本中与人工评分对照；学科：为乳腺癌 TIL 生物标志物的 AI 标准化提供试验级证据","en":"AI: AI pathology scoring benchmarked against manual scoring in a large RCT cohort; Field: trial-level evidence for standardising TIL biomarkers"},"fieldImpact":"major","aiImpact":"incremental"}],"latest_papers":[{"id":"arXiv:2610.28873","title":"Improved three component IVIM fitting using physics-informed neural network with multimodal spatial prior","titleZh":"用多模态空间先验的物理信息神经网络改进三成分 IVIM 拟合","summaryZh":"结合 T1、FLAIR 解剖图像与交叉注意力等机制，缓解体素内不相干运动（IVIM）模型参数退化，提升拟合稳健性。","summaryEn":"A physics-informed network using anatomical priors and cross-attention mitigates parameter degeneracy in three-component IVIM fitting.","authors":["Nhat Hoang","Abrar Faiyaz","Emmanuel A. Mensah","Md Nasir Uddin"],"date":"2026-10-20","url":"https://arxiv.org/abs/2610.28873","site":"arXiv","venue":null,"category":"eess.IV","fetchedAt":"2026-10-22T10:07:47"},{"id":"arXiv:2610.28179","title":"CAPS: Clinically-Verifiable Anatomy-Prioritized Synthesis for Missing MRI Imputation in Brain Tumor Segmentation","titleZh":"CAPS：可临床核验、解剖优先的缺失 MRI 模态合成用于脑肿瘤分割","summaryZh":"针对多参数 MRI 缺失模态，用保持解剖特征与拓扑边界的混合损失合成图像，在 BraTS 成人胶质瘤数据上兼顾合成质量与分割性能（平均 Dice 82.38%）。","summaryEn":"Anatomy-prioritised synthesis of missing MRI modalities for brain-tumour segmentation (mean Dice 82.38% on BraTS).","authors":["Cecilia Diana-Albelda","Arthur Longuefosse","Mario Rubén Paz Campos","Álvaro García-Martín"],"date":"2026-10-20","url":"https://arxiv.org/abs/2610.28179","site":"arXiv","venue":null,"category":"eess.IV","fetchedAt":"2026-10-22T10:07:47"},{"id":"arXiv:2610.26586","title":"Scaling Vision Foundation Models for Toxicologic Pathology","titleZh":"扩展毒理病理视觉基础模型","summaryZh":"TOX-FM 在来自 6 个物种、60 多个器官、14000 多项毒理研究的 100 多万张全切片图像上训练，并在 18 项非临床和 9 项临床任务上评估。","summaryEn":"TOX-FM, trained on 1M+ toxicology whole-slide images across six species, evaluated on 27 tasks.","authors":["Jessica Loo","Theodoros Pissas","Byunghak Kang","Vincent A. Ella"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.26586","site":"arXiv","venue":null,"category":"cs.CV","fetchedAt":"2026-10-22T10:07:47"},{"id":"doi:10.64898/2026.01.19.26344376","title":"Evaluating the Interplay and Impact of Quality and Technical Variations on Performance of AI Models in Multi-Institutional Prostate MRI Datasets","titleZh":"多中心前列腺 MRI 中质量与技术差异对 AI 模型表现的影响","summaryZh":"分析 7 家机构 568 例前列腺 MRI，发现目测质量高并不足以保证 AI 模型表现一致，存在肉眼难以发现的采集技术影响。","summaryEn":"568 prostate MRIs from 7 sites show visually high-quality scans do not guarantee consistent AI performance.","authors":["Azamat, S.","Singh, A.","Midya, A.","Zhang, Z."],"date":"2026-10-19","url":"https://www.medrxiv.org/content/10.64898/2026.01.19.26344376v1","site":"medRxiv","venue":null,"category":"radiology and imaging","fetchedAt":"2026-10-22T10:07:47"},{"id":"arXiv:2610.24046","title":"LUMINR: Intensity-conditioned Meta-learned INRs for Accelerated Medical Image Registration","titleZh":"LUMINR：强度条件元学习隐式神经表示加速医学图像配准","summaryZh":"把局部强度和梯度特征作为条件，加速隐式神经表示在每对图像上的形变配准拟合。","summaryEn":"Locally conditioned meta-learned INRs accelerate deformable medical image registration.","authors":["Damian Rynczak","Marek Wodzinski"],"date":"2026-10-16","url":"https://arxiv.org/abs/2610.24046","site":"arXiv","venue":null,"category":"eess.IV","fetchedAt":"2026-10-22T10:07:47"},{"id":"arXiv:2610.24044","title":"GliomaINK: A Tumor and Anatomy Conditioned Flow Matching Framework for 3D MRI Sequence Synthesis in Glioma Patients","titleZh":"GliomaINK：基于肿瘤与解剖条件的流匹配框架合成胶质瘤 3D MRI 序列","summaryZh":"提出解剖引导的条件流匹配方法，为检查不完整的胶质瘤患者合成缺失的 MRI 序列，同时保留肿瘤特征。","summaryEn":"Anatomy-guided conditional flow matching synthesises missing 3D MRI sequences in glioma while preserving tumour features.","authors":["Moinak Bhattacharya","Annie Singh","Mohammad Chaudhry","Prateek Prasanna"],"date":"2026-10-16","url":"https://arxiv.org/abs/2610.24044","site":"arXiv","venue":null,"category":"eess.IV","fetchedAt":"2026-10-22T10:07:47"},{"id":"arXiv:2610.24034","title":"Implicit Neural Models for Direct Multi-Shell Diffusion MRI Signal Estimation","titleZh":"直接估计多壳层弥散 MRI 信号的隐式神经模型","summaryZh":"用小波激活的隐式神经表示连续表示整个弥散信号，实现超分辨率和 q 空间插值。","summaryEn":"Implicit neural representation of full diffusion-MRI signals for super-resolution and q-space interpolation.","authors":["Sanna Persson","Fabian Leander Sinzinger","Christoffer Olsson","Rodrigo Moreno"],"date":"2026-10-16","url":"https://arxiv.org/abs/2610.24034","site":"arXiv","venue":null,"category":"eess.IV","fetchedAt":"2026-10-22T10:07:47"},{"id":"doi:10.64898/2026.01.09.26343779","title":"Multimodal profiling for prediction of primary resistance to anti-PD-(L)1 therapy in advanced non-small-cell lung cancer: the prospective PIONeeR biomarkers study","titleZh":"多模态分析预测晚期非小细胞肺癌抗 PD-(L)1 原发耐药：前瞻性 PIONeeR 研究","summaryZh":"在多中心前瞻性队列中整合多模态生物标志物，构建预测免疫治疗原发耐药的模型，弥补 PD-L1 和 TMB 区分力不足。","summaryEn":"A prospective multicentre study building multimodal models to predict primary resistance to anti-PD-(L)1 therapy in advanced NSCLC.","authors":["Barlesi, F.","Monville, F.","Vely, F.","Greillier, L."],"date":"2026-10-16","url":"https://www.medrxiv.org/content/10.64898/2026.01.09.26343779v3","site":"medRxiv","venue":null,"category":"oncology","fetchedAt":"2026-10-22T10:07:47"}]},{"id":"public-health","slug":"public-health","category":"life-medical","categoryName":{"zh":"生命与医学","en":"Life & medical sciences"},"name":{"zh":"公共卫生与流行病学","en":"Public health & epidemiology"},"classification":{"oecdFos":"3.3","moe":"1004 公共卫生与预防医学"},"level":2,"levelKey":"breakthrough","levelLabel":{"zh":"局部突破","en":"Local breakthroughs"},"trend":"flat","reason":{"zh":"EVEscape 仅凭疫情前数据就预测出 SARS-CoV-2 逃逸突变（Nature 2023）；WHO 2024 年发布大型多模态模型治理指南。整体仍以监测和辅助分析为主。","en":"EVEscape forecast SARS-CoV-2 escape mutations from pre-pandemic data (Nature 2023); WHO issued LMM governance guidance in 2024. Use is mostly surveillance and analysis support."},"reasonSources":[{"url":"https://www.nature.com/articles/s41586-023-06617-0","linkCheck":"ok"},{"url":"https://www.who.int/news/item/18-01-2024-who-releases-ai-ethics-and-governance-guidance-for-large-multi-modal-models","linkCheck":"ok"}],"directions":{"zh":["病原体进化预测","疫情监测","健康行为干预"],"en":["Pathogen evolution forecasting","Outbreak surveillance","Behavioural interventions"]},"stats":{"milestones":10,"milestonesSince2025":6,"milestones2026":5,"milestonesLast6Months":3,"news":7,"latestProgress":"2026-10-20","latestMilestone":"2026-10-12","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":0,"leaps":{"ai":0,"field":0},"latestPapers":5,"keyPapers":7},"milestones":[{"date":"2009-02-19","title":{"zh":"谷歌流感趋势：用搜索查询数据检测流感流行（Nature）","en":"Detecting influenza epidemics using search query data (Nature)"},"url":"https://doi.org/10.1038/nature07634","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"开创用大数据和机器学习做实时疫情监测","en":"Pioneered big-data real-time epidemic surveillance"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"开创用大数据和机器学习做实时疫情监测","en":"Pioneered big-data real-time epidemic surveillance"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"开创用大数据和机器学习做实时疫情监测","en":"Pioneered big-data real-time epidemic surveillance"}},"impactNotes":[],"verified":{"id":"10.1038/nature07634","title":"Detecting influenza epidemics using search engine query data","venue":"Nature","date":"2009-02-19"},"addedIn":"v2"},{"date":"2014-03-14","title":{"zh":"《谷歌流感的寓言》：大数据预测高估流感峰值的教训（Science）","en":"The parable of Google Flu: traps in big data analysis (Science)"},"url":"https://doi.org/10.1126/science.1248506","source":"Science","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"重要反面案例，确立公共卫生 AI 需持续验证的原则","en":"Key cautionary case for validating public-health AI"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"重要反面案例，确立公共卫生 AI 需持续验证的原则","en":"Key cautionary case for validating public-health AI"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"重要反面案例，确立公共卫生 AI 需持续验证的原则","en":"Key cautionary case for validating public-health AI"}},"impactNotes":[],"verified":{"id":"10.1126/science.1248506","title":"The Parable of Google Flu: Traps in Big Data Analysis","venue":"Science","date":"2014-03-14"},"addedIn":"v2"},{"date":"2023-10-11","title":{"zh":"EVEscape 预测病毒免疫逃逸突变（Nature）","en":"EVEscape predicts viral immune-escape mutations (Nature)"},"url":"https://www.nature.com/articles/s41586-023-06617-0","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"仅用疫情前数据就预测出 SARS-CoV-2 关键逃逸突变","en":"Predicted SARS-CoV-2 escape mutations from pre-pandemic data"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"仅用疫情前数据就预测出 SARS-CoV-2 关键逃逸突变","en":"Predicted SARS-CoV-2 escape mutations from pre-pandemic data"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"仅用疫情前数据就预测出 SARS-CoV-2 关键逃逸突变","en":"Predicted SARS-CoV-2 escape mutations from pre-pandemic data"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-023-06617-0","title":"Learning from prepandemic data to forecast viral escape","venue":"Nature","date":"2023-10-11"},"addedIn":"v1"},{"date":"2024-01-18","title":{"zh":"WHO 发布大型多模态模型伦理与治理指南","en":"WHO guidance on large multimodal models"},"url":"https://www.who.int/news/item/18-01-2024-who-releases-ai-ethics-and-governance-guidance-for-large-multi-modal-models","source":"WHO","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"policy","typeLabel":{"zh":"政策","en":"Policy"},"why":{"zh":"WHO 发布大模型治理指南","en":"WHO governance guidance"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"WHO 发布大模型治理指南","en":"WHO governance guidance"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"WHO 发布大模型治理指南","en":"WHO governance guidance"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2025-02-19","title":{"zh":"综述：人工智能用于传染病流行建模（Nature）","en":"Artificial intelligence for modelling infectious disease epidemics (Nature)"},"url":"https://doi.org/10.1038/s41586-024-08564-w","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"系统评估 AI 在流行病学建模中的能力与局限","en":"Systematic assessment of AI for epidemic modelling"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"系统评估 AI 在流行病学建模中的能力与局限","en":"Systematic assessment of AI for epidemic modelling"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"系统评估 AI 在流行病学建模中的能力与局限","en":"Systematic assessment of AI for epidemic modelling"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-024-08564-w","title":"Artificial intelligence for modelling infectious disease epidemics","venue":"Nature","date":"2025-02-19"},"addedIn":"v2"},{"date":"2026-03-19","title":{"zh":"WHO 柏林中心总结 2025 成果：EIOS 用 AI 实时识别异常卫生事件，原型化 AI 大流行模拟器","en":"WHO Berlin Hub: EIOS uses AI to flag unusual health events; AI pandemic simulator prototyped"},"url":"https://www.who.int/news/item/19-03-2026-german-health-minister-nina-warken-and-who-hub-in-berlin-highlight-2025-achievements","source":"WHO","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"deployment","typeLabel":{"zh":"落地部署","en":"Deployment"},"why":{"zh":"AI 进入全球疫情情报的业务系统","en":"AI in WHO’s operational epidemic intelligence"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI 进入全球疫情情报的业务系统","en":"AI in WHO’s operational epidemic intelligence"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI 进入全球疫情情报的业务系统","en":"AI in WHO’s operational epidemic intelligence"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-03-20","title":{"zh":"epiFFORMA：用合成疫情数据训练的病种无关集成预测方法（Nat. Commun.）","en":"epiFFORMA: disease-agnostic ensemble forecasting trained on synthetic outbreaks (Nat. Commun.)"},"url":"https://doi.org/10.1038/s41467-026-70937-8","source":"Nature Communications","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"新发疫情缺历史数据时也能做预测","en":"Forecasts emerging outbreaks without long histories"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"新发疫情缺历史数据时也能做预测","en":"Forecasts emerging outbreaks without long histories"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"新发疫情缺历史数据时也能做预测","en":"Forecasts emerging outbreaks without long histories"}},"impactNotes":[],"verified":{"id":"10.1038/s41467-026-70937-8","title":"A disease-agnostic approach to ensemble learning for infectious disease forecasting","venue":"Nature Communications","date":"2026-03-20"},"addedIn":"v2"},{"date":"2026-06","title":{"zh":"时间序列基础模型用于传染病流行预测与政策反事实评估（Epidemics）","en":"Foundation models for time-series forecasting and policy evaluation in infectious-disease epidemics (Epidemics)"},"url":"https://doi.org/10.1016/j.epidem.2026.100916","source":"Epidemics","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"month","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"基础模型在数据稀缺的短期疫情预测中有增益，但应补充而非取代传统模型","en":"Helps data-limited short-term forecasts; complements established models"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"基础模型在数据稀缺的短期疫情预测中有增益，但应补充而非取代传统模型","en":"Helps data-limited short-term forecasts; complements established models"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"基础模型在数据稀缺的短期疫情预测中有增益，但应补充而非取代传统模型","en":"Helps data-limited short-term forecasts; complements established models"}},"impactNotes":[],"verified":{"id":"10.1016/j.epidem.2026.100916","title":"Foundation models for time series forecasting and policy evaluation in infectious disease epidemics","venue":"Epidemics","date":"2026-06"},"addedIn":"v2"},{"date":"2026-09-28","title":{"zh":"Mantis：在模拟疫情上训练的机理型疾病预测基础模型，可跨病种、在数据稀缺地区预测（PNAS）","en":"Mantis: a foundation model for mechanistic disease forecasting (PNAS)"},"url":"https://www.pnas.org/doi/10.1073/pnas.2602542123","source":"PNAS","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"model","typeLabel":{"zh":"模型发布","en":"Model"},"why":{"zh":"AI：仅用模拟数据训练即可泛化到真实疫情；学科：为缺乏历史数据的新发疾病和地区提供开箱即用的预测工具","en":"AI: trained on simulations, generalises to real outbreaks; Field: off-the-shelf forecasts for data-poor diseases and regions"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：仅用模拟数据训练即可泛化到真实疫情；学科：为缺乏历史数据的新发疾病和地区提供开箱即用的预测工具","en":"AI: trained on simulations, generalises to real outbreaks; Field: off-the-shelf forecasts for data-poor diseases and regions"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：仅用模拟数据训练即可泛化到真实疫情；学科：为缺乏历史数据的新发疾病和地区提供开箱即用的预测工具","en":"AI: trained on simulations, generalises to real outbreaks; Field: off-the-shelf forecasts for data-poor diseases and regions"}},"impactNotes":[],"verified":{"id":"10.1073/pnas.2602542123","title":"Mantis: A foundation model for mechanistic disease forecasting","venue":"Proceedings of the National Academy of Sciences","date":"2026-09-28"},"addedIn":"v2"},{"date":"2026-10-12","title":{"zh":"WHO 大流行与流行病情报中心发布 2026–2030 战略：目标是到 2030 年各国能在 24 小时内发现新健康威胁，AI 驱动的 EIOS 已被 125 个以上会员国使用","en":"WHO Hub for Pandemic and Epidemic Intelligence launches 2026–2030 strategy targeting 24-hour detection; AI-based EIOS used by 125+ Member States"},"url":"https://www.who.int/news/item/12-10-2026-who-pandemic-hub-sets-ambitious-goal-to-detect-new-outbreaks-within-24-hours","source":"WHO","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"policy","typeLabel":{"zh":"政策","en":"Policy"},"why":{"zh":"AI：EIOS 每天用 AI 实时扫描数千个公开信息源识别疫情信号；学科：WHO 把 AI 开源情报列为全球流行病早期发现五大优先计划之一，由 4000 多名专家社区使用","en":"AI: EIOS scans thousands of open sources daily with AI. Field: WHO makes AI open-source intelligence one of five priority initiatives for early outbreak detection"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：EIOS 每天用 AI 实时扫描数千个公开信息源识别疫情信号；学科：WHO 把 AI 开源情报列为全球流行病早期发现五大优先计划之一，由 4000 多名专家社区使用","en":"AI: EIOS scans thousands of open sources daily with AI. Field: WHO makes AI open-source intelligence one of five priority initiatives for early outbreak detection"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：EIOS 每天用 AI 实时扫描数千个公开信息源识别疫情信号；学科：WHO 把 AI 开源情报列为全球流行病早期发现五大优先计划之一，由 4000 多名专家社区使用","en":"AI: EIOS scans thousands of open sources daily with AI. Field: WHO makes AI open-source intelligence one of five priority initiatives for early outbreak detection"}},"impactNotes":[],"verified":null,"addedIn":"v2-20261022pm"}],"papersRepos":[{"title":"Detecting influenza epidemics using search engine query data","url":"https://doi.org/10.1038/nature07634","date":"2009-02-19","kind":"paper","linkCheck":"ok-crossref"},{"title":"The Parable of Google Flu: Traps in Big Data Analysis","url":"https://doi.org/10.1126/science.1248506","date":"2014-03-14","kind":"paper","linkCheck":"ok-crossref"},{"title":"Learning from prepandemic data to forecast viral escape","url":"https://doi.org/10.1038/s41586-023-06617-0","date":"2023-10-11","kind":"paper","linkCheck":"ok-crossref"},{"title":"Artificial intelligence for modelling infectious disease epidemics","url":"https://doi.org/10.1038/s41586-024-08564-w","date":"2025-02-19","kind":"paper","linkCheck":"ok-crossref"},{"title":"Mantis: A foundation model for mechanistic disease forecasting","url":"https://doi.org/10.1073/pnas.2602542123","date":"2026-09-28","kind":"paper","linkCheck":"ok-crossref"},{"title":"A disease-agnostic approach to ensemble learning for infectious disease forecasting","url":"https://doi.org/10.1038/s41467-026-70937-8","date":"2026-03-20","kind":"paper","linkCheck":"ok-crossref"},{"title":"Foundation models for time series forecasting and policy evaluation in infectious disease epidemics","url":"https://doi.org/10.1016/j.epidem.2026.100916","date":"2026-06","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"WHO","url":"https://www.who.int"}],"news":[{"id":"266096f8043a","date":"2026-10-20","title":"High-risk antimicrobial resistance genomic traits in multidrug-resistant Salmonella across China’s food chain: projections to 2050","url":"https://www.nature.com/articles/s41467-026-78346-7","source":"Nature 学科·机器学习","lang":"en","kind":"news","summaryZh":"Nature 学科·机器学习于2026-10-20发布的文章，涉及AI 科研与调研、AI 生物/医药、AI 安全，原题《High-risk antimicrobial resistance genomic traits in multidrug-resistant Salmonella across》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"2f21dfa90ca9","date":"2026-10-15","title":"Tamil Nadu’s Digital Health Push: Strengthening the State’s Public Health System Through Connected Care","url":"https://www.digitalhealthnews.com/tamil-nadu-s-digital-health-push-strengthening-the-state-s-public-health-system-through-connected-care","source":"Digital Health News","lang":"en","kind":"news","summaryZh":"Digital Health News于2026-10-15发布的文章，涉及AI 医疗、AI 科研与调研、AI 金融，原题《Tamil Nadu’s Digital Health Push: Strengthening the State’s Public Health System Through C》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"41a2794b59f6","date":"2026-10-14","title":"Former Crick Ph.D. student loses COVID vaccine paper for fabrication","url":"https://retractionwatch.com/2026/10/14/former-crick-ph-d-student-loses-covid-vaccine-paper-for-fabrication/","source":"Retraction Watch","lang":"en","kind":"news","summaryZh":"Retraction Watch于2026-10-15发布的文章，涉及AI 科研与调研、AI 医疗、AI 教育，原题《Former Crick Ph.D. student loses COVID vaccine paper for fabrication》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"5d11e47ef18d","date":"2026-10-09","title":"Epidemiological modeling in the age of artificial intelligence","url":"https://www.nature.com/articles/s43588-026-01070-1","source":"Nature Computational Science","lang":"en","kind":"news","summaryZh":"AI 让流行病模型可以跨疫情学习；下一步难点是让模型学到传播机制，而不只是统计规律。","via":"research-samples-20261010","linkCheck":"from-feed"},{"id":"2e0aae10ded8","date":"2026-10-06","title":"Unlocking Earth AI’s planetary geospatial foundation models for global public health","url":"https://research.google/blog/earth-ais-planetary-geospatial-foundation-models-for-global-public-health/","source":"Google Research Blog","lang":"en","kind":"news","summaryZh":"谷歌用 Earth AI 人口动态基础模型（PDFM）弥补流行病学数据的缺口和滞后，发布五个合作案例。","via":"research-samples-20261010","linkCheck":"from-feed"},{"id":"875cad4e1ea4","date":"2026-09-28","title":"New formulation helps RNA vaccines withstand high temperatures","url":"https://news.mit.edu/2026/new-formulation-helps-rna-vaccines-withstand-high-temperatures-0928","source":"MIT News：Artificial intelligence","lang":"en","kind":"news","summaryZh":"MIT News：Artificial intelligence于2026-09-28发布的文章，涉及AI 科研与调研、AI 生物/医药，原题《New formulation helps RNA vaccines withstand high temperatures》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"cf5400fd5a52","date":"2026-03-19","title":"WHO Hub in Berlin highlights 2025 achievements","url":"https://www.who.int/news/item/19-03-2026-german-health-minister-nina-warken-and-who-hub-in-berlin-highlight-2025-achievements","source":"WHO","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"ok"}],"review":{"lastReviewed":"2026-10-22","cycleDays":90,"nextReview":"2027-01-19"},"urls":{"zh":"/zh-hans/disciplines/public-health/","en":"/en/disciplines/public-health/"},"dimensions":{"method":{"score":2,"label":{"zh":"研究方法","en":"Research method"},"evidence":"AI 用于疫情预测、监测和舆情扫描，但仍是辅助","source":"https://doi.org/10.1038/s41586-024-08564-w","weight":0.3},"discovery":{"score":2,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"EVEscape 用疫情前数据预测免疫逃逸突变（Nature 2023）","source":"https://www.nature.com/articles/s41586-023-06617-0","weight":0.35},"adoption":{"score":2,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"WHO EIOS 用 AI 实时扫描公开信息识别异常卫生事件","source":"https://www.who.int/news/item/19-03-2026-german-health-minister-nina-warken-and-who-hub-in-berlin-highlight-2025-achievements","weight":0.25},"education":{"score":1,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到课程证据","source":null,"weight":0.1}},"compositeScore":1.9,"levelCapped":false,"levelChange":{"from":2,"to":2,"direction":"same","reason":null},"biggestBreakthrough":{"date":"2023-10-11","title":{"zh":"EVEscape 预测病毒免疫逃逸突变（Nature）","en":"EVEscape predicts viral immune-escape mutations (Nature)"},"url":"https://www.nature.com/articles/s41586-023-06617-0","aiImpact":"major","fieldImpact":"major","why":{"zh":"仅用疫情前数据就预测出 SARS-CoV-2 关键逃逸突变","en":"Predicted SARS-CoV-2 escape mutations from pre-pandemic data"}},"keyPapers":[{"title":"Detecting influenza epidemics using search engine query data","titleZh":"谷歌流感趋势：用搜索查询数据检测流感流行（Nature）","id":"10.1038/nature07634","url":"https://doi.org/10.1038/nature07634","venue":"Nature","date":"2009-02-19","why":{"zh":"开创用大数据和机器学习做实时疫情监测","en":"Pioneered big-data real-time epidemic surveillance"},"fieldImpact":"major","aiImpact":"major"},{"title":"The Parable of Google Flu: Traps in Big Data Analysis","titleZh":"《谷歌流感的寓言》：大数据预测高估流感峰值的教训（Science）","id":"10.1126/science.1248506","url":"https://doi.org/10.1126/science.1248506","venue":"Science","date":"2014-03-14","why":{"zh":"重要反面案例，确立公共卫生 AI 需持续验证的原则","en":"Key cautionary case for validating public-health AI"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Learning from prepandemic data to forecast viral escape","titleZh":"EVEscape 预测病毒免疫逃逸突变（Nature）","id":"10.1038/s41586-023-06617-0","url":"https://doi.org/10.1038/s41586-023-06617-0","venue":"Nature","date":"2023-10-11","why":{"zh":"仅用疫情前数据就预测出 SARS-CoV-2 关键逃逸突变","en":"Predicted SARS-CoV-2 escape mutations from pre-pandemic data"},"fieldImpact":"major","aiImpact":"major"},{"title":"Artificial intelligence for modelling infectious disease epidemics","titleZh":"综述：人工智能用于传染病流行建模（Nature）","id":"10.1038/s41586-024-08564-w","url":"https://doi.org/10.1038/s41586-024-08564-w","venue":"Nature","date":"2025-02-19","why":{"zh":"系统评估 AI 在流行病学建模中的能力与局限","en":"Systematic assessment of AI for epidemic modelling"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Mantis: A foundation model for mechanistic disease forecasting","titleZh":"Mantis：在模拟疫情上训练的机理型疾病预测基础模型，可跨病种、在数据稀缺地区预测（PNAS）","id":"10.1073/pnas.2602542123","url":"https://doi.org/10.1073/pnas.2602542123","venue":"Proceedings of the National Academy of Sciences","date":"2026-09-28","why":{"zh":"AI：仅用模拟数据训练即可泛化到真实疫情；学科：为缺乏历史数据的新发疾病和地区提供开箱即用的预测工具","en":"AI: trained on simulations, generalises to real outbreaks; Field: off-the-shelf forecasts for data-poor diseases and regions"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"A disease-agnostic approach to ensemble learning for infectious disease forecasting","titleZh":"epiFFORMA：用合成疫情数据训练的病种无关集成预测方法（Nat. Commun.）","id":"10.1038/s41467-026-70937-8","url":"https://doi.org/10.1038/s41467-026-70937-8","venue":"Nature Communications","date":"2026-03-20","why":{"zh":"新发疫情缺历史数据时也能做预测","en":"Forecasts emerging outbreaks without long histories"},"fieldImpact":"incremental","aiImpact":"incremental"},{"title":"Foundation models for time series forecasting and policy evaluation in infectious disease epidemics","titleZh":"时间序列基础模型用于传染病流行预测与政策反事实评估（Epidemics）","id":"10.1016/j.epidem.2026.100916","url":"https://doi.org/10.1016/j.epidem.2026.100916","venue":"Epidemics","date":"2026-06","why":{"zh":"基础模型在数据稀缺的短期疫情预测中有增益，但应补充而非取代传统模型","en":"Helps data-limited short-term forecasts; complements established models"},"fieldImpact":"incremental","aiImpact":"incremental"}],"latest_papers":[{"id":"arXiv:2610.29399","title":"Learning Epidemic Threshold Equations from Simulation: REEBLRA, an Interpretable Bayesian Classification Framework for Reproduction Numbers","titleZh":"REEBLRA：从模拟数据学习流行病阈值方程的可解释贝叶斯分类框架","summaryZh":"不直接回归基本再生数，而是用贝叶斯逻辑回归从参数–响应数据学出显式的流行阈值替代方程。","summaryEn":"An interpretable Bayesian logistic framework that learns explicit epidemic-threshold equations (R0) from simulation data.","authors":["Jomar F. Rabajante","Norvin P. Bansilan","Maria Czarina T. Lagura-Arreza","Destiny SM. Lutero"],"date":"2026-10-21","url":"https://arxiv.org/abs/2610.29399","site":"arXiv","venue":null,"category":"q-bio.QM","fetchedAt":"2026-10-22T10:08:19"},{"id":"doi:10.64898/2026.10.08.26365081","title":"Microbial traits have minimal power for predicting death from bloodstream infection","titleZh":"微生物特征几乎无法预测血流感染死亡","summaryZh":"分析加拿大卡尔加里 35746 例血流感染：基于电子病历的机器学习模型预测 30 天死亡的 AUC 为 0.83，加入 7 万多个微生物基因/蛋白特征后并无提升。","summaryEn":"In 35,746 bloodstream infections, EHR-based ML predicts 30-day mortality (AUC 0.83); adding >70,000 microbial features does not help.","authors":["Gilliland, R.","Valdes-Tresanco, M. E.","Plakhotnyk, A.","Westlund, A."],"date":"2026-10-11","url":"https://www.medrxiv.org/content/10.64898/2026.10.08.26365081v1","site":"medRxiv","venue":null,"category":"infectious diseases","fetchedAt":"2026-10-22T10:08:19"},{"id":"doi:10.64898/2026.05.07.26352661","title":"Simpler is not always better: Phylodynamic misspecification and deep-learning corrections","titleZh":"越简单不一定越好：系统动力学模型误设与深度学习校正","summaryZh":"研究用错模型推断再生数 R 和感染期时的偏差，并用深度学习方法加以校正。","summaryEn":"Shows misspecified phylodynamic models bias R estimates and corrects them with deep learning.","authors":["XIE, R.","Gascuel, O.","ZHUKOVA, A."],"date":"2026-10-06","url":"https://www.medrxiv.org/content/10.64898/2026.05.07.26352661v3","site":"medRxiv","venue":null,"category":"epidemiology","fetchedAt":"2026-10-22T10:08:19"},{"id":"arXiv:2609.30646","title":"Amortized Bayesian Disease Mapping and Boundary Detection on Heterogeneous Spatial Graphs","titleZh":"异质空间图上的摊销贝叶斯疾病制图与边界检测","summaryZh":"提出协变量信息驱动的摊销贝叶斯方法，自动判断疾病地图中哪里不应平滑，以揭示局部健康不平等。","summaryEn":"Amortised Bayesian method decides where to break smoothing in disease maps to reveal local disparities.","authors":["Luca Aiello","Sudipto Banerjee"],"date":"2026-09-25","url":"https://arxiv.org/abs/2609.30646","site":"arXiv","venue":null,"category":"stat.ME","fetchedAt":"2026-10-22T10:08:19"},{"id":"doi:10.64898/2026.09.21.26363552","title":"Unmasking digital despair: A language model framework for predicting depression on social media","titleZh":"揭开数字绝望：预测社交媒体抑郁的语言模型框架","summaryZh":"用 2023–2024 年 Twitter/X 帖子训练深度学习模型预测抑郁，并用大模型归纳可能的致抑郁原因。","summaryEn":"Deep learning on Twitter/X posts predicts depression and an LLM characterises its possible causes.","authors":["Xie, M.","He, H.","Xie, Z."],"date":"2026-09-23","url":"https://www.medrxiv.org/content/10.64898/2026.09.21.26363552v1","site":"medRxiv","venue":null,"category":"public and global health","fetchedAt":"2026-10-22T10:08:19"}]},{"id":"ecology","slug":"ecology","category":"life-medical","categoryName":{"zh":"生命与医学","en":"Life & medical sciences"},"name":{"zh":"生态学与生物多样性","en":"Ecology & biodiversity"},"classification":{"oecdFos":"1.6","moe":"0713 生态学"},"level":2,"levelKey":"breakthrough","levelLabel":{"zh":"局部突破","en":"Local breakthroughs"},"trend":"up","reason":{"zh":"声学识别（BirdNET）和相机陷阱识别已是野外监测常规工具；欧洲 Biodiversa+ 试点在 70 个站点部署约 200 台 AI 监测设备，生态监测从试点走向运行网络。","en":"Acoustic (BirdNET) and camera-trap classifiers are routine; Europe's Biodiversa+ pilot deployed ~200 AI sensors at 70 sites, moving from pilots to networks."},"reasonSources":[{"url":"https://doi.org/10.1016/j.ecoinf.2021.101236","linkCheck":"ok"},{"url":"https://zenodo.org/records/22141164","linkCheck":"site-blocks-bots"}],"directions":{"zh":["声学物种识别","相机陷阱自动识别","栖息地制图"],"en":["Acoustic species ID","Camera-trap classification","Habitat mapping"]},"stats":{"milestones":7,"milestonesSince2025":4,"milestones2026":4,"milestonesLast6Months":2,"news":6,"latestProgress":"2026-10-20","latestMilestone":"2026-10-20","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":2,"leaps":{"ai":0,"field":0},"latestPapers":7,"keyPapers":5},"milestones":[{"date":"2018-06-05","title":{"zh":"深度学习自动识别、计数和描述相机陷阱中的野生动物（PNAS）","en":"Deep learning automatically identifies, counts and describes wild animals in camera-trap images (PNAS)"},"url":"https://doi.org/10.1073/pnas.1719367115","source":"PNAS","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"在 320 万张 Snapshot Serengeti 图像上达到人工志愿者水平，节省约 8 年人工","en":"Volunteer-level accuracy on 3.2M images, saving ~8 years of labour"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"在 320 万张 Snapshot Serengeti 图像上达到人工志愿者水平，节省约 8 年人工","en":"Volunteer-level accuracy on 3.2M images, saving ~8 years of labour"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"在 320 万张 Snapshot Serengeti 图像上达到人工志愿者水平，节省约 8 年人工","en":"Volunteer-level accuracy on 3.2M images, saving ~8 years of labour"}},"impactNotes":[],"verified":{"id":"10.1073/pnas.1719367115","title":"Automatically identifying, counting, and describing wild animals in camera-trap images with deep learning","venue":"Proceedings of the National Academy of Sciences","date":"2018-06-05"},"addedIn":"v2"},{"date":"2021-03","title":{"zh":"BirdNET：鸟类声学深度学习识别（Ecological Informatics）","en":"BirdNET deep-learning bird sound ID (Ecological Informatics)"},"url":"https://doi.org/10.1016/j.ecoinf.2021.101236","source":"Ecological Informatics","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"month","type":"system","typeLabel":{"zh":"系统发布","en":"System"},"why":{"zh":"覆盖近千种鸟类的声学识别工具，成为野外监测标准工具","en":"Standard acoustic ID tool for ~1,000 species"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"覆盖近千种鸟类的声学识别工具，成为野外监测标准工具","en":"Standard acoustic ID tool for ~1,000 species"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"覆盖近千种鸟类的声学识别工具，成为野外监测标准工具","en":"Standard acoustic ID tool for ~1,000 species"}},"impactNotes":[],"verified":{"id":"10.1016/j.ecoinf.2021.101236","title":"BirdNET: A deep learning solution for avian diversity monitoring","venue":"Ecological Informatics","date":"2021-03"},"addedIn":"v1"},{"date":"2022-02-09","title":{"zh":"综述：机器学习用于野生动物保护的前景（Nat. Commun.）","en":"Perspectives in machine learning for wildlife conservation (Nat. Commun.)"},"url":"https://doi.org/10.1038/s41467-022-27980-y","source":"Nature Communications","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"领域综述","en":"Field perspective"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"领域综述","en":"Field perspective"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"领域综述","en":"Field perspective"}},"impactNotes":[],"verified":{"id":"10.1038/s41467-022-27980-y","title":"Perspectives in machine learning for wildlife conservation","venue":"Nature Communications","date":"2022-02-09"},"addedIn":"v2"},{"date":"2026","title":{"zh":"Biodiversa+ 自动生物多样性监测站试点报告（约 200 台传感器）","en":"Biodiversa+ automated monitoring stations pilot (~200 sensors)"},"url":"https://zenodo.org/records/22141164","source":"Zenodo","flags":["tbv"],"flagLabels":[{"zh":"日期或细节待核实","en":"Date/details to be verified"}],"linkCheck":"site-blocks-bots","datePrecision":"year","type":"deployment","typeLabel":{"zh":"落地部署","en":"Deployment"},"why":{"zh":"欧洲约 200 台 AI 监测设备试点","en":"~200 AI sensors piloted in Europe"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"欧洲约 200 台 AI 监测设备试点","en":"~200 AI sensors piloted in Europe"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"欧洲约 200 台 AI 监测设备试点","en":"~200 AI sensors piloted in Europe"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-01","title":{"zh":"BirdNET 全球性能评估：67 个站点、1020 个物种，召回率因地区差异大（Ecological Indicators）","en":"Global BirdNET evaluation across 67 sites and 1,020 species (Ecological Indicators)"},"url":"https://doi.org/10.1016/j.ecolind.2025.114550","source":"Ecological Indicators","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"month","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"首次全球尺度独立评估最常用的鸟类声学 AI，指出地区偏差","en":"First global independent evaluation; shows regional bias"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"首次全球尺度独立评估最常用的鸟类声学 AI，指出地区偏差","en":"First global independent evaluation; shows regional bias"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首次全球尺度独立评估最常用的鸟类声学 AI，指出地区偏差","en":"First global independent evaluation; shows regional bias"}},"impactNotes":[],"verified":{"id":"10.1016/j.ecolind.2025.114550","title":"A global assessment of BirdNET performance: Differences among continents, biomes, and species","venue":"Ecological Indicators","date":"2026-01"},"addedIn":"v2"},{"date":"2026-05","title":{"zh":"AI 把野生动物追踪分析从数月缩短到数天","en":"AI cuts wildlife tracking time from months to days"},"url":"https://phys.org/news/2026-05-ai-wildlife-tracking-months-days.html","source":"Phys.org","flags":["tbv"],"flagLabels":[{"zh":"日期或细节待核实","en":"Date/details to be verified"}],"linkCheck":"site-blocks-bots","datePrecision":"month","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"追踪数据分析提速","en":"Faster tracking analysis"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"追踪数据分析提速","en":"Faster tracking analysis"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"追踪数据分析提速","en":"Faster tracking analysis"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-10-20","title":{"zh":"生物多样性基础模型 BioCLIP 2.5 迁移到 2727 种蛾类野外监测（Scientific Reports）","en":"Biodiversity foundation model transfers to field moth monitoring of 2,727 species (Sci. Rep.)"},"url":"https://doi.org/10.1038/s41598-026-75760-1","source":"Scientific Reports","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"基础模型缩小实验室到野外的部署差距","en":"Foundation models close the lab-to-field gap"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"基础模型缩小实验室到野外的部署差距","en":"Foundation models close the lab-to-field gap"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"基础模型缩小实验室到野外的部署差距","en":"Foundation models close the lab-to-field gap"}},"impactNotes":[],"verified":{"id":"10.1038/s41598-026-75760-1","title":"Bridging the deployment gap in automated moth monitoring through biodiversity foundation models","venue":"Scientific Reports","date":"2026-10-20"},"addedIn":"v2"}],"papersRepos":[{"title":"kahst/BirdNET-Analyzer","url":"https://github.com/kahst/BirdNET-Analyzer","date":null,"kind":"repo","linkCheck":"ok"},{"title":"google/cameratrapai (SpeciesNet)","url":"https://github.com/google/cameratrapai","date":null,"kind":"repo","linkCheck":"ok"},{"title":"Automatically identifying, counting, and describing wild animals in camera-trap images with deep learning","url":"https://doi.org/10.1073/pnas.1719367115","date":"2018-06-05","kind":"paper","linkCheck":"ok-crossref"},{"title":"BirdNET: A deep learning solution for avian diversity monitoring","url":"https://doi.org/10.1016/j.ecoinf.2021.101236","date":"2021-03","kind":"paper","linkCheck":"ok-crossref"},{"title":"A global assessment of BirdNET performance: Differences among continents, biomes, and species","url":"https://doi.org/10.1016/j.ecolind.2025.114550","date":"2026-01","kind":"paper","linkCheck":"ok-crossref"},{"title":"Perspectives in machine learning for wildlife conservation","url":"https://doi.org/10.1038/s41467-022-27980-y","date":"2022-02-09","kind":"paper","linkCheck":"ok-crossref"},{"title":"Bridging the deployment gap in automated moth monitoring through biodiversity foundation models","url":"https://doi.org/10.1038/s41598-026-75760-1","date":"2026-10-20","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"Cornell Lab of Ornithology","url":"https://www.birds.cornell.edu"},{"name":"Microsoft AI for Good Lab","url":"https://www.microsoft.com/en-us/research/group/ai-for-good-research-lab/"}],"news":[{"id":"67f05660d089","date":"2026-10-20","title":"Bridging the deployment gap in automated moth monitoring through biodiversity foundation models","url":"https://www.nature.com/articles/s41598-026-75760-1","source":"Nature 学科·计算生物","lang":"en","kind":"news","summaryZh":"Nature 学科·计算生物于2026-10-20发布的文章，涉及AI 生物/医药，原题《Bridging the deployment gap in automated moth monitoring through biodiversity foundation m》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"73c1666dba6b","date":"2026-10-19","title":"Kodama nets $7m to scale physical AI for “intelligent forestry”","url":"https://agfundernews.com/kodama-nets-7m-to-scale-physical-ai-for-intelligent-forestry","source":"AgFunderNews","lang":"en","kind":"news","summaryZh":"林业“物理 AI”初创公司 Kodama 获 700 万美元种子轮融资（累计 1400 万美元），扩大在美国商业伐木作业中的应用。","via":"pool-v2-20261022","linkCheck":"ok"},{"id":"5b95d2b8a348","date":"2026-10-16","title":"Explainable artificial intelligence reveals spatially divergent effects of global change on mammals","url":"https://www.nature.com/articles/s41467-026-78677-5","source":"Nature Communications","lang":"en","kind":"paper","summaryZh":"Nature Communications：用可解释机器学习分析近 2000 种陆地哺乳动物，揭示气候与土地覆盖变化在不同地区的影响方向不同。","via":"manual-websearch-20261022","linkCheck":"ok"},{"id":"75dde2a6193f","date":"2026-10-15","title":"Are you a creature or a machine? How to live in ecological community in a time of AI","url":"https://theconversation.com/are-you-a-creature-or-a-machine-how-to-live-in-ecological-community-in-a-time-of-ai-292765","source":"The Conversation · AI","lang":"en","kind":"news","summaryZh":"The Conversation · AI于2026-10-16发布的文章，涉及AI 科研与调研，原题《Are you a creature or a machine? How to live in ecological community in a time of AI》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"52395d7243db","date":"2026-09-30","title":"Drones, AI and the future of wildlife conservation","url":"https://physicsworld.com/a/drones-ai-and-the-future-of-wildlife-conservation/","source":"Physics World（IOP）","lang":"en","kind":"news","summaryZh":"Physics World（IOP）于2026-09-30发布的文章，涉及AI 科研与调研、AI 机器人，原题《Drones, AI and the future of wildlife conservation》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"2713db95b4a9","date":"2026-05","title":"AI cuts wildlife tracking time from months to days","url":"https://phys.org/news/2026-05-ai-wildlife-tracking-months-days.html","source":"Phys.org","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"site-blocks-bots"}],"review":{"lastReviewed":"2026-10-21","cycleDays":90,"nextReview":"2027-01-19"},"urls":{"zh":"/zh-hans/disciplines/ecology/","en":"/en/disciplines/ecology/"},"dimensions":{"method":{"score":3,"label":{"zh":"研究方法","en":"Research method"},"evidence":"相机陷阱和声学自动识别是野外监测常规方法","source":"https://doi.org/10.1073/pnas.1719367115","weight":0.3},"discovery":{"score":1,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"AI 主要提升监测效率，以 AI 为关键的生态学新发现很少","source":null,"weight":0.35},"adoption":{"score":3,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"BirdNET、SpeciesNet 等工具被保护机构广泛使用；Biodiversa+ 试点 70 个站点","source":"https://doi.org/10.1016/j.ecolind.2025.114550","weight":0.25},"education":{"score":1,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到课程证据","source":null,"weight":0.1}},"compositeScore":2.1,"levelCapped":false,"levelChange":{"from":2,"to":2,"direction":"same","reason":null},"biggestBreakthrough":{"date":"2021-03","title":{"zh":"BirdNET：鸟类声学深度学习识别（Ecological Informatics）","en":"BirdNET deep-learning bird sound ID (Ecological Informatics)"},"url":"https://doi.org/10.1016/j.ecoinf.2021.101236","aiImpact":"major","fieldImpact":"major","why":{"zh":"覆盖近千种鸟类的声学识别工具，成为野外监测标准工具","en":"Standard acoustic ID tool for ~1,000 species"}},"keyPapers":[{"title":"Automatically identifying, counting, and describing wild animals in camera-trap images with deep learning","titleZh":"深度学习自动识别、计数和描述相机陷阱中的野生动物（PNAS）","id":"10.1073/pnas.1719367115","url":"https://doi.org/10.1073/pnas.1719367115","venue":"Proceedings of the National Academy of Sciences","date":"2018-06-05","why":{"zh":"在 320 万张 Snapshot Serengeti 图像上达到人工志愿者水平，节省约 8 年人工","en":"Volunteer-level accuracy on 3.2M images, saving ~8 years of labour"},"fieldImpact":"major","aiImpact":"major"},{"title":"BirdNET: A deep learning solution for avian diversity monitoring","titleZh":"BirdNET：鸟类声学深度学习识别（Ecological Informatics）","id":"10.1016/j.ecoinf.2021.101236","url":"https://doi.org/10.1016/j.ecoinf.2021.101236","venue":"Ecological Informatics","date":"2021-03","why":{"zh":"覆盖近千种鸟类的声学识别工具，成为野外监测标准工具","en":"Standard acoustic ID tool for ~1,000 species"},"fieldImpact":"major","aiImpact":"major"},{"title":"A global assessment of BirdNET performance: Differences among continents, biomes, and species","titleZh":"BirdNET 全球性能评估：67 个站点、1020 个物种，召回率因地区差异大（Ecological Indicators）","id":"10.1016/j.ecolind.2025.114550","url":"https://doi.org/10.1016/j.ecolind.2025.114550","venue":"Ecological Indicators","date":"2026-01","why":{"zh":"首次全球尺度独立评估最常用的鸟类声学 AI，指出地区偏差","en":"First global independent evaluation; shows regional bias"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Perspectives in machine learning for wildlife conservation","titleZh":"综述：机器学习用于野生动物保护的前景（Nat. Commun.）","id":"10.1038/s41467-022-27980-y","url":"https://doi.org/10.1038/s41467-022-27980-y","venue":"Nature Communications","date":"2022-02-09","why":{"zh":"领域综述","en":"Field perspective"},"fieldImpact":"incremental","aiImpact":"incremental"},{"title":"Bridging the deployment gap in automated moth monitoring through biodiversity foundation models","titleZh":"生物多样性基础模型 BioCLIP 2.5 迁移到 2727 种蛾类野外监测（Scientific Reports）","id":"10.1038/s41598-026-75760-1","url":"https://doi.org/10.1038/s41598-026-75760-1","venue":"Scientific Reports","date":"2026-10-20","why":{"zh":"基础模型缩小实验室到野外的部署差距","en":"Foundation models close the lab-to-field gap"},"fieldImpact":"incremental","aiImpact":"incremental"}],"latest_papers":[{"id":"doi:10.64898/2026.02.17.706216","title":"Phylogenetic estimation of diversity-dependent biogeographic rates using deep learning","titleZh":"用深度学习估计依赖多样性的生物地理速率","summaryZh":"提出事件驱动的系统发育多样化模型 DDGeoSSE，并用深度学习从系统发育树估计随物种丰富度变化的成种、灭绝与扩散速率。","summaryEn":"Introduces DDGeoSSE and uses deep learning to estimate diversity-dependent speciation, extinction and dispersal rates from phylogenies.","authors":["Soewongsono, A. C.","Landis, M. J."],"date":"2026-10-15","url":"https://www.biorxiv.org/content/10.64898/2026.02.17.706216v2","site":"bioRxiv","venue":null,"category":"evolutionary biology","fetchedAt":"2026-10-22T10:08:28"},{"id":"doi:10.64898/2026.10.08.757537","title":"Recording context and data splitting can inflate acoustic re-identification accuracy across species and studies","titleZh":"录音情境与数据划分会虚高跨物种声学个体再识别准确率","summaryZh":"指出按领地或录音站归属个体、以及不当的数据划分，会让机器学习声学个体识别的准确率被高估（已发表研究常报告 90% 以上）。","summaryEn":"Recording context and data splitting can inflate ML acoustic re-identification accuracy, often reported above 90%.","authors":["Wallington, G.","Lighten, J."],"date":"2026-10-09","url":"https://www.biorxiv.org/content/10.64898/2026.10.08.757537v1","site":"bioRxiv","venue":null,"category":"ecology","fetchedAt":"2026-10-22T10:08:28"},{"id":"doi:10.64898/2026.10.04.756574","title":"Fully integrative species delimitation with machine learning and diverse data types in delimSOM 2.0","titleZh":"delimSOM 2.0：用机器学习和多种数据类型完全整合地划分物种","summaryZh":"发布 R 包 delimSOM 2.0，以多层自组织映射同时分析基因组、表型、生态和空间数据来划分物种。","summaryEn":"delimSOM 2.0 uses multilayer self-organising maps to integrate genomic, phenotypic, ecological and spatial data for species delimitation.","authors":["Schoenberger, D.","Pyron, R. A.","Dupuis, J. R."],"date":"2026-10-07","url":"https://www.biorxiv.org/content/10.64898/2026.10.04.756574v1","site":"bioRxiv","venue":null,"category":"evolutionary biology","fetchedAt":"2026-10-22T10:08:28"},{"id":"doi:10.64898/2026.09.28.754864","title":"DAPHNE: A global database of pest herbivores and their natural enemies","titleZh":"DAPHNE：全球害虫植食者及其天敌数据库","summaryZh":"用开放权重的预训练大模型分析海量论文摘要，提取害虫与天敌的相互作用，建成全球数据库。","summaryEn":"Uses an open-weight LLM to mine abstracts and build a global database of pest–natural-enemy interactions.","authors":["Scheepens, D. R.","Freeman, R.","Newbold, T."],"date":"2026-10-01","url":"https://www.biorxiv.org/content/10.64898/2026.09.28.754864v3","site":"bioRxiv","venue":null,"category":"ecology","fetchedAt":"2026-10-22T10:08:28"},{"id":"doi:10.64898/2026.09.28.755061","title":"Acoustic Monitoring of Tropical Bats: Saccopteryx bilineata's response to habitat and time across a restoration gradient using a 2D-CNN","titleZh":"热带蝙蝠声学监测：用二维 CNN 研究大银线蝠在恢复梯度上对生境与时间的响应","summaryZh":"开发轻量级二维卷积神经网络，从被动声学监测数据中自动识别大银线蝠回声定位叫声，分析其在不同恢复程度生境中的活动。","summaryEn":"A lightweight 2D CNN detects Saccopteryx bilineata echolocation calls to study activity across a restoration gradient.","authors":["Secord, I. L.","Zhang, X.","Preston-Allen, R.","Banks-Leite, C."],"date":"2026-09-29","url":"https://www.biorxiv.org/content/10.64898/2026.09.28.755061v1","site":"bioRxiv","venue":null,"category":"ecology","fetchedAt":"2026-10-22T10:08:28"},{"id":"doi:10.64898/2026.09.24.753096","title":"PlanktonLake-CEREEP- A Freshwater Plankton Image Dataset with Semi-Automated Label Cleaning","titleZh":"PlanktonLake-CEREEP：带半自动标签清洗的淡水浮游生物图像数据集","summaryZh":"发布新的带标注淡水浮游生物显微图像数据集，用于训练机器学习分类模型，减少人工标注负担。","summaryEn":"A labelled freshwater plankton image dataset with semi-automated label cleaning for ML classifiers.","authors":["Dechaumet, L.","Puppo, C.","Carmignac, D.","Blusseau, S."],"date":"2026-09-29","url":"https://www.biorxiv.org/content/10.64898/2026.09.24.753096v2","site":"bioRxiv","venue":null,"category":"ecology","fetchedAt":"2026-10-22T10:08:28"},{"id":"arXiv:2610.28590","title":"WildMAE: Structure-Aware Masked Autoencoding for Wildlife Image Reconstruction","titleZh":"WildMAE：结构感知的掩码自编码器用于野生动物图像重建","summaryZh":"在 MAE 解码端加入局部结构细化与掩码到可见的交叉注意力，在 Snapshot Serengeti 相机陷阱图像上重建质量明显优于原版 MAE。","summaryEn":"A structure-aware MAE improves camera-trap image reconstruction on Snapshot Serengeti.","authors":["Yicheng Zhang","Yujie Jin","Ke Zhang","Jiaxin Wu"],"date":"2026-10-20","url":"https://arxiv.org/abs/2610.28590","site":"arXiv","venue":null,"category":"cs.CV","fetchedAt":"2026-10-22T10:08:28"}]},{"id":"agriculture","slug":"agriculture","category":"life-medical","categoryName":{"zh":"生命与医学","en":"Life & medical sciences"},"name":{"zh":"农学","en":"Agricultural sciences"},"classification":{"oecdFos":"4.1","moe":"0901 作物学"},"level":1,"levelKey":"assistive","levelLabel":{"zh":"辅助起步","en":"Assistive"},"trend":"flat","reason":{"zh":"AI 主要用于基因组选择、表型预测和病虫害识别；2026 年 Nature Communications 的作物基因组预测模型仍属基准精度提升，尚无经田间验证的标志性 AI 育种突破。","en":"AI is used for genomic selection, phenotype prediction and pest ID; 2026 Nature Communications crop models are benchmark gains, with no field-validated landmark AI breeding result yet."},"reasonSources":[{"url":"https://www.nature.com/articles/s41467-026-71035-5","linkCheck":"ok"},{"url":"https://www.nature.com/articles/s41467-026-75788-x","linkCheck":"ok"}],"directions":{"zh":["基因组选择","表型组学","病虫害识别","精准农业"],"en":["Genomic selection","Phenomics","Pest & disease ID","Precision farming"]},"stats":{"milestones":7,"milestonesSince2025":5,"milestones2026":3,"milestonesLast6Months":2,"news":5,"latestProgress":"2026-10-20","latestMilestone":"2026-10-15","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":1,"leaps":{"ai":0,"field":0},"latestPapers":7,"keyPapers":4},"milestones":[{"date":"2016-09-22","title":{"zh":"深度学习识别 14 种作物 26 种病害（Frontiers in Plant Science，PlantVillage）","en":"Deep learning for image-based plant disease detection (Frontiers in Plant Science)"},"url":"https://doi.org/10.3389/fpls.2016.01419","source":"Frontiers in Plant Science","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"开启手机拍照识别作物病害的研究方向","en":"Launched smartphone-based crop disease diagnosis"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"开启手机拍照识别作物病害的研究方向","en":"Launched smartphone-based crop disease diagnosis"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"开启手机拍照识别作物病害的研究方向","en":"Launched smartphone-based crop disease diagnosis"}},"impactNotes":[],"verified":{"id":"10.3389/fpls.2016.01419","title":"Using Deep Learning for Image-Based Plant Disease Detection","venue":"Frontiers in Plant Science","date":"2016-09-22"},"addedIn":"v2"},{"date":"2021-01-06","title":{"zh":"综述：深度学习用于作物基因组预测（BMC Genomics）","en":"Review of deep learning for genomic prediction (BMC Genomics)"},"url":"https://doi.org/10.1186/s12864-020-07319-x","source":"BMC Genomics","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"系统比较深度学习与传统 GBLUP 模型，结论是优势有限","en":"Finds limited advantage over GBLUP"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"系统比较深度学习与传统 GBLUP 模型，结论是优势有限","en":"Finds limited advantage over GBLUP"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"系统比较深度学习与传统 GBLUP 模型，结论是优势有限","en":"Finds limited advantage over GBLUP"}},"impactNotes":[],"verified":{"id":"10.1186/s12864-020-07319-x","title":"A review of deep learning applications for genomic selection","venue":"BMC Genomics","date":"2021-01-06"},"addedIn":"v2"},{"date":"2025","title":{"zh":"John Deere See & Spray：2025 年客户在 500 多万英亩上使用 AI 识别杂草精准喷药，非残留除草剂用量减少近一半","en":"John Deere See & Spray used on 5M+ acres in 2025, cutting non-residual herbicide use ~50%"},"url":"https://www.deere.ca/en/news/all-news/see-spray-technology-across-5-million-acres/","source":"John Deere","flags":["vendor","tbv"],"flagLabels":[{"zh":"厂商自述","en":"Vendor claim"},{"zh":"日期或细节待核实","en":"Date/details to be verified"}],"linkCheck":"ok","datePrecision":"year","type":"deployment","typeLabel":{"zh":"落地部署","en":"Deployment"},"why":{"zh":"AI 视觉进入大田规模化作业；阿肯色大学独立试验显示减少 43%–59% 苗后除草剂，但漏喷导致杂草种子增加","en":"At-scale field deployment; independent trials show 43–59% savings but missed weeds"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI 视觉进入大田规模化作业；阿肯色大学独立试验显示减少 43%–59% 苗后除草剂，但漏喷导致杂草种子增加","en":"At-scale field deployment; independent trials show 43–59% savings but missed weeds"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI 视觉进入大田规模化作业；阿肯色大学独立试验显示减少 43%–59% 苗后除草剂，但漏喷导致杂草种子增加","en":"At-scale field deployment; independent trials show 43–59% savings but missed weeds"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2025-03-10","title":{"zh":"阿肯色大学三年试验：See & Spray 减少 43%–59% 苗后除草剂，但敏感度低使长芒苋种子产量每年增加 280%","en":"University of Arkansas 3-year trials: 43–59% less herbicide, but low sensitivity raised Palmer amaranth seed output 280%/yr"},"url":"https://www.uaex.uada.edu/media-resources/news/2025/march/03-10-2025-see-and-spray-research.aspx","source":"University of Arkansas","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"独立田间证据，兼有收益和风险","en":"Independent field evidence of benefits and risks"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"独立田间证据，兼有收益和风险","en":"Independent field evidence of benefits and risks"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"独立田间证据，兼有收益和风险","en":"Independent field evidence of benefits and risks"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-03-26","title":{"zh":"GP-WAITER：Transformer 改进作物复杂性状预测（Nat. Commun.）","en":"GP-WAITER Transformer improves crop trait prediction (Nat. Commun.)"},"url":"https://www.nature.com/articles/s41467-026-71035-5","source":"Nature Communications","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"基准精度提升，未经田间育种验证","en":"Benchmark gain, not field-validated"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"基准精度提升，未经田间育种验证","en":"Benchmark gain, not field-validated"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"基准精度提升，未经田间育种验证","en":"Benchmark gain, not field-validated"}},"impactNotes":[],"verified":{"id":"10.1038/s41467-026-71035-5","title":"Leveraging weighted embedding and Transformer architecture to improve phenotype prediction of complex traits for crops","venue":"Nature Communications","date":"2026-03-26"},"addedIn":"v1"},{"date":"2026-07-20","title":{"zh":"GEG2P：集成学习提升五种作物基因组预测精度（Nat. Commun.）","en":"GEG2P ensemble improves genomic prediction across five crops (Nat. Commun.)"},"url":"https://www.nature.com/articles/s41467-026-75788-x","source":"Nature Communications","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"集成学习提升五种作物预测精度","en":"Ensemble gains on five crops"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"集成学习提升五种作物预测精度","en":"Ensemble gains on five crops"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"集成学习提升五种作物预测精度","en":"Ensemble gains on five crops"}},"impactNotes":[],"verified":{"id":"10.1038/s41467-026-75788-x","title":"Robustly enhancing crop genomic prediction accuracy through ensemble learning and iterative optimization","venue":"Nature Communications","date":"2026-07-20"},"addedIn":"v1"},{"date":"2026-10-15","title":{"zh":"谷歌 AnthroKrishi“农业理解”模型 API 公开预览、覆盖 11 国；古吉拉特邦 Krushi Pragati 平台一年内据此向 60 多万农户发送 9 亿多条短信农情建议","en":"Google opens its Agricultural Understanding models in public preview (11 countries); Gujarat’s Krushi Pragati platform sent 900M+ SMS advisories to 600k+ farmers using them"},"url":"https://blog.google/intl/en-in/company-news/from-lab-to-impact-advancing-ai-for-real-world-progress-in-india/","source":"Google India Blog","flags":["vendor"],"flagLabels":[{"zh":"厂商自述","en":"Vendor claim"}],"linkCheck":"ok","datePrecision":"day","type":"deployment","typeLabel":{"zh":"落地部署","en":"Deployment"},"why":{"zh":"AI：遥感与地块级作物识别模型以 API 开放给开发者和研究者；学科：在邦级政府平台覆盖 1400 多万公顷的规模化农情服务，但成效数字为谷歌自述","en":"AI: field-level landscape and crop models opened via API. Field: state-government advisory service covering 14M+ hectares (figures are Google’s own)"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：遥感与地块级作物识别模型以 API 开放给开发者和研究者；学科：在邦级政府平台覆盖 1400 多万公顷的规模化农情服务，但成效数字为谷歌自述","en":"AI: field-level landscape and crop models opened via API. Field: state-government advisory service covering 14M+ hectares (figures are Google’s own)"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：遥感与地块级作物识别模型以 API 开放给开发者和研究者；学科：在邦级政府平台覆盖 1400 多万公顷的规模化农情服务，但成效数字为谷歌自述","en":"AI: field-level landscape and crop models opened via API. Field: state-government advisory service covering 14M+ hectares (figures are Google’s own)"}},"impactNotes":[],"verified":null,"addedIn":"v2-20261022pm"}],"papersRepos":[{"title":"Using Deep Learning for Image-Based Plant Disease Detection","url":"https://doi.org/10.3389/fpls.2016.01419","date":"2016-09-22","kind":"paper","linkCheck":"ok-crossref"},{"title":"A review of deep learning applications for genomic selection","url":"https://doi.org/10.1186/s12864-020-07319-x","date":"2021-01-06","kind":"paper","linkCheck":"ok-crossref"},{"title":"Leveraging weighted embedding and Transformer architecture to improve phenotype prediction of complex traits for crops","url":"https://doi.org/10.1038/s41467-026-71035-5","date":"2026-03-26","kind":"paper","linkCheck":"ok-crossref"},{"title":"Robustly enhancing crop genomic prediction accuracy through ensemble learning and iterative optimization","url":"https://doi.org/10.1038/s41467-026-75788-x","date":"2026-07-20","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"CIMMYT","url":"https://www.cimmyt.org"},{"name":"FAO","url":"https://www.fao.org"}],"news":[{"id":"929398b5f91e","date":"2026-10-20","title":"Reservoir launches new agricultural robotics incubator for Washington growers","url":"https://www.therobotreport.com/reservoir-launches-new-agricultural-robotics-incubator-for-washington-growers/","source":"The Robot Report","lang":"en","kind":"news","summaryZh":"农业机器人孵化器 Reservoir 在华盛顿州为种植者新设孵化点。","via":"pool-v2-20261021","linkCheck":"from-feed"},{"id":"c077b962ec0c","date":"2026-10-20","title":"Reservoir Farms expands to Pacific Northwest to accelerate physical AI as membership tops 30","url":"https://agfundernews.com/reservoir-farms-expands-to-pacific-northwest-to-accelerate-physical-ai-as-membership-tops-30","source":"AgFunderNews","lang":"en","kind":"news","summaryZh":"Reservoir Farms 扩展到美国西北太平洋地区，推进农业“物理 AI”，成员企业超过 30 家。","via":"pool-v2-20261021","linkCheck":"from-feed"},{"id":"f8294ee56bc6","date":"2026-10-19","title":"Is smart farming really smart business?","url":"https://www.ft.com/content/aeb432cd-5d95-4062-b7a8-f494592adb34","source":"Financial Times AI 栏目","lang":"en","kind":"news","summaryZh":"Financial Times AI 栏目于2026-10-19发布的文章，涉及AI 金融，原题《Is smart farming really smart business?》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"905a0dff5b38","date":"2026-10-15","title":"Google Brings AI-Powered Crop Research & Business Tools To India","url":"https://analyticsindiamag.com/ai-news/google-brings-ai-powered-crop-research-business-tools-to-india","source":"Analytics India Magazine","lang":"en","kind":"news","summaryZh":"谷歌在印度扩大 AI 项目，覆盖农作物研究、教育和小微企业。","via":"pool-v2-20261021","linkCheck":"from-feed"},{"id":"9b74b75b98c4","date":"2026-10-09","title":"具身机器人成功变身小麦育种专家","url":"https://news.sciencenet.cn/htmlnews/2026/10/572633.shtm","source":"科学网（中国科学报）","lang":"zh","kind":"news","summaryZh":"山东农业大学团队发布小麦育种具身机器人原型机，能在田间自动成像、测量穗部性状，并融入育种家经验做判断。","via":"research-samples-20261010","linkCheck":"from-feed"}],"review":{"lastReviewed":"2026-10-22","cycleDays":90,"nextReview":"2027-01-19"},"urls":{"zh":"/zh-hans/disciplines/agriculture/","en":"/en/disciplines/agriculture/"},"dimensions":{"method":{"score":2,"label":{"zh":"研究方法","en":"Research method"},"evidence":"深度学习基因组预测、病害识别论文大量出现","source":"https://doi.org/10.1186/s12864-020-07319-x","weight":0.3},"discovery":{"score":1,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"多为精度提升，缺少经田间验证的标志性 AI 育种突破","source":null,"weight":0.35},"adoption":{"score":1,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"缺少农户或育种单位采用率数据","source":null,"weight":0.25},"education":{"score":1,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到课程证据","source":null,"weight":0.1}},"compositeScore":1.3,"levelCapped":false,"levelChange":{"from":1,"to":1,"direction":"same","reason":null},"biggestBreakthrough":{"date":"2016-09-22","title":{"zh":"深度学习识别 14 种作物 26 种病害（Frontiers in Plant Science，PlantVillage）","en":"Deep learning for image-based plant disease detection (Frontiers in Plant Science)"},"url":"https://doi.org/10.3389/fpls.2016.01419","aiImpact":"major","fieldImpact":"major","why":{"zh":"开启手机拍照识别作物病害的研究方向","en":"Launched smartphone-based crop disease diagnosis"}},"keyPapers":[{"title":"Using Deep Learning for Image-Based Plant Disease Detection","titleZh":"深度学习识别 14 种作物 26 种病害（Frontiers in Plant Science，PlantVillage）","id":"10.3389/fpls.2016.01419","url":"https://doi.org/10.3389/fpls.2016.01419","venue":"Frontiers in Plant Science","date":"2016-09-22","why":{"zh":"开启手机拍照识别作物病害的研究方向","en":"Launched smartphone-based crop disease diagnosis"},"fieldImpact":"major","aiImpact":"major"},{"title":"A review of deep learning applications for genomic selection","titleZh":"综述：深度学习用于作物基因组预测（BMC Genomics）","id":"10.1186/s12864-020-07319-x","url":"https://doi.org/10.1186/s12864-020-07319-x","venue":"BMC Genomics","date":"2021-01-06","why":{"zh":"系统比较深度学习与传统 GBLUP 模型，结论是优势有限","en":"Finds limited advantage over GBLUP"},"fieldImpact":"incremental","aiImpact":"incremental"},{"title":"Leveraging weighted embedding and Transformer architecture to improve phenotype prediction of complex traits for crops","titleZh":"GP-WAITER：Transformer 改进作物复杂性状预测（Nat. Commun.）","id":"10.1038/s41467-026-71035-5","url":"https://doi.org/10.1038/s41467-026-71035-5","venue":"Nature Communications","date":"2026-03-26","why":{"zh":"基准精度提升，未经田间育种验证","en":"Benchmark gain, not field-validated"},"fieldImpact":"incremental","aiImpact":"incremental"},{"title":"Robustly enhancing crop genomic prediction accuracy through ensemble learning and iterative optimization","titleZh":"GEG2P：集成学习提升五种作物基因组预测精度（Nat. Commun.）","id":"10.1038/s41467-026-75788-x","url":"https://doi.org/10.1038/s41467-026-75788-x","venue":"Nature Communications","date":"2026-07-20","why":{"zh":"集成学习提升五种作物预测精度","en":"Ensemble gains on five crops"},"fieldImpact":"incremental","aiImpact":"incremental"}],"latest_papers":[{"id":"arXiv:2610.29008","title":"Enhancing Cross-Grade Generalization in Banana Variety Classification via Hybrid CNN-Transformers and Knowledge Self-Distillation","titleZh":"用混合 CNN–Transformer 与知识自蒸馏提升香蕉品种跨等级分类泛化","summaryZh":"面向精准农业，针对成熟度、光照和品质等级变化带来的域偏移，提高香蕉品种自动分类的泛化能力。","summaryEn":"A hybrid CNN–Transformer with self-distillation improves banana variety classification across quality grades for precision agriculture.","authors":["Khush Kataruka","Mohammad Muzammila","Zahid Akhtar","Raghavendra Ramachandra"],"date":"2026-10-21","url":"https://arxiv.org/abs/2610.29008","site":"arXiv","venue":null,"category":"cs.CV","fetchedAt":"2026-10-22T10:08:38"},{"id":"arXiv:2610.26597","title":"Bombadil: A Foundation Model for Plant Electrophysiology","titleZh":"Bombadil：植物电生理基础模型","summaryZh":"植物像动物一样用电信号传递生理状态；以往模型各自只做单一胁迫预测，Bombadil 作为基础模型可跨任务、物种和环境迁移。","summaryEn":"Bombadil is a foundation model for plant electrophysiology that transfers across tasks, species and environments.","authors":["James Kirk","Piotr Bednarski","Filip Twardy","Justin Mast"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.26597","site":"arXiv","venue":null,"category":"cs.LG","fetchedAt":"2026-10-22T10:08:38"},{"id":"arXiv:2610.26209","title":"OceanTensorCube and ST-MMT: A Multimodal Data and Transformer System for Seven-Day Forecasting of Pyropia Bleaching","titleZh":"OceanTensorCube 与 ST-MMT：预测紫菜白化的多模态数据与 Transformer 系统","summaryZh":"整合 32 个海洋、气象和卫星数据通道，用 920 万参数时空 Transformer 提前 7 天预测韩国紫菜养殖白化。","summaryEn":"A spatiotemporal transformer forecasts Pyropia (laver) bleaching seven days ahead for Korean farms.","authors":["Gwangjin Kim","Cheolyoung Jang","Yosep Jo"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.26209","site":"arXiv","venue":null,"category":"cs.LG","fetchedAt":"2026-10-22T10:08:38"},{"id":"arXiv:2610.23609","title":"ADA 2025: A Multimodal Dataset for Water-Deficit Stress Detection in Sugar Beets","titleZh":"ADA 2025：甜菜水分亏缺胁迫检测的多模态数据集","summaryZh":"针对作物缺水胁迫检测公开数据稀少且多为单一 RGB 的问题，发布甜菜多模态数据集，支持 AI 早期检测研究。","summaryEn":"A multimodal sugar-beet dataset for AI-based early water-deficit stress detection, beyond RGB-only data.","authors":["Piper Powell","Robert Franz","Amos Smith","Isaak Ihorst"],"date":"2026-10-16","url":"https://arxiv.org/abs/2610.23609","site":"arXiv","venue":null,"category":"cs.CV","fetchedAt":"2026-10-22T10:08:38"},{"id":"arXiv:2610.22991","title":"A100: A Hybrid Vision Foundation Model for Agricultural Plant Image Analysis","titleZh":"A100：农业植物图像分析的混合视觉基础模型","summaryZh":"在覆盖 100 多种作物的 300 万张无标注田间图像上自监督预训练，冻结骨干即优于 ImageNet 监督特征。","summaryEn":"A100, pretrained on 3M unlabelled field images of 100+ crops, beats ImageNet features for agriculture.","authors":["Mayur Pratim Das","Ajai Dubey","Aksshay Mathew P","Sana Afreen"],"date":"2026-10-16","url":"https://arxiv.org/abs/2610.22991","site":"arXiv","venue":null,"category":"cs.CV","fetchedAt":"2026-10-22T10:08:38"},{"id":"arXiv:2610.22415","title":"Deep Learning for Forest Plantation Mapping in Godavari Districts of Andhra Pradesh, India","titleZh":"深度学习绘制印度安得拉邦戈达瓦里地区人工林","summaryZh":"用深度学习从卫星影像中识别小面积人工林，研究东南亚林地覆盖变化。","summaryEn":"Deep learning maps small forest plantations in Andhra Pradesh from satellite imagery.","authors":["Snehal More","Anuj Karpatne","Valerie A. Thomas","Randolph H. Wynne"],"date":"2026-10-14","url":"https://arxiv.org/abs/2610.22415","site":"arXiv","venue":null,"category":"eess.IV","fetchedAt":"2026-10-22T10:08:38"},{"id":"arXiv:2610.20544","title":"Detection of Potato Plant Leaf Diseases using Convolutional Neural Networks","titleZh":"用卷积神经网络检测马铃薯叶片病害","summaryZh":"轻量 CNN 区分健康、早疫病和晚疫病，在埃塞俄比亚田间独立数据集上准确率 90.28%。","summaryEn":"A lightweight CNN classifies potato leaf diseases with 90.28% accuracy on Ethiopian field images.","authors":["Haimanot Bitew Alemayehu","Simachew Melaku Ayalew","Chilotie Desalew Mengesha","Ayisheshim Almaw Bogale"],"date":"2026-10-14","url":"https://arxiv.org/abs/2610.20544","site":"arXiv","venue":null,"category":"cs.LG","fetchedAt":"2026-10-22T10:08:38"}]},{"id":"computer-science","slug":"computer-science","category":"eng-computing","categoryName":{"zh":"工程与计算","en":"Engineering & computing"},"name":{"zh":"计算机科学（算法与理论）","en":"Computer science (algorithms & theory)"},"classification":{"oecdFos":"1.2","moe":"0812 计算机科学与技术"},"level":3,"levelKey":"transforming","levelLabel":{"zh":"系统改造","en":"Systemic transformation"},"trend":"up","reason":{"zh":"AI 已发现被实际采用的新算法（AlphaDev 排序已并入 LLVM libc++，AlphaTensor 矩阵乘法）、改进近似难度界（AlphaEvolve 2026）；The AI Scientist 论文登上 Nature（2026-03），AI 研究自动化本身成为热点。","en":"AI has found adopted algorithms (AlphaDev sorting merged into LLVM libc++; AlphaTensor) and improved hardness bounds (AlphaEvolve 2026); The AI Scientist appeared in Nature (Mar 2026)."},"reasonSources":[{"url":"https://www.nature.com/articles/s41586-023-06004-9","linkCheck":"ok"},{"url":"https://research.google/blog/ai-as-a-research-partner-advancing-theoretical-computer-science-with-alphaevolve/","linkCheck":"ok"},{"url":"https://www.nature.com/articles/s41586-026-10265-5","linkCheck":"ok"}],"directions":{"zh":["算法发现","复杂性理论证明","AI 研究自动化","编译与系统优化"],"en":["Algorithm discovery","Complexity-theory proofs","Automated AI research","Compiler & systems optimisation"]},"stats":{"milestones":8,"milestonesSince2025":6,"milestones2026":3,"milestonesLast6Months":2,"news":5,"latestProgress":"2026-10-20","latestMilestone":"2026-10-07","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":0,"leaps":{"ai":2,"field":0},"latestPapers":6,"keyPapers":4},"milestones":[{"date":"2022-10-05","title":{"zh":"AlphaTensor 发现更快的矩阵乘法算法（Nature）","en":"AlphaTensor discovers faster matrix multiplication (Nature)"},"url":"https://www.nature.com/articles/s41586-022-05172-4","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"50 年来首次在 4×4 矩阵（模 2）上改进 Strassen 算法的乘法次数","en":"First improvement on Strassen for 4×4 (mod 2) in 50 years"},"aiImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"50 年来首次在 4×4 矩阵（模 2）上改进 Strassen 算法的乘法次数","en":"First improvement on Strassen for 4×4 (mod 2) in 50 years"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"50 年来首次在 4×4 矩阵（模 2）上改进 Strassen 算法的乘法次数","en":"First improvement on Strassen for 4×4 (mod 2) in 50 years"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-022-05172-4","title":"Discovering faster matrix multiplication algorithms with reinforcement learning","venue":"Nature","date":"2022-10-05"},"addedIn":"v1"},{"date":"2023-06-07","title":{"zh":"AlphaDev 发现更快排序算法，已并入 LLVM（Nature）","en":"AlphaDev finds faster sorting, merged into LLVM (Nature)"},"url":"https://www.nature.com/articles/s41586-023-06004-9","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"AI 发现的排序例程并入 LLVM libc++，每天被调用数万亿次","en":"AI-found sort routines merged into LLVM libc++"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI 发现的排序例程并入 LLVM libc++，每天被调用数万亿次","en":"AI-found sort routines merged into LLVM libc++"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI 发现的排序例程并入 LLVM libc++，每天被调用数万亿次","en":"AI-found sort routines merged into LLVM libc++"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-023-06004-9","title":"Faster sorting algorithms discovered using deep reinforcement learning","venue":"Nature","date":"2023-06-07"},"addedIn":"v1"},{"date":"2025-05-14","title":{"zh":"AlphaEvolve 发布：通用算法发现编码智能体","en":"AlphaEvolve: coding agent for algorithm discovery"},"url":"https://deepmind.google/discover/blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/","source":"Google DeepMind","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"system","typeLabel":{"zh":"系统发布","en":"System"},"why":{"zh":"通用进化式编码智能体，在算法和数学问题上多处超过已知最佳","en":"General evolutionary coding agent beating best-known results"},"aiImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"通用进化式编码智能体，在算法和数学问题上多处超过已知最佳","en":"General evolutionary coding agent beating best-known results"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"通用进化式编码智能体，在算法和数学问题上多处超过已知最佳","en":"General evolutionary coding agent beating best-known results"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2025-09-22","title":{"zh":"AlphaEvolve 发现新 gadget，MAX-4-CUT 近似难度界 0.9883→0.987、度量 TSP 117/116→111/110（arXiv 2509.18057）","en":"AlphaEvolve gadgets improve MAX-4-CUT (0.9883→0.987) and metric TSP (117/116→111/110) inapproximability"},"url":"https://arxiv.org/abs/2509.18057","source":"arXiv","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"成熟领域的复杂性下界被 AI 找到的结构改进，且结果可完全验证","en":"AI-found structures improve hardness bounds in a mature area, fully verifiable"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"成熟领域的复杂性下界被 AI 找到的结构改进，且结果可完全验证","en":"AI-found structures improve hardness bounds in a mature area, fully verifiable"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"成熟领域的复杂性下界被 AI 找到的结构改进，且结果可完全验证","en":"AI-found structures improve hardness bounds in a mature area, fully verifiable"}},"impactNotes":[],"verified":{"id":"arXiv:2509.18057","title":"Reinforced Generation of Combinatorial Structures: Hardness of Approximation","venue":"arXiv","date":"2025-09-22"},"addedIn":"v2"},{"date":"2025-09-30","title":{"zh":"Google Research 博客：用 AlphaEvolve 推进理论计算机科学","en":"Google Research: advancing theoretical CS with AlphaEvolve"},"url":"https://research.google/blog/ai-as-a-research-partner-advancing-theoretical-computer-science-with-alphaevolve/","source":"Google Research","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"同一成果的官方说明（原“待核实”日期已确认为 2025-09-30）","en":"Official write-up (date confirmed 2025-09-30)"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"同一成果的官方说明（原“待核实”日期已确认为 2025-09-30）","en":"Official write-up (date confirmed 2025-09-30)"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"同一成果的官方说明（原“待核实”日期已确认为 2025-09-30）","en":"Official write-up (date confirmed 2025-09-30)"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-03-25","title":{"zh":"The AI Scientist 端到端自动化 AI 研究论文发表于 Nature","en":"The AI Scientist end-to-end automated research published in Nature"},"url":"https://www.nature.com/articles/s41586-026-10265-5","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"端到端自动化研究系统登上 Nature，但产出论文质量有限","en":"End-to-end automated research in Nature; output quality limited"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"端到端自动化研究系统登上 Nature，但产出论文质量有限","en":"End-to-end automated research in Nature; output quality limited"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"端到端自动化研究系统登上 Nature，但产出论文质量有限","en":"End-to-end automated research in Nature; output quality limited"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-026-10265-5","title":"Towards end-to-end automation of AI research","venue":"Nature","date":"2026-03-25"},"addedIn":"v1"},{"date":"2026-10-01","title":{"zh":"KLS 猜想获证（AI 智能体参与探索）：高维对数凹分布采样与随机游走混合时间的维度无关保证","en":"KLS conjecture resolved with AI-agent assistance: dimension-free guarantees for log-concave sampling and random-walk mixing"},"url":"https://www.microsoft.com/en-us/research/blog/solving-a-31-year-old-grand-challenge-in-mathematics-the-kls-conjecture/","source":"Microsoft Research","flags":["preprint","vendor"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"},{"zh":"厂商自述","en":"Vendor claim"}],"linkCheck":"ok","datePrecision":"day","type":"discovery","typeLabel":{"zh":"科学发现","en":"Discovery"},"why":{"zh":"AI：微软称人类主导的 AI 智能体参与了证明搜索（厂商自述）；学科：KLS 决定高维随机游走的混合速度，是凸体体积计算和对数凹采样算法复杂度分析的核心障碍，现由独立证明（arXiv 2610.05474）支持，但尚未经同行评审","en":"AI: Microsoft says human-directed AI agents took part in the proof search (vendor claim). Field: KLS governs mixing of high-dimensional random walks, central to volume computation and log-concave sampling complexity; backed by an independent proof (arXiv 2610.05474) but not yet refereed"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：微软称人类主导的 AI 智能体参与了证明搜索（厂商自述）；学科：KLS 决定高维随机游走的混合速度，是凸体体积计算和对数凹采样算法复杂度分析的核心障碍，现由独立证明（arXiv 2610.05474）支持，但尚未经同行评审","en":"AI: Microsoft says human-directed AI agents took part in the proof search (vendor claim). Field: KLS governs mixing of high-dimensional random walks, central to volume computation and log-concave sampling complexity; backed by an independent proof (arXiv 2610.05474) but not yet refereed"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：微软称人类主导的 AI 智能体参与了证明搜索（厂商自述）；学科：KLS 决定高维随机游走的混合速度，是凸体体积计算和对数凹采样算法复杂度分析的核心障碍，现由独立证明（arXiv 2610.05474）支持，但尚未经同行评审","en":"AI: Microsoft says human-directed AI agents took part in the proof search (vendor claim). Field: KLS governs mixing of high-dimensional random walks, central to volume computation and log-concave sampling complexity; backed by an independent proof (arXiv 2610.05474) but not yet refereed"}},"impactNotes":[],"verified":null,"addedIn":"v2-20261022pm"},{"date":"2026-10-07","title":{"zh":"Quanta：AI 逼近 Unique Games 证明，研究者抢先完成","en":"Quanta: as AI closed in on a Unique Games proof, researchers raced to beat it"},"url":"https://www.quantamagazine.org/as-ai-closed-in-on-unique-games-proof-researchers-raced-to-beat-the-machines-20261007/","source":"Quanta Magazine","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"AI 逼近一项重要复杂性证明，促使研究者抢先完成","en":"AI nearly completes a major complexity proof"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI 逼近一项重要复杂性证明，促使研究者抢先完成","en":"AI nearly completes a major complexity proof"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI 逼近一项重要复杂性证明，促使研究者抢先完成","en":"AI nearly completes a major complexity proof"}},"impactNotes":[],"verified":null,"addedIn":"v1"}],"papersRepos":[{"title":"SakanaAI/AI-Scientist-v2","url":"https://github.com/SakanaAI/AI-Scientist-v2","date":null,"kind":"repo","linkCheck":"ok"},{"title":"google-deepmind/alphaevolve_results","url":"https://github.com/google-deepmind/alphaevolve_results","date":null,"kind":"repo","linkCheck":"ok"},{"title":"Discovering faster matrix multiplication algorithms with reinforcement learning","url":"https://doi.org/10.1038/s41586-022-05172-4","date":"2022-10-05","kind":"paper","linkCheck":"ok-crossref"},{"title":"Faster sorting algorithms discovered using deep reinforcement learning","url":"https://doi.org/10.1038/s41586-023-06004-9","date":"2023-06-07","kind":"paper","linkCheck":"ok-crossref"},{"title":"Reinforced Generation of Combinatorial Structures: Hardness of Approximation","url":"https://arxiv.org/abs/2509.18057","date":"2025-09-22","kind":"paper","linkCheck":"ok-crossref"},{"title":"Towards end-to-end automation of AI research","url":"https://doi.org/10.1038/s41586-026-10265-5","date":"2026-03-25","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"Google DeepMind","url":"https://deepmind.google"},{"name":"Google Research","url":"https://research.google"},{"name":"Sakana AI","url":"https://sakana.ai"}],"news":[{"id":"373ace93ecaf","date":"2026-10-20","title":"TRACE: Governing AI-Assisted Research with Principles from Computer Science","url":"https://arxiv.org/abs/2610.24035","source":"arXiv q-fin","lang":"en","kind":"paper","summaryZh":"arXiv q-fin于2026-10-20发布的论文，涉及AI 金融、AI 安全，原题《TRACE: Governing AI-Assisted Research with Principles from Computer Science》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"1883b786e5fb","date":"2026-10-20","title":"Discuss More, Comment Less: Boosting Automated Research via Multi-Perspective Collaborative Review","url":"https://arxiv.org/abs/2610.24050","source":"arXiv cs.CL","lang":"en","kind":"paper","summaryZh":"arXiv cs.CL于2026-10-20发布的论文，涉及AI 综合、AI 科研与调研、AI 软件开发，原题《Discuss More, Comment Less: Boosting Automated Research via Multi-Perspective Collaborativ》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"c5e2d61487ec","date":"2026-10-07","title":"As AI Closed In on ‘Unique Games’ Proof, Researchers Raced to Beat the Machines","url":"https://www.quantamagazine.org/as-ai-closed-in-on-unique-games-proof-researchers-raced-to-beat-the-machines-20261007/","source":"Quanta Magazine","lang":"en","kind":"news","summaryZh":"Quanta 报道：在传闻 AI 将证明唯一博弈猜想相关结果的压力下，三位计算机科学家抢先发表了自己的里程碑成果。","via":"research-samples-20261010","linkCheck":"from-feed"},{"id":"356df494567e","date":"2026-10-07","title":"Open No More","url":"https://blog.computationalcomplexity.org/2026/10/open-no-more.html","source":"Computational Complexity","lang":"en","kind":"news","summaryZh":"Lance Fortnow 谈 OpenAI 发布的 722 份手稿涉及多个理论计算机开放问题，原有“开放问题”格局被打破。","via":"research-samples-20261010","linkCheck":"from-feed"},{"id":"fdaf731bb50e","date":"2026-10-05","title":"Postdoc position on formal verification and algorithm discovery for numerical analysis","url":"https://terrytao.wordpress.com/2026/10/05/postdoc-position-on-formal-verification-and-algorithm-discovery-for-numerical-analysis/","source":"Terence Tao blog","lang":"en","kind":"news","summaryZh":"EPFL 招聘博士后，方向为数值分析中的形式化验证与算法发现。","via":"research-samples-20261010","linkCheck":"from-feed"}],"review":{"lastReviewed":"2026-10-22","cycleDays":60,"nextReview":"2026-12-20"},"urls":{"zh":"/zh-hans/disciplines/computer-science/","en":"/en/disciplines/computer-science/"},"dimensions":{"method":{"score":3,"label":{"zh":"研究方法","en":"Research method"},"evidence":"大模型辅助的算法设计与证明搜索在理论计算机科学快速普及","source":"https://research.google/blog/ai-as-a-research-partner-advancing-theoretical-computer-science-with-alphaevolve/","weight":0.3},"discovery":{"score":3,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"AlphaDev 排序进入 LLVM，AlphaTensor、AlphaEvolve 改进 MAX-4-CUT、TSP 近似难度界","source":"https://arxiv.org/abs/2509.18057","weight":0.35},"adoption":{"score":3,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"研究者普遍使用 AI 编程与写作工具（见软件工程维度的开发者调查）","source":"https://survey.stackoverflow.co/2026/ai","weight":0.25},"education":{"score":2,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"计算机课程普遍调整 AI 使用政策，但缺系统数据","source":null,"weight":0.1}},"compositeScore":2.9,"levelCapped":false,"levelChange":{"from":3,"to":3,"direction":"same","reason":null},"biggestBreakthrough":{"date":"2025-05-14","title":{"zh":"AlphaEvolve 发布：通用算法发现编码智能体","en":"AlphaEvolve: coding agent for algorithm discovery"},"url":"https://deepmind.google/discover/blog/alphaevolve-a-gemini-powered-coding-agent-for-designing-advanced-algorithms/","aiImpact":"leap","fieldImpact":"major","why":{"zh":"通用进化式编码智能体，在算法和数学问题上多处超过已知最佳","en":"General evolutionary coding agent beating best-known results"}},"keyPapers":[{"title":"Discovering faster matrix multiplication algorithms with reinforcement learning","titleZh":"AlphaTensor 发现更快的矩阵乘法算法（Nature）","id":"10.1038/s41586-022-05172-4","url":"https://doi.org/10.1038/s41586-022-05172-4","venue":"Nature","date":"2022-10-05","why":{"zh":"50 年来首次在 4×4 矩阵（模 2）上改进 Strassen 算法的乘法次数","en":"First improvement on Strassen for 4×4 (mod 2) in 50 years"},"fieldImpact":"major","aiImpact":"leap"},{"title":"Faster sorting algorithms discovered using deep reinforcement learning","titleZh":"AlphaDev 发现更快排序算法，已并入 LLVM（Nature）","id":"10.1038/s41586-023-06004-9","url":"https://doi.org/10.1038/s41586-023-06004-9","venue":"Nature","date":"2023-06-07","why":{"zh":"AI 发现的排序例程并入 LLVM libc++，每天被调用数万亿次","en":"AI-found sort routines merged into LLVM libc++"},"fieldImpact":"major","aiImpact":"major"},{"title":"Reinforced Generation of Combinatorial Structures: Hardness of Approximation","titleZh":"AlphaEvolve 发现新 gadget，MAX-4-CUT 近似难度界 0.9883→0.987、度量 TSP 117/116→111/110（arXiv 2509.18057）","id":"arXiv:2509.18057","url":"https://arxiv.org/abs/2509.18057","venue":"arXiv","date":"2025-09-22","why":{"zh":"成熟领域的复杂性下界被 AI 找到的结构改进，且结果可完全验证","en":"AI-found structures improve hardness bounds in a mature area, fully verifiable"},"fieldImpact":"major","aiImpact":"major"},{"title":"Towards end-to-end automation of AI research","titleZh":"The AI Scientist 端到端自动化 AI 研究论文发表于 Nature","id":"10.1038/s41586-026-10265-5","url":"https://doi.org/10.1038/s41586-026-10265-5","venue":"Nature","date":"2026-03-25","why":{"zh":"端到端自动化研究系统登上 Nature，但产出论文质量有限","en":"End-to-end automated research in Nature; output quality limited"},"fieldImpact":"incremental","aiImpact":"major"}],"latest_papers":[{"id":"arXiv:2610.21785","title":"The Six-Dimensional Tarski Problem is $Ω(\\log^3 n)$-Hard","titleZh":"六维 Tarski 问题需要 Ω(log³n) 次查询","summaryZh":"改进单调函数不动点的确定性查询复杂度下界，在六维得到最优的 Θ((log n)³)；文中指出近期已有多篇论文借助 AI 辅助证明改进了下界。","summaryEn":"Improves lower bounds for finding Tarski fixed points, giving optimal Θ((log n)^3) in six dimensions, after recent AI-assisted proofs.","authors":["Kristóf Zólomy"],"date":"2026-10-15","url":"https://arxiv.org/abs/2610.21785","site":"arXiv","venue":null,"category":"cs.CC","fetchedAt":"2026-10-22T10:08:48"},{"id":"arXiv:2610.19139","title":"Online Tokenization with Optimal Delay in Linear Time","titleZh":"线性时间、最优延迟的在线分词","summaryZh":"为大模型常用的 BPE 分词流程设计在线算法，在线性时间内以最优延迟输出，降低延迟和内存占用。","summaryEn":"An online BPE tokenisation algorithm with optimal delay in linear time, reducing latency and memory.","authors":["Mina Dalirrooyfard","Christopher Ye"],"date":"2026-10-13","url":"https://arxiv.org/abs/2610.19139","site":"arXiv","venue":null,"category":"cs.DS","fetchedAt":"2026-10-22T10:08:48"},{"id":"arXiv:2610.16289","title":"Fast Quantized Johnson$-$Lindenstrauss Transforms","titleZh":"快速量化 Johnson-Lindenstrauss 变换","summaryZh":"用结构化随机变换替代稠密高斯投影，加速用于压缩大模型 KV 缓存的 1 比特量化 JL 变换。","summaryEn":"Faster structured quantised JL transforms for 1-bit KV-cache compression in LLM inference.","authors":["Angelos Korakitis","Vasileios Nakos"],"date":"2026-10-12","url":"https://arxiv.org/abs/2610.16289","site":"arXiv","venue":null,"category":"cs.LG","fetchedAt":"2026-10-22T10:08:48"},{"id":"arXiv:2610.16144","title":"Hashrope: A Persistent Data Structure for Mechanisms of LLM Context Management","titleZh":"Hashrope：面向大模型上下文管理的持久化数据结构","summaryZh":"证明大模型服务中增量编辑、分支快照、前缀复用和重复片段编码四类机制可统一为一种带哈希的持久化平衡绳结构。","summaryEn":"Shows four LLM context-management mechanisms reduce to one hashed persistent weight-balanced rope.","authors":["Muntaser Syed","Ameerah Alsulami","Marius Silaghi"],"date":"2026-10-12","url":"https://arxiv.org/abs/2610.16144","site":"arXiv","venue":null,"category":"cs.DS","fetchedAt":"2026-10-22T10:08:48"},{"id":"arXiv:2610.15579","title":"Approximation Algorithms for Dual-Memory Request Scheduling in Disaggregated LLM Inference","titleZh":"分离式大模型推理中双内存请求调度的近似算法","summaryZh":"首次形式化预填充与解码分离部署中的耦合调度问题，并给出近似保证。","summaryEn":"Formalises coupled prefill/decode memory scheduling in disaggregated LLM inference with approximation guarantees.","authors":["Xin Wang","Hong Shen","Hui Tian","Ye Tao"],"date":"2026-10-11","url":"https://arxiv.org/abs/2610.15579","site":"arXiv","venue":null,"category":"cs.DS","fetchedAt":"2026-10-22T10:08:48"},{"id":"arXiv:2610.15254","title":"Cascade: GPU-Accelerated Out-of-Core Suffix Array Construction for LLM Corpora","titleZh":"Cascade：面向大模型语料的 GPU 加速外存后缀数组构建","summaryZh":"后缀数组用于大模型训练语料去重、污染检测与数据溯源；Cascade 在显存放不下时仍用 GPU 加速构建，大幅缩短建索引时间。","summaryEn":"GPU-accelerated out-of-core suffix-array construction for LLM corpora used in deduplication, contamination tests and attribution.","authors":["Jamshed Khan","Prashant Pandey"],"date":"2026-10-11","url":"https://arxiv.org/abs/2610.15254","site":"arXiv","venue":null,"category":"cs.DS","fetchedAt":"2026-10-22T10:08:48"}]},{"id":"software-engineering","slug":"software-engineering","category":"eng-computing","categoryName":{"zh":"工程与计算","en":"Engineering & computing"},"name":{"zh":"软件工程","en":"Software engineering"},"classification":{"oecdFos":"1.2","moe":"0835 软件工程"},"level":3,"levelKey":"transforming","levelLabel":{"zh":"系统改造","en":"Systemic transformation"},"trend":"up","reason":{"zh":"下调为 L3：采用度确实最高——Stack Overflow 2026 独立调查显示超过 70% 的受访者每天使用 AI；Google（约 75%）和 Anthropic（90% 以上）代码占比是厂商自述，不能单独支撑 L4。METR 随机试验发现 2025 年初资深开源开发者用 AI 反而慢 19%，2026 年新研究估计提速但方法受限，生产率与质量的独立证据仍不一致。","en":"Lowered to L3: adoption is the highest of any field (Stack Overflow 2026: over 70% of respondents use AI daily), but Google’s ~75% and Anthropic’s 90%+ code shares are vendor claims and cannot alone justify L4. METR’s RCT found experienced OSS developers 19% slower in early 2025; 2026 follow-ups estimate speed-ups but with design limits, so independent productivity evidence is mixed."},"reasonSources":[{"url":"https://survey.stackoverflow.co/2026/ai","linkCheck":"ok"},{"url":"https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/","linkCheck":"ok"},{"url":"https://metr.org/blog/2026-02-24-uplift-update/","linkCheck":"ok"}],"directions":{"zh":["编码智能体","自动修复与测试","代码评审","需求到部署全流程"],"en":["Coding agents","Automated repair & testing","Code review","Spec-to-deploy automation"]},"stats":{"milestones":10,"milestonesSince2025":5,"milestones2026":4,"milestonesLast6Months":3,"news":12,"latestProgress":"2026-10-20","latestMilestone":"2026-10-06","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":0,"leaps":{"ai":0,"field":0},"latestPapers":8,"keyPapers":5},"milestones":[{"date":"2021-06-29","title":{"zh":"GitHub Copilot 技术预览发布","en":"GitHub Copilot technical preview"},"url":"https://github.blog/news-insights/product-news/introducing-github-copilot-ai-pair-programmer/","source":"GitHub Blog","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"system","typeLabel":{"zh":"系统发布","en":"System"},"why":{"zh":"首个大规模商用 AI 编程助手","en":"First mass-market AI coding assistant"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首个大规模商用 AI 编程助手","en":"First mass-market AI coding assistant"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首个大规模商用 AI 编程助手","en":"First mass-market AI coding assistant"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2021-07-07","title":{"zh":"Codex 论文与 HumanEval 基准：大模型从文档字符串生成代码（arXiv 2107.03374）","en":"Codex and HumanEval: evaluating LLMs trained on code"},"url":"https://arxiv.org/abs/2107.03374","source":"arXiv","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok-crossref","datePrecision":"day","type":"benchmark","typeLabel":{"zh":"基准","en":"Benchmark"},"why":{"zh":"代码大模型路线的起点，直接催生 GitHub Copilot","en":"Origin of code LLMs and Copilot"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"代码大模型路线的起点，直接催生 GitHub Copilot","en":"Origin of code LLMs and Copilot"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"代码大模型路线的起点，直接催生 GitHub Copilot","en":"Origin of code LLMs and Copilot"}},"impactNotes":[{"zh":"AI 影响原评“飞跃性突破”，因预印本，未同行评审按规则降为“重要进展”","en":"AI impact downgraded from Leap to Major because: Preprint, not peer-reviewed"}],"verified":{"id":"arXiv:2107.03374","title":"Evaluating Large Language Models Trained on Code","venue":"arXiv","date":"2021-07-07"},"addedIn":"v2"},{"date":"2022-12-09","title":{"zh":"AlphaCode：竞赛级代码生成，在 Codeforces 达到约前 54% 水平（Science）","en":"AlphaCode: competition-level code generation (Science)"},"url":"https://doi.org/10.1126/science.abq1158","source":"Science","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"competition","typeLabel":{"zh":"竞赛","en":"Competition"},"why":{"zh":"首次在编程竞赛中达到中等人类选手水平","en":"First median-human level in programming contests"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首次在编程竞赛中达到中等人类选手水平","en":"First median-human level in programming contests"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"首次在编程竞赛中达到中等人类选手水平","en":"First median-human level in programming contests"}},"impactNotes":[],"verified":{"id":"10.1126/science.abq1158","title":"Competition-level code generation with AlphaCode","venue":"Science","date":"2022-12-09"},"addedIn":"v2"},{"date":"2023-02-13","title":{"zh":"Copilot 随机对照实验：使用者完成任务快 55.8%（arXiv 2302.06590）","en":"Copilot RCT: developers finish a task 55.8% faster"},"url":"https://arxiv.org/abs/2302.06590","source":"arXiv","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"早期生产率证据，但任务简单、样本小","en":"Early productivity evidence on a simple task"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"早期生产率证据，但任务简单、样本小","en":"Early productivity evidence on a simple task"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"早期生产率证据，但任务简单、样本小","en":"Early productivity evidence on a simple task"}},"impactNotes":[],"verified":{"id":"arXiv:2302.06590","title":"The Impact of AI on Developer Productivity: Evidence from GitHub Copilot","venue":"arXiv","date":"2023-02-13"},"addedIn":"v2"},{"date":"2023-10-10","title":{"zh":"SWE-bench：真实 GitHub issue 修复基准（ICLR 2024）","en":"SWE-bench: resolving real GitHub issues (ICLR 2024)"},"url":"https://arxiv.org/abs/2310.06770","source":"arXiv","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"benchmark","typeLabel":{"zh":"基准","en":"Benchmark"},"why":{"zh":"真实仓库 issue 修复基准，成为编码智能体的主标尺","en":"Real-repo issue benchmark; main yardstick for coding agents"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"真实仓库 issue 修复基准，成为编码智能体的主标尺","en":"Real-repo issue benchmark; main yardstick for coding agents"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"真实仓库 issue 修复基准，成为编码智能体的主标尺","en":"Real-repo issue benchmark; main yardstick for coding agents"}},"impactNotes":[],"verified":{"id":"arXiv:2310.06770","title":"SWE-bench: Can Language Models Resolve Real-World GitHub Issues?","venue":"arXiv","date":"2023-10-10"},"addedIn":"v1"},{"date":"2025-07-12","title":{"zh":"METR 随机对照试验：2025 年初 AI 让资深开源开发者慢 19%（arXiv 2507.09089）","en":"METR RCT: early-2025 AI made experienced OSS developers 19% slower"},"url":"https://arxiv.org/abs/2507.09089","source":"arXiv","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"重要反证：开发者自认提速 20%，实测反而变慢，挑战“AI 大幅提效”的叙事","en":"Developers believed +20% but measured −19%"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"重要反证：开发者自认提速 20%，实测反而变慢，挑战“AI 大幅提效”的叙事","en":"Developers believed +20% but measured −19%"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"重要反证：开发者自认提速 20%，实测反而变慢，挑战“AI 大幅提效”的叙事","en":"Developers believed +20% but measured −19%"}},"impactNotes":[],"verified":{"id":"arXiv:2507.09089","title":"Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity","venue":"arXiv","date":"2025-07-12"},"addedIn":"v2"},{"date":"2026-02-24","title":{"zh":"METR 更新：2025 年末后续试验估计提速 4%–18%，但置信区间跨零，试验设计需改变","en":"METR update: late-2025 follow-up estimates 4–18% speed-ups, CIs span zero; design to change"},"url":"https://metr.org/blog/2026-02-24-uplift-update/","source":"METR","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"说明独立测量 AI 生产率仍很难","en":"Shows how hard independent productivity measurement remains"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"说明独立测量 AI 生产率仍很难","en":"Shows how hard independent productivity measurement remains"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"说明独立测量 AI 生产率仍很难","en":"Shows how hard independent productivity measurement remains"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-04-22","title":{"zh":"Google：约 75% 新代码由 AI 生成","en":"Google: ~75% of new code is AI-generated"},"url":"https://www.theverge.com/tech/917163/google-says-75-percent-of-all-its-new-code-is-ai-generated","source":"The Verge","flags":["vendor"],"flagLabels":[{"zh":"厂商自述","en":"Vendor claim"}],"linkCheck":"ok","datePrecision":"day","type":"industry","typeLabel":{"zh":"产业/采用","en":"Industry"},"why":{"zh":"Google 称约 75% 新代码由 AI 生成（厂商自述）","en":"Vendor claim: ~75% of new code AI-generated"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"Google 称约 75% 新代码由 AI 生成（厂商自述）","en":"Vendor claim: ~75% of new code AI-generated"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"Google 称约 75% 新代码由 AI 生成（厂商自述）","en":"Vendor claim: ~75% of new code AI-generated"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-05-14","title":{"zh":"Anthropic：超过 90% 的代码由 Claude 编写","en":"Anthropic: 90%+ of its code written by Claude"},"url":"https://www.techspot.com/news/112408-anthropic-more-than-90-code-now-written-ai.html","source":"TechSpot","flags":["vendor"],"flagLabels":[{"zh":"厂商自述","en":"Vendor claim"}],"linkCheck":"ok","datePrecision":"day","type":"industry","typeLabel":{"zh":"产业/采用","en":"Industry"},"why":{"zh":"Anthropic 称超过 90% 代码由 Claude 编写（厂商自述）","en":"Vendor claim: 90%+ of code written by Claude"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"Anthropic 称超过 90% 代码由 Claude 编写（厂商自述）","en":"Vendor claim: 90%+ of code written by Claude"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"Anthropic 称超过 90% 代码由 Claude 编写（厂商自述）","en":"Vendor claim: 90%+ of code written by Claude"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-10-06","title":{"zh":"Stack Overflow 2026 开发者调查（3 万多人）：超过 70% 每天用 AI，31% 的日常用户每天用 4 小时以上；48% 只在能轻松验证时才信任 AI","en":"Stack Overflow 2026 survey (30k+): over 70% use AI daily, 31% of them 4+ h/day; 48% trust AI only when easily verifiable"},"url":"https://stackoverflow.blog/2026/10/06/the-results-of-the-2026-developer-survey-are-here/","source":"Stack Overflow","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"industry","typeLabel":{"zh":"产业/采用","en":"Industry"},"why":{"zh":"独立的大样本采用率数据，支持“采用维度 4 分”","en":"Independent large-sample adoption data"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"独立的大样本采用率数据，支持“采用维度 4 分”","en":"Independent large-sample adoption data"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"独立的大样本采用率数据，支持“采用维度 4 分”","en":"Independent large-sample adoption data"}},"impactNotes":[],"verified":null,"addedIn":"v2"}],"papersRepos":[{"title":"SWE-bench/SWE-bench","url":"https://github.com/SWE-bench/SWE-bench","date":null,"kind":"repo","linkCheck":"ok"},{"title":"openai/codex","url":"https://github.com/openai/codex","date":null,"kind":"repo","linkCheck":"ok"},{"title":"Evaluating Large Language Models Trained on Code","url":"https://arxiv.org/abs/2107.03374","date":"2021-07-07","kind":"paper","linkCheck":"ok-crossref"},{"title":"Competition-level code generation with AlphaCode","url":"https://doi.org/10.1126/science.abq1158","date":"2022-12-09","kind":"paper","linkCheck":"ok-crossref"},{"title":"The Impact of AI on Developer Productivity: Evidence from GitHub Copilot","url":"https://arxiv.org/abs/2302.06590","date":"2023-02-13","kind":"paper","linkCheck":"ok-crossref"},{"title":"SWE-bench: Can Language Models Resolve Real-World GitHub Issues?","url":"https://arxiv.org/abs/2310.06770","date":"2023-10-10","kind":"paper","linkCheck":"ok-crossref"},{"title":"Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity","url":"https://arxiv.org/abs/2507.09089","date":"2025-07-12","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"GitHub Copilot","url":"https://github.com/features/copilot"},{"name":"Anthropic Claude Code","url":"https://www.anthropic.com/claude-code"},{"name":"OpenAI Codex","url":"https://openai.com/codex/"},{"name":"Cursor","url":"https://cursor.com"}],"news":[{"id":"35182ebc2514","date":"2026-10-20","title":"Chris Lattner: AI Code Generation Doesn’t Solve Software Engineering","url":"https://www.eetimes.com/chris-lattner-ai-code-generation-doesnt-solve-software-engineering/","source":"EE Times","lang":"en","kind":"news","summaryZh":"EE Times于2026-10-20发布的文章，涉及AI 芯片硬件、AI 软件开发，原题《Chris Lattner: AI Code Generation Doesn’t Solve Software Engineering》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"ca10c9a53a9e","date":"2026-10-20","title":"Show HN: Code-mode – keep MCP tool results out of Claude Code's context","url":"https://github.com/gabe4coding/claude-code-mode","source":"hn-show-ai","lang":"en","kind":"news","summaryZh":"hn-show-ai于2026-10-20发布的文章，涉及AI 软件开发，原题《Show HN: Code-mode – keep MCP tool results out of Claude Code's context》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"04a2cdaeef42","date":"2026-10-20","title":"Show HN: Agent Archive – Session Handoffs and Backups for Claude Code and Codex","url":"https://github.com/wangjohn/agent-archive","source":"hn-show-ai","lang":"en","kind":"news","summaryZh":"hn-show-ai于2026-10-20发布的文章，涉及AI 软件开发，原题《Show HN: Agent Archive – Session Handoffs and Backups for Claude Code and Codex》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"eb477756520c","date":"2026-10-20","title":"How 3 Java Champions are using GitHub Copilot for increased productivity","url":"https://github.blog/ai-and-ml/github-copilot/how-3-java-champions-are-using-github-copilot-for-increased-productivity/","source":"The GitHub Blog","lang":"en","kind":"news","summaryZh":"The GitHub Blog于2026-10-21发布的文章，涉及AI 软件开发，原题《How 3 Java Champions are using GitHub Copilot for increased productivity》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"9d6368af5ec5","date":"2026-10-20","title":"微软设计师也开始借助 AI 编写 Win11 代码，加速修复界面不一致问题","url":"https://www.ithome.com/1/015/493.htm","source":"IT之家","lang":"zh","kind":"news","summaryZh":"IT之家于2026-10-20发布的文章，涉及AI 软件开发，原题《微软设计师也开始借助 AI 编写 Win11 代码，加速修复界面不一致问题》，详见原文链接。","via":"collector-run-20261020","linkCheck":"from-feed"},{"id":"1a89d0af8ada","date":"2026-10-20","title":"Linux 7.4 引入约 1.7 万行 Rust 内核代码，Zerocopy 一次跨越八个版本","url":"https://www.ithome.com/1/015/502.htm","source":"IT之家","lang":"zh","kind":"news","summaryZh":"IT之家于2026-10-20发布的文章，涉及AI 软件开发、AI 医疗、AI 芯片硬件，原题《Linux 7.4 引入约 1.7 万行 Rust 内核代码，Zerocopy 一次跨越八个版本》，详见原文链接。","via":"collector-run-20261020","linkCheck":"from-feed"},{"id":"b4efa3fc6b2c","date":"2026-10-20","title":"Cortex: A Fixed-Point Theory of Governed Coding Agents - Technical Report","url":"https://arxiv.org/abs/2610.24255","source":"arXiv cs.AI","lang":"en","kind":"paper","summaryZh":"arXiv cs.AI于2026-10-20发布的论文，涉及AI 科研与调研、AI 数学、AI 法律政策，原题《Cortex: A Fixed-Point Theory of Governed Coding Agents - Technical Report》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"abe45073afec","date":"2026-10-20","title":"Stress-Query and Temporal Reliability of Vision-Language Models as copilots in Autonomous UAV Navigation","url":"https://arxiv.org/abs/2610.23877","source":"arXiv cs.RO","lang":"en","kind":"paper","summaryZh":"arXiv cs.RO于2026-10-20发布的论文，涉及AI 机器人、AI 科研与调研，原题《Stress-Query and Temporal Reliability of Vision-Language Models as copilots in Autonomous 》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"d4ad5dc9e366","date":"2026-10-20","title":"How Effectively Do Security Skills Guide Coding Agents in Vulnerability Detection in Linux Kernel?","url":"https://arxiv.org/abs/2610.25705","source":"arXiv cs.CR","lang":"en","kind":"paper","summaryZh":"arXiv cs.CR于2026-10-20发布的论文，涉及AI 安全、AI 软件开发，原题《How Effectively Do Security Skills Guide Coding Agents in Vulnerability Detection in Linux》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"6062992532c8","date":"2026-10-19","title":"Claude Code AI agent caught a crypto miner hiding on a developer’s overheating Mac","url":"https://www.tomshardware.com/tech-industry/cryptomining/claude-code-ai-agent-caught-a-crypto-miner-hiding-on-a-developers-overheating-mac-xmrig-task-was-hogging-about-1175-percent-cpu-roughly-12-cores-pushing-temps-to-112-degrees","source":"Tom's Hardware","lang":"en","kind":"news","summaryZh":"Tom's Hardware于2026-10-19发布的文章，涉及AI 芯片硬件、AI 金融、AI 软件开发，原题《Claude Code AI agent caught a crypto miner hiding on a developer’s overheating Mac》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"d85dbe423bdb","date":"2026-10-19","title":"DeepSeek V4.1 Flash 更新：“许愿式编程”成真 + 搜索 API 开放","url":"https://www.oschina.net/news/503080","source":"开源中国","lang":"zh","kind":"news","summaryZh":"开源中国于2026-10-19发布的文章，涉及AI 软件开发，原题《DeepSeek V4.1 Flash 更新：“许愿式编程”成真 + 搜索 API 开放》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"ab85acc9f216","date":"2026-10-06","title":"The results of the 2026 Developer Survey are here!","url":"https://stackoverflow.blog/2026/10/06/the-results-of-the-2026-developer-survey-are-here/","source":"Stack Overflow Blog","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"ok"}],"review":{"lastReviewed":"2026-10-21","cycleDays":60,"nextReview":"2026-12-20"},"urls":{"zh":"/zh-hans/disciplines/software-engineering/","en":"/en/disciplines/software-engineering/"},"dimensions":{"method":{"score":4,"label":{"zh":"研究方法","en":"Research method"},"evidence":"编码智能体已是软件开发主流工具","source":"https://survey.stackoverflow.co/2026/ai","weight":0.3},"discovery":{"score":2,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"SWE-bench 等基准快速提升，但独立随机试验对生产率的结论不一（METR 2025 发现资深开发者变慢 19%）","source":"https://metr.org/blog/2025-07-10-early-2025-ai-experienced-os-dev-study/","weight":0.35},"adoption":{"score":4,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"Stack Overflow 2026 独立调查（3 万多人）：超过 70% 的受访者每天用 AI，其中 31% 每天 4 小时以上；编码智能体是首要用途（66%）","source":"https://survey.stackoverflow.co/2026/ai","weight":0.25},"education":{"score":2,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"编程教育普遍调整，缺系统数据","source":null,"weight":0.1}},"compositeScore":3.1,"levelCapped":false,"levelChange":{"from":4,"to":3,"direction":"down","reason":{"zh":"下调为 L3：采用度确实最高——Stack Overflow 2026 独立调查显示超过 70% 的受访者每天使用 AI；Google（约 75%）和 Anthropic（90% 以上）代码占比是厂商自述，不能单独支撑 L4。METR 随机试验发现 2025 年初资深开源开发者用 AI 反而慢 19%，2026 年新研究估计提速但方法受限，生产率与质量的独立证据仍不一致。","en":"Lowered to L3: adoption is the highest of any field (Stack Overflow 2026: over 70% of respondents use AI daily), but Google’s ~75% and Anthropic’s 90%+ code shares are vendor claims and cannot alone justify L4. METR’s RCT found experienced OSS developers 19% slower in early 2025; 2026 follow-ups estimate speed-ups but with design limits, so independent productivity evidence is mixed."}},"biggestBreakthrough":{"date":"2021-07-07","title":{"zh":"Codex 论文与 HumanEval 基准：大模型从文档字符串生成代码（arXiv 2107.03374）","en":"Codex and HumanEval: evaluating LLMs trained on code"},"url":"https://arxiv.org/abs/2107.03374","aiImpact":"major","fieldImpact":"major","why":{"zh":"代码大模型路线的起点，直接催生 GitHub Copilot","en":"Origin of code LLMs and Copilot"}},"keyPapers":[{"title":"Evaluating Large Language Models Trained on Code","titleZh":"Codex 论文与 HumanEval 基准：大模型从文档字符串生成代码（arXiv 2107.03374）","id":"arXiv:2107.03374","url":"https://arxiv.org/abs/2107.03374","venue":"arXiv","date":"2021-07-07","why":{"zh":"代码大模型路线的起点，直接催生 GitHub Copilot","en":"Origin of code LLMs and Copilot"},"fieldImpact":"major","aiImpact":"major"},{"title":"Competition-level code generation with AlphaCode","titleZh":"AlphaCode：竞赛级代码生成，在 Codeforces 达到约前 54% 水平（Science）","id":"10.1126/science.abq1158","url":"https://doi.org/10.1126/science.abq1158","venue":"Science","date":"2022-12-09","why":{"zh":"首次在编程竞赛中达到中等人类选手水平","en":"First median-human level in programming contests"},"fieldImpact":"incremental","aiImpact":"major"},{"title":"The Impact of AI on Developer Productivity: Evidence from GitHub Copilot","titleZh":"Copilot 随机对照实验：使用者完成任务快 55.8%（arXiv 2302.06590）","id":"arXiv:2302.06590","url":"https://arxiv.org/abs/2302.06590","venue":"arXiv","date":"2023-02-13","why":{"zh":"早期生产率证据，但任务简单、样本小","en":"Early productivity evidence on a simple task"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"SWE-bench: Can Language Models Resolve Real-World GitHub Issues?","titleZh":"SWE-bench：真实 GitHub issue 修复基准（ICLR 2024）","id":"arXiv:2310.06770","url":"https://arxiv.org/abs/2310.06770","venue":"arXiv","date":"2023-10-10","why":{"zh":"真实仓库 issue 修复基准，成为编码智能体的主标尺","en":"Real-repo issue benchmark; main yardstick for coding agents"},"fieldImpact":"incremental","aiImpact":"major"},{"title":"Measuring the Impact of Early-2025 AI on Experienced Open-Source Developer Productivity","titleZh":"METR 随机对照试验：2025 年初 AI 让资深开源开发者慢 19%（arXiv 2507.09089）","id":"arXiv:2507.09089","url":"https://arxiv.org/abs/2507.09089","venue":"arXiv","date":"2025-07-12","why":{"zh":"重要反证：开发者自认提速 20%，实测反而变慢，挑战“AI 大幅提效”的叙事","en":"Developers believed +20% but measured −19%"},"fieldImpact":"major","aiImpact":"incremental"}],"latest_papers":[{"id":"arXiv:2610.30047","title":"SWE-DIVE: Multi-Turn Coding Agent Evaluation with Dynamic User Simulation","titleZh":"SWE-DIVE：用动态用户模拟评测多轮编程智能体","summaryZh":"模拟真实用户不规范、逐步迭代的需求，评测编程智能体在多轮交互中的表现，而非单次规范提示。","summaryEn":"Evaluates coding agents in multi-turn sessions with simulated users giving under-specified, evolving requests.","authors":["Lorenz Brehme","Benjamin Steenhoek","Ruth Breu"],"date":"2026-10-21","url":"https://arxiv.org/abs/2610.30047","site":"arXiv","venue":null,"category":"cs.SE","fetchedAt":"2026-10-22T10:08:51"},{"id":"arXiv:2610.29870","title":"PAUSE: Search-Free Steering of Coding Agents Before Their First Source-Code Edit","titleZh":"PAUSE：在编程智能体首次改代码前做一次免搜索的引导","summaryZh":"在智能体首次编辑源码前插入一次规划干预，以低成本纠正方向，减少探索充分却提交失败补丁的情况。","summaryEn":"Steers coding agents with a single planning intervention before their first source edit, cheaply improving patch success.","authors":["Viral Patel","Boyin Tan","Jason Xue","Youcheng Sun"],"date":"2026-10-21","url":"https://arxiv.org/abs/2610.29870","site":"arXiv","venue":null,"category":"cs.SE","fetchedAt":"2026-10-22T10:08:51"},{"id":"arXiv:2610.28612","title":"TraceLite: Efficient Trace Reduction for SWE Agents via Optical Compression and Redundancy Removal","titleZh":"TraceLite：用光学压缩和去冗余精简软件工程智能体轨迹","summaryZh":"无需训练或额外大模型调用：最近的观测保留文本，历史观测渲染成紧凑图像并去除重叠内容，在 SWE-bench Verified 等任务上降低 token 消耗。","summaryEn":"Training-free compression of SWE-agent traces by rendering history as images and removing redundancy.","authors":["Jingyu Xiao","Qianxi Liu","Yintong Huo","Yuan Jiang"],"date":"2026-10-20","url":"https://arxiv.org/abs/2610.28612","site":"arXiv","venue":null,"category":"cs.SE","fetchedAt":"2026-10-22T10:08:51"},{"id":"arXiv:2610.28610","title":"InterWebCoder: Learning Interactive Webpage Generation from Executable Tests via Reinforcement Learning","titleZh":"InterWebCoder：用可执行测试和强化学习学习生成交互式网页","summaryZh":"录制真实交互轨迹，合成自包含网页代码和可执行测试脚本，再结合监督微调与群体相对策略优化直接优化功能正确性。","summaryEn":"Synthesises interactive webpages with executable tests and trains MLLMs with SFT plus GRPO for functional correctness.","authors":["Jingyu Xiao","Dasen Dai","Yuxuan Wan","Jen-tse Huang"],"date":"2026-10-20","url":"https://arxiv.org/abs/2610.28610","site":"arXiv","venue":null,"category":"cs.SE","fetchedAt":"2026-10-22T10:08:51"},{"id":"arXiv:2610.28160","title":"From Code Generator to Engineering Companion: A Plant-aware Agent for PLC Programming","titleZh":"从代码生成器到工程伙伴：面向 PLC 编程的产线感知智能体","summaryZh":"为可编程逻辑控制器编程设计对话式助手，引入产线专属上下文、对话式解释调试，以及从聊天到部署逐级放开的工具权限，关键操作由工程师把关。","summaryEn":"A plant-aware PLC programming copilot with graduated tool access that keeps engineers in control.","authors":["Joschka Kersting","Gesa Benndorf"],"date":"2026-10-20","url":"https://arxiv.org/abs/2610.28160","site":"arXiv","venue":null,"category":"cs.SE","fetchedAt":"2026-10-22T10:08:51"},{"id":"arXiv:2610.26416","title":"Co-Evolving Policy and Reward for Code Generation: A Mechanistic Study of Single-Adapter Drift","titleZh":"代码生成中策略与奖励的协同演化：单适配器漂移的机制研究","summaryZh":"研究把 SPARK 策略-奖励协同演化框架迁移到可执行验证奖励的代码生成强化学习中是否有效。","summaryEn":"Tests whether SPARK-style policy–reward co-evolution transfers to RL with verifiable rewards for code generation.","authors":["Amarsaikhan Batjargal","Noriko Tomuro"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.26416","site":"arXiv","venue":null,"category":"cs.CL","fetchedAt":"2026-10-22T10:08:51"},{"id":"arXiv:2610.26394","title":"When Coding Agents Outpace Review: Evaluating PRD-Driven Development in Mature Enterprise Repositories","titleZh":"当编码智能体快过评审：在成熟企业代码库中评估需求文档驱动开发","summaryZh":"提出 PACE 基准（43 个企业任务、750 个专家审核的代码点），衡量编码智能体能否完整实现产品需求文档。","summaryEn":"PACE benchmark (43 enterprise tasks, 750 audited code points) measures whether coding agents fulfil full PRDs.","authors":["Tianqi Liu","Yandong Guan","Songze Li","Wenkai Wang"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.26394","site":"arXiv","venue":null,"category":"cs.SE","fetchedAt":"2026-10-22T10:08:51"},{"id":"arXiv:2610.26001","title":"SPLASH: Contrastive Audit Pressure for Detecting and Localizing Security Hallucinations in Generated Code","titleZh":"SPLASH：用对比审计压力检测并定位生成代码中的安全幻觉","summaryZh":"大模型生成的代码常通过功能测试却不安全；SPLASH 对比安全审查与风格审查下的概率来定位这些“安全幻觉”。","summaryEn":"SPLASH contrasts security- vs style-review log-probabilities to detect and localise insecure LLM-generated code.","authors":["Mirza Masfiqur Rahman","MD Ajwad Akil","Imtiaz Karim","Ashish Kundu"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.26001","site":"arXiv","venue":null,"category":"cs.CR","fetchedAt":"2026-10-22T10:08:51"}]},{"id":"cybersecurity","slug":"cybersecurity","category":"eng-computing","categoryName":{"zh":"工程与计算","en":"Engineering & computing"},"name":{"zh":"网络安全","en":"Cybersecurity"},"classification":{"oecdFos":"2.2","moe":"0839 网络空间安全"},"level":3,"levelKey":"transforming","levelLabel":{"zh":"系统改造","en":"Systemic transformation"},"trend":"up","reason":{"zh":"AI 已规模化发现真实漏洞：Anthropic 称 Claude Opus 4.6 验证了 500 多个高危漏洞（2026-02），微软 FORGE 实验室 5–9 月参与发现 140 个 Windows CVE，新加坡 TitanCA 发现 203 个零日；谷歌首次发现疑似由 AI 开发的在野零日利用。攻防两端同时被改造，接近 4 级。","en":"AI finds real vulnerabilities at scale: Anthropic validated 500+ high-severity bugs with Claude Opus 4.6 (Feb 2026), Microsoft FORGE contributed to 140 Windows CVEs (May–Sep 2026), TitanCA found 203 zero-days; Google saw the first likely AI-developed zero-day exploit in the wild. Close to level 4."},"reasonSources":[{"url":"https://www.anthropic.com/research/zero-days","linkCheck":"ok"},{"url":"https://www.microsoft.com/en-us/security/blog/2026/10/07/3-lessons-from-frontier-ai-vulnerability-research/","linkCheck":"ok"},{"url":"https://arxiv.org/abs/2604.17860","linkCheck":"ok"},{"url":"https://cloud.google.com/blog/topics/threat-intelligence/ai-vulnerability-exploitation-initial-access","linkCheck":"ok"}],"directions":{"zh":["漏洞挖掘智能体","自动修补","威胁情报","AI 驱动攻击的检测"],"en":["Vulnerability-hunting agents","Automated patching","Threat intelligence","Detecting AI-driven attacks"]},"stats":{"milestones":9,"milestonesSince2025":6,"milestones2026":5,"milestonesLast6Months":3,"news":12,"latestProgress":"2026-10-20","latestMilestone":"2026-10-16","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":0,"leaps":{"ai":2,"field":0},"latestPapers":8,"keyPapers":3},"milestones":[{"date":"2024-04-11","title":{"zh":"大模型智能体可自主利用一日漏洞，GPT-4 成功率 87%（arXiv 2404.08144）","en":"LLM agents can autonomously exploit one-day vulnerabilities (87% with GPT-4)"},"url":"https://arxiv.org/abs/2404.08144","source":"arXiv","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"首次系统表明大模型可把公开漏洞描述变成可用攻击","en":"First systematic evidence LLMs can weaponise CVE descriptions"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首次系统表明大模型可把公开漏洞描述变成可用攻击","en":"First systematic evidence LLMs can weaponise CVE descriptions"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首次系统表明大模型可把公开漏洞描述变成可用攻击","en":"First systematic evidence LLMs can weaponise CVE descriptions"}},"impactNotes":[],"verified":{"id":"arXiv:2404.08144","title":"LLM Agents can Autonomously Exploit One-day Vulnerabilities","venue":"arXiv","date":"2024-04-11"},"addedIn":"v2"},{"date":"2024-08-15","title":{"zh":"Cybench：专业级 CTF 网络安全能力评测基准","en":"Cybench: professional-level CTF benchmark for LM cyber capability"},"url":"https://arxiv.org/abs/2408.08926","source":"arXiv","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok-crossref","datePrecision":"day","type":"benchmark","typeLabel":{"zh":"基准","en":"Benchmark"},"why":{"zh":"成为政府 AI 安全评估使用的标准基准之一","en":"Adopted in government AI safety evaluations"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"成为政府 AI 安全评估使用的标准基准之一","en":"Adopted in government AI safety evaluations"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"成为政府 AI 安全评估使用的标准基准之一","en":"Adopted in government AI safety evaluations"}},"impactNotes":[],"verified":{"id":"arXiv:2408.08926","title":"Cybench: A Framework for Evaluating Cybersecurity Capabilities and Risks of Language Models","venue":"arXiv","date":"2024-08-15"},"addedIn":"v2"},{"date":"2024-11-01","title":{"zh":"Google Big Sleep 用大模型发现 SQLite 真实漏洞","en":"Google Big Sleep finds a real SQLite vulnerability with an LLM"},"url":"https://googleprojectzero.blogspot.com/2024/10/from-naptime-to-big-sleep.html","source":"Google Project Zero","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"首次公开报道 AI 在广泛使用的软件（SQLite）中发现此前未知的可利用漏洞","en":"First public AI-found exploitable bug in widely used software"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首次公开报道 AI 在广泛使用的软件（SQLite）中发现此前未知的可利用漏洞","en":"First public AI-found exploitable bug in widely used software"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首次公开报道 AI 在广泛使用的软件（SQLite）中发现此前未知的可利用漏洞","en":"First public AI-found exploitable bug in widely used software"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2025-08-08","title":{"zh":"DARPA AIxCC 决赛：Team Atlanta 夺冠，7 支队伍的自主系统发现约 80% 植入漏洞、修补 60% 以上并全部开源","en":"DARPA AIxCC final: Team Atlanta wins; autonomous systems find ~80% of planted bugs and patch 60%+, all open-sourced"},"url":"https://www.darpa.mil/news/2025/aixcc-results","source":"DARPA","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"competition","typeLabel":{"zh":"竞赛","en":"Competition"},"why":{"zh":"政府组织、独立评分的大型赛事证明 AI 可自主发现并修补真实开源软件漏洞（原“待核实”已确认 2025-08-08）","en":"Independently scored government challenge shows autonomous find-and-patch"},"aiImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"政府组织、独立评分的大型赛事证明 AI 可自主发现并修补真实开源软件漏洞（原“待核实”已确认 2025-08-08）","en":"Independently scored government challenge shows autonomous find-and-patch"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"政府组织、独立评分的大型赛事证明 AI 可自主发现并修补真实开源软件漏洞（原“待核实”已确认 2025-08-08）","en":"Independently scored government challenge shows autonomous find-and-patch"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-02-05","title":{"zh":"Anthropic：Claude Opus 4.6 验证 500 多个高危零日漏洞","en":"Anthropic: Claude Opus 4.6 validated 500+ high-severity zero-days"},"url":"https://www.anthropic.com/research/zero-days","source":"Anthropic","flags":["vendor"],"flagLabels":[{"zh":"厂商自述","en":"Vendor claim"}],"linkCheck":"ok","datePrecision":"day","type":"industry","typeLabel":{"zh":"产业/采用","en":"Industry"},"why":{"zh":"厂商称验证 500 多个高危零日","en":"Vendor: 500+ validated high-severity zero-days"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"厂商称验证 500 多个高危零日","en":"Vendor: 500+ validated high-severity zero-days"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"厂商称验证 500 多个高危零日","en":"Vendor: 500+ validated high-severity zero-days"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-04-20","title":{"zh":"TitanCA 多智能体发现 203 个零日、118 个 CVE","en":"TitanCA: 203 zero-days, 118 CVEs"},"url":"https://arxiv.org/abs/2604.17860","source":"arXiv","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"多智能体发现 203 个零日、118 个 CVE（预印本）","en":"203 zero-days, 118 CVEs (preprint)"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"多智能体发现 203 个零日、118 个 CVE（预印本）","en":"203 zero-days, 118 CVEs (preprint)"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"多智能体发现 203 个零日、118 个 CVE（预印本）","en":"203 zero-days, 118 CVEs (preprint)"}},"impactNotes":[],"verified":{"id":"arXiv:2604.17860","title":"TitanCA: Lessons from Orchestrating LLM Agents to Discover 100+ CVEs","venue":"arXiv","date":"2026-04-20"},"addedIn":"v1"},{"date":"2026-05-11","title":{"zh":"Google GTIG 首次发现疑似由 AI 开发的在野零日利用","en":"Google GTIG identifies first likely AI-developed zero-day exploit"},"url":"https://cloud.google.com/blog/topics/threat-intelligence/ai-vulnerability-exploitation-initial-access","source":"Google Cloud","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"industry","typeLabel":{"zh":"产业/采用","en":"Industry"},"why":{"zh":"首次在野发现疑似 AI 开发的零日利用，攻击方也被改造","en":"First in-the-wild zero-day exploit likely AI-developed"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首次在野发现疑似 AI 开发的零日利用，攻击方也被改造","en":"First in-the-wild zero-day exploit likely AI-developed"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首次在野发现疑似 AI 开发的零日利用，攻击方也被改造","en":"First in-the-wild zero-day exploit likely AI-developed"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-10-07","title":{"zh":"微软 FORGE：5–9 月参与发现 140 个 Windows CVE","en":"Microsoft FORGE: 140 Windows CVEs, May–Sep 2026"},"url":"https://www.microsoft.com/en-us/security/blog/2026/10/07/3-lessons-from-frontier-ai-vulnerability-research/","source":"Microsoft Security","flags":["vendor"],"flagLabels":[{"zh":"厂商自述","en":"Vendor claim"}],"linkCheck":"ok","datePrecision":"day","type":"industry","typeLabel":{"zh":"产业/采用","en":"Industry"},"why":{"zh":"厂商称参与发现 140 个 Windows CVE","en":"Vendor: 140 Windows CVEs"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"厂商称参与发现 140 个 Windows CVE","en":"Vendor: 140 Windows CVEs"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"厂商称参与发现 140 个 Windows CVE","en":"Vendor: 140 Windows CVEs"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-10-16","title":{"zh":"Anthropic 团队用 Claude 破解 NIST P-256 等椭圆曲线种子来源之谜：1997–98 年用 drand48 以时间戳生成，领走悬赏","en":"Claude traces the NIST elliptic-curve seeds to drand48 seeded with 1997–98 timestamps; bounty claimed"},"url":"https://www.anthropic.com/research/ecdsa-curve-origins","source":"Anthropic","flags":["vendor"],"flagLabels":[{"zh":"厂商自述","en":"Vendor claim"}],"linkCheck":"ok","datePrecision":"day","type":"discovery","typeLabel":{"zh":"科学发现","en":"Discovery"},"why":{"zh":"AI：困扰密码学界 25 年的种子来源由 AI 辅助团队解出，悬赏发起人 Filippo Valsorda 已独立重实现并核验；学科：排除了 Bernstein–Lange 设想的种子操纵后门空间（自由度仅约 30 比特），恢复对 NIST 曲线的信任，但不改变其安全性","en":"AI: a 25-year mystery solved by an AI-assisted team, independently re-implemented and verified by bounty organiser Filippo Valsorda; Field: rules out seed-manipulation backdoors (≈30 bits of freedom) without changing curve security"},"aiImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"AI：困扰密码学界 25 年的种子来源由 AI 辅助团队解出，悬赏发起人 Filippo Valsorda 已独立重实现并核验；学科：排除了 Bernstein–Lange 设想的种子操纵后门空间（自由度仅约 30 比特），恢复对 NIST 曲线的信任，但不改变其安全性","en":"AI: a 25-year mystery solved by an AI-assisted team, independently re-implemented and verified by bounty organiser Filippo Valsorda; Field: rules out seed-manipulation backdoors (≈30 bits of freedom) without changing curve security"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：困扰密码学界 25 年的种子来源由 AI 辅助团队解出，悬赏发起人 Filippo Valsorda 已独立重实现并核验；学科：排除了 Bernstein–Lange 设想的种子操纵后门空间（自由度仅约 30 比特），恢复对 NIST 曲线的信任，但不改变其安全性","en":"AI: a 25-year mystery solved by an AI-assisted team, independently re-implemented and verified by bounty organiser Filippo Valsorda; Field: rules out seed-manipulation backdoors (≈30 bits of freedom) without changing curve security"}},"impactNotes":[{"zh":"厂商发布，但悬赏发起人 Filippo Valsorda 已用 Go 独立重实现并核验全部种子，悬赏已兑付","en":"Vendor post, but bounty organiser Filippo Valsorda independently re-implemented and verified the seeds; bounty paid"}],"verified":null,"addedIn":"v2"}],"papersRepos":[{"title":"TitanCA: Lessons from Orchestrating LLM Agents to Discover 100+ CVEs (arXiv 2604.17860)","url":"https://arxiv.org/abs/2604.17860","date":"2026-04-20","kind":"paper","linkCheck":"ok"},{"title":"ZeroDayBench (arXiv 2603.02297)","url":"https://arxiv.org/abs/2603.02297","date":"2026-03","kind":"paper","linkCheck":"ok"},{"title":"LLM Agents can Autonomously Exploit One-day Vulnerabilities","url":"https://arxiv.org/abs/2404.08144","date":"2024-04-11","kind":"paper","linkCheck":"ok-crossref"},{"title":"Cybench: A Framework for Evaluating Cybersecurity Capabilities and Risks of Language Models","url":"https://arxiv.org/abs/2408.08926","date":"2024-08-15","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"DARPA AIxCC","url":"https://aicyberchallenge.com"},{"name":"Google Project Zero","url":"https://googleprojectzero.blogspot.com"},{"name":"Microsoft Security","url":"https://www.microsoft.com/en-us/security/blog/"},{"name":"XBOW","url":"https://xbow.com"},{"name":"Anthropic","url":"https://www.anthropic.com"}],"news":[{"id":"46122b369cdc","date":"2026-10-20","title":"AI Accelerates Cyber Threat Landscape, Poses New Risks for Insurers and Businesses","url":"https://www.insurancejournal.com/news/national/2026/10/20/889665.htm","source":"Insurance Journal","lang":"en","kind":"news","summaryZh":"Insurance Journal于2026-10-20发布的文章，涉及AI 金融、AI 安全，原题《AI Accelerates Cyber Threat Landscape, Poses New Risks for Insurers and Businesses》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"59252d156fbe","date":"2026-10-20","title":"Army Cyber Command's ECHOTRIBBLE would turn AI agents into self-spreading cyberweapons","url":"https://the-decoder.com/army-cyber-commands-echotribble-would-turn-ai-agents-into-self-spreading-cyberweapons/","source":"The Decoder","lang":"en","kind":"news","summaryZh":"The Decoder于2026-10-20发布的文章，涉及AI 安全、AI 商业融资，原题《Army Cyber Command's ECHOTRIBBLE would turn AI agents into self-spreading cyberweapons》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"0dfa738d1635","date":"2026-10-20","title":"Hackers Use Autonomous AI Agents to Breach Company Websites, Target Over 1,200 Sites","url":"https://thehackernews.com/2026/10/hackers-use-autonomous-ai-agents-to.html","source":"The Hacker News","lang":"en","kind":"news","summaryZh":"The Hacker News于2026-10-20发布的文章，涉及AI 安全，原题《Hackers Use Autonomous AI Agents to Breach Company Websites, Target Over 1,200 Sites》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"dfadf5c326ba","date":"2026-10-20","title":"京东暂停双 11 期间外部安全测试及漏洞收录，未经授权测试将按违规处理","url":"https://www.ithome.com/1/015/590.htm","source":"IT之家","lang":"zh","kind":"news","summaryZh":"IT之家于2026-10-20发布的文章，涉及AI 安全、AI 法律政策，原题《京东暂停双 11 期间外部安全测试及漏洞收录，未经授权测试将按违规处理》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"91169d021095","date":"2026-10-20","title":"Hackers Abuse AI Pen-Test Tool Cairn to Automate Massive Attack Spree","url":"https://www.darkreading.com/cyberattacks-data-breaches/hackers-abuse-ai-pen-test-tool-cairn","source":"Dark Reading","lang":"en","kind":"news","summaryZh":"Dark Reading于2026-10-20发布的文章，涉及AI 安全，原题《Hackers Abuse AI Pen-Test Tool Cairn to Automate Massive Attack Spree》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"75467ee74dca","date":"2026-10-20","title":"网络安全初创公司 Blackwall 完成 1 亿美元融资，估值达 22 亿美元","url":"https://www.ithome.com/1/015/369.htm","source":"IT之家","lang":"zh","kind":"news","summaryZh":"IT之家于2026-10-20发布的文章，涉及AI 安全、AI 商业融资、AI 金融，原题《网络安全初创公司 Blackwall 完成 1 亿美元融资，估值达 22 亿美元》，详见原文链接。","via":"collector-run-20261020","linkCheck":"from-feed"},{"id":"9ec44bd063b0","date":"2026-10-20","title":"Federated Learning for Malware Image Classification under Data Heterogeneity","url":"https://arxiv.org/abs/2610.25264","source":"arXiv cs.CR","lang":"en","kind":"paper","summaryZh":"arXiv cs.CR于2026-10-20发布的论文，涉及AI 安全、AI 科研与调研，原题《Federated Learning for Malware Image Classification under Data Heterogeneity》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"8a67864a69bb","date":"2026-10-20","title":"AdaFedXplain: Adaptive Federated Learning with Explainable AI for Ransomware Detection","url":"https://arxiv.org/abs/2610.25318","source":"arXiv cs.CR","lang":"en","kind":"paper","summaryZh":"arXiv cs.CR于2026-10-20发布的论文，涉及AI 安全、AI 商业融资，原题《AdaFedXplain: Adaptive Federated Learning with Explainable AI for Ransomware Detection》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"9e57ed1bc6e8","date":"2026-10-20","title":"Adaptive Cyber-Defense for Evolving Attacker Capabilities using Reinforcement Learning and Behavior Trees","url":"https://arxiv.org/abs/2610.25420","source":"arXiv cs.CR","lang":"en","kind":"paper","summaryZh":"arXiv cs.CR于2026-10-20发布的论文，涉及AI 安全，原题《Adaptive Cyber-Defense for Evolving Attacker Capabilities using Reinforcement Learning and》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"0618ca744b65","date":"2026-10-19","title":"A Chinese hacker used AI agents to break into more than 100 company websites","url":"https://www.tomshardware.com/tech-industry/artificial-intelligence/a-chinese-hacker-used-ai-agents-to-break-into-more-than-100-company-websites-deere-and-sharkninja-among-websites-hit-by-open-source-ai-powered-problem-solving-platform-demo","source":"Tom's Hardware AI","lang":"en","kind":"news","summaryZh":"Tom's Hardware AI于2026-10-19发布的文章，涉及AI 芯片硬件，原题《A Chinese hacker used AI agents to break into more than 100 company websites》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"0534f1906337","date":"2026-10-19","title":"Fake Web3 Accelerator Uses npm Packages to Spread Malware and Hijack AI Agent Traffic","url":"https://thehackernews.com/2026/10/fake-web3-accelerator-uses-npm-packages.html","source":"The Hacker News","lang":"en","kind":"news","summaryZh":"The Hacker News于2026-10-19发布的文章，涉及AI 安全、AI 芯片硬件、AI 软件开发，原题《Fake Web3 Accelerator Uses npm Packages to Spread Malware and Hijack AI Agent Traffic》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"03d3ba9d36cc","date":"2026-10-19","title":"⚡ Weekly Recap: 98 Zero-Days, Rogue AI Agents, $100M Crypto Theft, 700 Supermarkets Hit and More","url":"https://thehackernews.com/2026/10/weekly-recap-98-zero-days-rogue-ai.html","source":"The Hacker News","lang":"en","kind":"news","summaryZh":"The Hacker News于2026-10-19发布的文章，涉及AI 安全、AI 金融，原题《⚡ Weekly Recap: 98 Zero-Days, Rogue AI Agents, $100M Crypto Theft, 700 Supermarkets Hit an》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"}],"review":{"lastReviewed":"2026-10-21","cycleDays":60,"nextReview":"2026-12-20"},"urls":{"zh":"/zh-hans/disciplines/cybersecurity/","en":"/en/disciplines/cybersecurity/"},"dimensions":{"method":{"score":3,"label":{"zh":"研究方法","en":"Research method"},"evidence":"大模型漏洞挖掘、自动修补成为攻防研究主线","source":"https://www.darpa.mil/news/2025/aixcc-results","weight":0.3},"discovery":{"score":3,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"AIxCC 系统在比赛中发现约 80% 植入漏洞并修补 60% 以上；Big Sleep 发现真实漏洞；多家称规模化发现零日","source":"https://www.darpa.mil/news/2025/aixcc-results","weight":0.35},"adoption":{"score":3,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"微软、谷歌等将 AI 漏洞研究纳入生产流程；GTIG 观察到攻击方使用 AI","source":"https://cloud.google.com/blog/topics/threat-intelligence/ai-vulnerability-exploitation-initial-access","weight":0.25},"education":{"score":1,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到课程证据","source":null,"weight":0.1}},"compositeScore":2.8,"levelCapped":false,"levelChange":{"from":3,"to":3,"direction":"same","reason":null},"biggestBreakthrough":{"date":"2026-10-16","title":{"zh":"Anthropic 团队用 Claude 破解 NIST P-256 等椭圆曲线种子来源之谜：1997–98 年用 drand48 以时间戳生成，领走悬赏","en":"Claude traces the NIST elliptic-curve seeds to drand48 seeded with 1997–98 timestamps; bounty claimed"},"url":"https://www.anthropic.com/research/ecdsa-curve-origins","aiImpact":"leap","fieldImpact":"major","why":{"zh":"AI：困扰密码学界 25 年的种子来源由 AI 辅助团队解出，悬赏发起人 Filippo Valsorda 已独立重实现并核验；学科：排除了 Bernstein–Lange 设想的种子操纵后门空间（自由度仅约 30 比特），恢复对 NIST 曲线的信任，但不改变其安全性","en":"AI: a 25-year mystery solved by an AI-assisted team, independently re-implemented and verified by bounty organiser Filippo Valsorda; Field: rules out seed-manipulation backdoors (≈30 bits of freedom) without changing curve security"}},"keyPapers":[{"title":"LLM Agents can Autonomously Exploit One-day Vulnerabilities","titleZh":"大模型智能体可自主利用一日漏洞，GPT-4 成功率 87%（arXiv 2404.08144）","id":"arXiv:2404.08144","url":"https://arxiv.org/abs/2404.08144","venue":"arXiv","date":"2024-04-11","why":{"zh":"首次系统表明大模型可把公开漏洞描述变成可用攻击","en":"First systematic evidence LLMs can weaponise CVE descriptions"},"fieldImpact":"major","aiImpact":"major"},{"title":"TitanCA: Lessons from Orchestrating LLM Agents to Discover 100+ CVEs","titleZh":"TitanCA 多智能体发现 203 个零日、118 个 CVE","id":"arXiv:2604.17860","url":"https://arxiv.org/abs/2604.17860","venue":"arXiv","date":"2026-04-20","why":{"zh":"多智能体发现 203 个零日、118 个 CVE（预印本）","en":"203 zero-days, 118 CVEs (preprint)"},"fieldImpact":"major","aiImpact":"major"},{"title":"Cybench: A Framework for Evaluating Cybersecurity Capabilities and Risks of Language Models","titleZh":"Cybench：专业级 CTF 网络安全能力评测基准","id":"arXiv:2408.08926","url":"https://arxiv.org/abs/2408.08926","venue":"arXiv","date":"2024-08-15","why":{"zh":"成为政府 AI 安全评估使用的标准基准之一","en":"Adopted in government AI safety evaluations"},"fieldImpact":"incremental","aiImpact":"incremental"}],"latest_papers":[{"id":"arXiv:2610.29869","title":"Compaction Laundering: Provenance Destruction in Context-Compacting Coding Agents","titleZh":"压缩洗白：会压缩上下文的编程智能体中的来源信息破坏","summaryZh":"发现主流开源编程智能体在压缩上下文时把对话扁平化成纯文本，抹掉消息来源，攻击者可借此把注入内容伪装成用户指令。","summaryEn":"Shows context compaction in open-source coding agents destroys message provenance, letting injected text masquerade as user instructions.","authors":["Arshan Dabirsiaghi"],"date":"2026-10-21","url":"https://arxiv.org/abs/2610.29869","site":"arXiv","venue":null,"category":"cs.CR","fetchedAt":"2026-10-22T10:08:56"},{"id":"arXiv:2610.28810","title":"Combining Long-Term Memory and LLMs Comes at a High Cost: From Prompt Injection to Persistent Compromise","titleZh":"长期记忆加大模型代价高昂：从提示注入到持久性入侵","summaryZh":"指出带持久记忆的对话系统会让一次提示注入跨会话长期生效，分析其攻击面与风险。","summaryEn":"Shows persistent memory turns one-off prompt injection into lasting, cross-session compromise of LLM assistants.","authors":["Pedro H. Barcha Correia","Matheus Azevedo de Sá","Paulo H. Gomes de Senna","Giovanni Mandel Martignago"],"date":"2026-10-20","url":"https://arxiv.org/abs/2610.28810","site":"arXiv","venue":null,"category":"cs.CR","fetchedAt":"2026-10-22T10:08:56"},{"id":"arXiv:2610.27653","title":"A Unified Attack Framework against Statistical Generative Watermarks for Large Language Models","titleZh":"针对大模型统计生成式水印的统一攻击框架","summaryZh":"发现 KGW、SynthID 等多类水印的检测分数与生成时的概率偏移存在固有关联，攻击者只需黑盒查询即可生成通顺且逃避检测的文本，并证明该攻击在这些水印族上最优。","summaryEn":"A black-box attack exploiting a shared weakness across LLM watermarks (incl. SynthID), proven optimal across covered families.","authors":["Romina Omidi","Nima Naderloui","Yuan Hong","Binghui Wang"],"date":"2026-10-20","url":"https://arxiv.org/abs/2610.27653","site":"arXiv","venue":null,"category":"cs.CR","fetchedAt":"2026-10-22T10:08:56"},{"id":"arXiv:2610.27574","title":"The Show-Up Attack: When Subagents Win Over Their Supervisor","titleZh":"“越位攻击”：子智能体压过主管","summaryZh":"提出恶意子智能体诱导主管让其参与超出授权任务的攻击，构建含 188 个用例的 SUA-Arena，6 个前沿模型全部中招，作为主管的 GPT-5.6-Sol 攻击成功率最高达 0.93。","summaryEn":"Malicious subagents manipulate supervisors beyond scope; all six frontier LLMs vulnerable, ASR up to 0.93.","authors":["Maosen Zhang","Jianshuo Dong","Yutong Zhang","Jie Zhang"],"date":"2026-10-20","url":"https://arxiv.org/abs/2610.27574","site":"arXiv","venue":null,"category":"cs.AI","fetchedAt":"2026-10-22T10:08:56"},{"id":"arXiv:2610.27353","title":"Exposing Vulnerability Monoculture Through Systemic Adversarial Robustness","titleZh":"从系统性对抗鲁棒性看“漏洞单一化”","summaryZh":"提出系统性对抗鲁棒性（SAR）指标，衡量攻破整个模型池需要多少种不同攻击，揭示基础模型共享训练流程与数据带来的集中风险。","summaryEn":"Introduces Systemic Adversarial Robustness to measure how few attacks compromise a whole pool of foundation models.","authors":["Leo Elmecker-Plakolm","Lukasz Szpruch","Avishek Joey Bose","Matthew Wicker"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.27353","site":"arXiv","venue":null,"category":"cs.CR","fetchedAt":"2026-10-22T10:08:56"},{"id":"arXiv:2610.25791","title":"Towards Architectural-Level Safety Control: Relocating Hazardous Knowledge into Removable Experts in Mixture-of-Experts Models","titleZh":"迈向架构层面的安全控制：把危险知识迁移到混合专家模型中可移除的专家","summaryZh":"把危险知识集中到可删除的专家模块中，而不是只抑制表达，以免被后续微调恢复。","summaryEn":"Relocates hazardous knowledge into removable MoE experts so it can be truly deleted rather than suppressed.","authors":["Cheng Jin","Zhongtian Ma","Xiao Wang","Han Qi"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.25791","site":"arXiv","venue":null,"category":"cs.CR","fetchedAt":"2026-10-22T10:08:56"},{"id":"arXiv:2610.24896","title":"Peer Pressure: Execution-Verified Co-Evolution of Attacker and Defender Populations for Code Security","titleZh":"同伴压力：以执行验证让攻击者与防御者群体协同演化提升代码安全","summaryZh":"攻击方需在真实仓库中复现真实漏洞利用才得分，与防御方一起训练，全部奖励由执行结果决定。","summaryEn":"Co-trains attacker and defender LLM populations with execution-verified rewards on real repositories.","authors":["Khushbu Pahwa","Yubin Ge","Daniele Bonadiman","Xiaofei Ma"],"date":"2026-10-17","url":"https://arxiv.org/abs/2610.24896","site":"arXiv","venue":null,"category":"cs.CR","fetchedAt":"2026-10-22T10:08:56"},{"id":"arXiv:2610.24623","title":"Narratives Mislead: An Empirical Study of How LLMs Misjudge CVSS Severity in Bug Bounty Reports","titleZh":"叙述会误导：大模型如何误判漏洞赏金报告的 CVSS 严重度","summaryZh":"基于 4281 份报告，发现大模型的 CVSS 评分错误集中在研究者叙述与 CVSS 正式定义不一致的地方。","summaryEn":"Across 4,281 bug-bounty reports, LLM CVSS errors concentrate where narrative impact diverges from formal metric meaning.","authors":["Ruitao Feng","Yuekang Li","Yao Zhang"],"date":"2026-10-17","url":"https://arxiv.org/abs/2610.24623","site":"arXiv","venue":null,"category":"cs.CR","fetchedAt":"2026-10-22T10:08:56"}]},{"id":"chip-design","slug":"chip-design","category":"eng-computing","categoryName":{"zh":"工程与计算","en":"Engineering & computing"},"name":{"zh":"电子与芯片设计","en":"Electronics & chip design"},"classification":{"oecdFos":"2.2","moe":"1401 集成电路科学与工程"},"level":2,"levelKey":"breakthrough","levelLabel":{"zh":"局部突破","en":"Local breakthroughs"},"trend":"up","reason":{"zh":"强化学习芯片布局（AlphaChip）用于多代 TPU（Nature 2021），但可复现性遭质疑（CACM 复评）；2026 年 Synopsys 推进智能体 EDA，AMD 评估称调试周期缩短 25–40%（厂商口径）。","en":"RL chip placement (AlphaChip) was used for several TPU generations (Nature 2021) but its reproducibility is disputed (CACM re-evaluation); in 2026 Synopsys pushes agentic EDA, with AMD reporting 25–40% shorter debug cycles (vendor figures)."},"reasonSources":[{"url":"https://www.nature.com/articles/s41586-021-03544-w","linkCheck":"ok"},{"url":"https://doi.org/10.1145/3676845","linkCheck":"ok-crossref"},{"url":"https://news.synopsys.com/2026-07-27-Synopsys-Advances-Agentic-AI-Chip-Design-with-AMD-and-Microsoft","linkCheck":"ok"}],"directions":{"zh":["布局布线","验证与调试智能体","RTL 生成"],"en":["Placement & routing","Verification/debug agents","RTL generation"]},"stats":{"milestones":8,"milestonesSince2025":4,"milestones2026":4,"milestonesLast6Months":4,"news":7,"latestProgress":"2026-10-19","latestMilestone":"2026-10-14","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":0,"leaps":{"ai":0,"field":0},"latestPapers":8,"keyPapers":4},"milestones":[{"date":"2021-06-09","title":{"zh":"强化学习芯片布局（Nature）","en":"RL chip floorplanning (Nature)"},"url":"https://www.nature.com/articles/s41586-021-03544-w","source":"Nature","flags":["contested"],"flagLabels":[{"zh":"有争议","en":"Contested"}],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"强化学习数小时生成芯片布局，用于多代 TPU；可复现性遭质疑","en":"RL floorplanning used in TPUs; reproducibility disputed"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"强化学习数小时生成芯片布局，用于多代 TPU；可复现性遭质疑","en":"RL floorplanning used in TPUs; reproducibility disputed"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"强化学习数小时生成芯片布局，用于多代 TPU；可复现性遭质疑","en":"RL floorplanning used in TPUs; reproducibility disputed"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-021-03544-w","title":"A graph placement methodology for fast chip design","venue":"Nature","date":"2021-06-09"},"addedIn":"v1"},{"date":"2023-10-31","title":{"zh":"ChipNeMo：领域适配大模型用于芯片设计（NVIDIA，arXiv 2311.00176）","en":"ChipNeMo: domain-adapted LLMs for chip design (NVIDIA)"},"url":"https://arxiv.org/abs/2311.00176","source":"arXiv","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"芯片企业内部首次系统报告大模型用于设计助手、EDA 脚本与缺陷分析","en":"First systematic in-house LLM use for chip design"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"芯片企业内部首次系统报告大模型用于设计助手、EDA 脚本与缺陷分析","en":"First systematic in-house LLM use for chip design"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"芯片企业内部首次系统报告大模型用于设计助手、EDA 脚本与缺陷分析","en":"First systematic in-house LLM use for chip design"}},"impactNotes":[],"verified":{"id":"arXiv:2311.00176","title":"ChipNeMo: Domain-Adapted LLMs for Chip Design","venue":"arXiv","date":"2023-10-31"},"addedIn":"v2"},{"date":"2024-09-26","title":{"zh":"Google：AlphaChip 如何改变芯片设计","en":"Google: how AlphaChip transformed chip design"},"url":"https://deepmind.google/discover/blog/how-alphachip-transformed-computer-chip-design/","source":"Google DeepMind","flags":["vendor"],"flagLabels":[{"zh":"厂商自述","en":"Vendor claim"}],"linkCheck":"ok","datePrecision":"day","type":"industry","typeLabel":{"zh":"产业/采用","en":"Industry"},"why":{"zh":"Google 称 AlphaChip 用于多代 TPU（厂商自述）","en":"Vendor claim"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"Google 称 AlphaChip 用于多代 TPU（厂商自述）","en":"Vendor claim"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"Google 称 AlphaChip 用于多代 TPU（厂商自述）","en":"Vendor claim"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2024-10-25","title":{"zh":"复评 Google 强化学习宏单元布局（CACM）","en":"Reevaluating Google's RL for IC macro placement (CACM)"},"url":"https://doi.org/10.1145/3676845","source":"Communications of the ACM","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"独立复评质疑原论文方法与结论，是重要反证","en":"Independent re-evaluation challenging the original"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"独立复评质疑原论文方法与结论，是重要反证","en":"Independent re-evaluation challenging the original"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"独立复评质疑原论文方法与结论，是重要反证","en":"Independent re-evaluation challenging the original"}},"impactNotes":[],"verified":{"id":"10.1145/3676845","title":"Reevaluating Google’s Reinforcement Learning for IC Macro Placement","venue":"Communications of the ACM","date":"2024-10-25"},"addedIn":"v1"},{"date":"2026-04-23","title":{"zh":"Verkor Design Conductor：智能体从 219 词规格 12 小时生成可布局的 RISC-V CPU（仅仿真，未流片）","en":"Verkor Design Conductor builds a RISC-V CPU from a 219-word spec to GDSII in 12 hours (simulation only)"},"url":"https://www.tomshardware.com/tech-industry/artificial-intelligence/ai-agent-designs-a-complete-risc-v-cpu-from-a-219-word-spec-in-just-12-hours","source":"Tom’s Hardware","flags":["vendor"],"flagLabels":[{"zh":"厂商自述","en":"Vendor claim"}],"linkCheck":"ok","datePrecision":"day","type":"system","typeLabel":{"zh":"系统发布","en":"System"},"why":{"zh":"首次由自主智能体完成从规格到 GDSII 的完整 CPU；但未制造实物，作者估计仍需 5–10 名专家指导","en":"First spec-to-GDSII CPU by an agent; no silicon"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首次由自主智能体完成从规格到 GDSII 的完整 CPU；但未制造实物，作者估计仍需 5–10 名专家指导","en":"First spec-to-GDSII CPU by an agent; no silicon"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首次由自主智能体完成从规格到 GDSII 的完整 CPU；但未制造实物，作者估计仍需 5–10 名专家指导","en":"First spec-to-GDSII CPU by an agent; no silicon"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-07-27","title":{"zh":"Synopsys 与 AMD、微软推进智能体芯片设计","en":"Synopsys advances agentic AI chip design with AMD & Microsoft"},"url":"https://news.synopsys.com/2026-07-27-Synopsys-Advances-Agentic-AI-Chip-Design-with-AMD-and-Microsoft","source":"Synopsys","flags":["vendor"],"flagLabels":[{"zh":"厂商自述","en":"Vendor claim"}],"linkCheck":"ok","datePrecision":"day","type":"industry","typeLabel":{"zh":"产业/采用","en":"Industry"},"why":{"zh":"EDA 厂商推进智能体设计流程（厂商口径）","en":"Vendor agentic EDA push"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"EDA 厂商推进智能体设计流程（厂商口径）","en":"Vendor agentic EDA push"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"EDA 厂商推进智能体设计流程（厂商口径）","en":"Vendor agentic EDA push"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-08-26","title":{"zh":"Architect Labs Redwood：AI 两周内从零设计、验证并在 FPGA 上部署 AI 加速器（arXiv 2608.26418）","en":"Architect Labs Redwood: AI accelerator designed, verified and FPGA-deployed by AI in two weeks"},"url":"https://arxiv.org/abs/2608.26418","source":"arXiv","flags":["vendor","preprint"],"flagLabels":[{"zh":"厂商自述","en":"Vendor claim"},{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok-crossref","datePrecision":"day","type":"system","typeLabel":{"zh":"系统发布","en":"System"},"why":{"zh":"RTL、验证、固件全部由 AI 生成，运行十亿参数模型推理；尚未流片","en":"All RTL/verification/firmware AI-generated; not yet taped out"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"RTL、验证、固件全部由 AI 生成，运行十亿参数模型推理；尚未流片","en":"All RTL/verification/firmware AI-generated; not yet taped out"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"RTL、验证、固件全部由 AI 生成，运行十亿参数模型推理；尚未流片","en":"All RTL/verification/firmware AI-generated; not yet taped out"}},"impactNotes":[],"verified":{"id":"arXiv:2608.26418","title":"Redwood: A Frontier AI Accelerator Designed, Verified, and Deployed from Scratch in 2 Weeks by AI","venue":"arXiv","date":"2026-08-26"},"addedIn":"v2"},{"date":"2026-10-14","title":{"zh":"ChipAgents 与 NVIDIA 推出 Build-A-Chip-Bench：从规格到 GDS 的端到端芯片设计智能体基准","en":"ChipAgents & NVIDIA launch Build-A-Chip-Bench for end-to-end autonomous chip design"},"url":"https://chipagents.ai/newsroom/chipagents-nvidia-build-a-chip-bench","source":"ChipAgents","flags":["vendor"],"flagLabels":[{"zh":"厂商自述","en":"Vendor claim"}],"linkCheck":"ok","datePrecision":"day","type":"benchmark","typeLabel":{"zh":"基准","en":"Benchmark"},"why":{"zh":"AI：覆盖 RTL、验证、综合、物理实现与签核；学科：初步结果显示智能体功能验证通过率约 40%，满足 PPA 目标仅约 10%","en":"AI: covers RTL through signoff; Field: early agents pass functional verification ~40% but meet PPA targets only ~10%"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：覆盖 RTL、验证、综合、物理实现与签核；学科：初步结果显示智能体功能验证通过率约 40%，满足 PPA 目标仅约 10%","en":"AI: covers RTL through signoff; Field: early agents pass functional verification ~40% but meet PPA targets only ~10%"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：覆盖 RTL、验证、综合、物理实现与签核；学科：初步结果显示智能体功能验证通过率约 40%，满足 PPA 目标仅约 10%","en":"AI: covers RTL through signoff; Field: early agents pass functional verification ~40% but meet PPA targets only ~10%"}},"impactNotes":[],"verified":null,"addedIn":"v2"}],"papersRepos":[{"title":"google-research/circuit_training","url":"https://github.com/google-research/circuit_training","date":null,"kind":"repo","linkCheck":"ok"},{"title":"A graph placement methodology for fast chip design","url":"https://doi.org/10.1038/s41586-021-03544-w","date":"2021-06-09","kind":"paper","linkCheck":"ok-crossref"},{"title":"ChipNeMo: Domain-Adapted LLMs for Chip Design","url":"https://arxiv.org/abs/2311.00176","date":"2023-10-31","kind":"paper","linkCheck":"ok-crossref"},{"title":"Reevaluating Google’s Reinforcement Learning for IC Macro Placement","url":"https://doi.org/10.1145/3676845","date":"2024-10-25","kind":"paper","linkCheck":"ok-crossref"},{"title":"Redwood: A Frontier AI Accelerator Designed, Verified, and Deployed from Scratch in 2 Weeks by AI","url":"https://arxiv.org/abs/2608.26418","date":"2026-08-26","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"Google DeepMind","url":"https://deepmind.google"},{"name":"Synopsys","url":"https://www.synopsys.com"},{"name":"Cadence","url":"https://www.cadence.com"},{"name":"NVIDIA Research","url":"https://research.nvidia.com"}],"news":[{"id":"26159c4a5e5e","date":"2026-10-19","title":"ChipAgents Expands AI Across Chip Design. The Test Is Faster, Verified Results","url":"https://semiwiki.com/eda/chipagents-ai/375026-chipagents-expands-ai-across-chip-design-the-test-is-faster-verified-results/","source":"SemiWiki","lang":"en","kind":"news","summaryZh":"SemiWiki于2026-10-20发布的文章，涉及AI 芯片硬件，原题《ChipAgents Expands AI Across Chip Design. The Test Is Faster, Verified Results》，详见原文链接。","via":"collector-run-20261020","linkCheck":"from-feed"},{"id":"96560e491ed3","date":"2026-10-18","title":"A Composable Thermal Solver For Heterogeneous 2.5D/3D-IC Packages (Cadence, Notre Dame)","url":"https://semiengineering.com/a-composable-thermal-solver-for-heterogeneous-2-5d-3d-ic-packages-cadence-notre-dame/","source":"Semiconductor Engineering","lang":"en","kind":"news","summaryZh":"Semiconductor Engineering于2026-10-19发布的文章，涉及AI 芯片硬件、AI 教育，原题《A Composable Thermal Solver For Heterogeneous 2.5D/3D-IC Packages (Cadence, Notre Dame)》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"5ff5c30e38d3","date":"2026-10-16","title":"Physics AI pushes chip design beyond traditional simulation","url":"https://siliconangle.com/2026/10/16/vinci4d-physics-models-semiconwest/","source":"SiliconANGLE AI","lang":"en","kind":"news","summaryZh":"SiliconANGLE AI于2026-10-17发布的文章，涉及AI 芯片硬件、AI 商业融资，原题《Physics AI pushes chip design beyond traditional simulation》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"f5e4dbcd6767","date":"2026-10-15","title":"Vinci targets faster chip design and simulation with physics-driven AI","url":"https://siliconangle.com/2026/10/15/simulation-vinci-physics-driven-ai-thecube-semiconwest/","source":"SiliconANGLE AI","lang":"en","kind":"news","summaryZh":"SiliconANGLE AI于2026-10-16发布的文章，涉及AI 芯片硬件、AI 商业融资，原题《Vinci targets faster chip design and simulation with physics-driven AI》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"ef03e1cd1e54","date":"2026-10-14","title":"NVIDIA and Semiconductor Leaders Advance AI Across Chip Design and Manufacturing for AI-Defined Fabs","url":"https://nvidianews.nvidia.com/news/nvidia-and-semiconductor-leaders-advance-ai-across-chip-design-and-manufacturing-for-ai-defined-labs","source":"NVIDIA Newsroom","lang":"en","kind":"news","summaryZh":"NVIDIA Newsroom于2026-10-15发布的文章，涉及AI 芯片硬件、AI 软件开发，原题《NVIDIA and Semiconductor Leaders Advance AI Across Chip Design and Manufacturing for AI-De》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"587d8045aa32","date":"2026-10-06","title":"OpenAI and Synopsys partner to build \"GPT-Synopsys\" for autonomous chip design","url":"https://www.tomshardware.com/tech-industry/artificial-intelligence/openai-and-synopsys-partner-to-build-gpt-synopsys-for-autonomous-chip-design-specialized-ai-model-will-operate-eda-tools-allowing-engineers-to-deliver-more-sophisticated-chips-faster","source":"Tom's Hardware AI","lang":"en","kind":"news","summaryZh":"OpenAI 与 Synopsys 合作开发能直接操作 EDA 工具的芯片设计专用模型 GPT-Synopsys。","via":"research-samples-20261010","linkCheck":"from-feed"},{"id":"e83559f1afc2","date":"2026-04-22","title":"AI chip design","url":"https://spectrum.ieee.org/ai-chip-design","source":"IEEE Spectrum","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"ok"}],"review":{"lastReviewed":"2026-10-21","cycleDays":90,"nextReview":"2027-01-19"},"urls":{"zh":"/zh-hans/disciplines/chip-design/","en":"/en/disciplines/chip-design/"},"dimensions":{"method":{"score":2,"label":{"zh":"研究方法","en":"Research method"},"evidence":"强化学习布局、AI 优化 EDA 已在工业使用，大模型生成 RTL 处于研究阶段","source":"https://arxiv.org/abs/2311.00176","weight":0.3},"discovery":{"score":2,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"2026 年出现 AI 智能体从规格到 GDSII 的完整 CPU 设计和 FPGA 上运行的 AI 加速器，但均未流片","source":"https://spectrum.ieee.org/ai-chip-design","weight":0.35},"adoption":{"score":2,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"EDA 厂商 AI 工具被采用，数字为厂商口径","source":null,"weight":0.25},"education":{"score":1,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到课程证据","source":null,"weight":0.1}},"compositeScore":1.9,"levelCapped":false,"levelChange":{"from":2,"to":2,"direction":"same","reason":null},"biggestBreakthrough":{"date":"2026-08-26","title":{"zh":"Architect Labs Redwood：AI 两周内从零设计、验证并在 FPGA 上部署 AI 加速器（arXiv 2608.26418）","en":"Architect Labs Redwood: AI accelerator designed, verified and FPGA-deployed by AI in two weeks"},"url":"https://arxiv.org/abs/2608.26418","aiImpact":"major","fieldImpact":"major","why":{"zh":"RTL、验证、固件全部由 AI 生成，运行十亿参数模型推理；尚未流片","en":"All RTL/verification/firmware AI-generated; not yet taped out"}},"keyPapers":[{"title":"A graph placement methodology for fast chip design","titleZh":"强化学习芯片布局（Nature）","id":"10.1038/s41586-021-03544-w","url":"https://doi.org/10.1038/s41586-021-03544-w","venue":"Nature","date":"2021-06-09","why":{"zh":"强化学习数小时生成芯片布局，用于多代 TPU；可复现性遭质疑","en":"RL floorplanning used in TPUs; reproducibility disputed"},"fieldImpact":"major","aiImpact":"major"},{"title":"ChipNeMo: Domain-Adapted LLMs for Chip Design","titleZh":"ChipNeMo：领域适配大模型用于芯片设计（NVIDIA，arXiv 2311.00176）","id":"arXiv:2311.00176","url":"https://arxiv.org/abs/2311.00176","venue":"arXiv","date":"2023-10-31","why":{"zh":"芯片企业内部首次系统报告大模型用于设计助手、EDA 脚本与缺陷分析","en":"First systematic in-house LLM use for chip design"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Reevaluating Google’s Reinforcement Learning for IC Macro Placement","titleZh":"复评 Google 强化学习宏单元布局（CACM）","id":"10.1145/3676845","url":"https://doi.org/10.1145/3676845","venue":"Communications of the ACM","date":"2024-10-25","why":{"zh":"独立复评质疑原论文方法与结论，是重要反证","en":"Independent re-evaluation challenging the original"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Redwood: A Frontier AI Accelerator Designed, Verified, and Deployed from Scratch in 2 Weeks by AI","titleZh":"Architect Labs Redwood：AI 两周内从零设计、验证并在 FPGA 上部署 AI 加速器（arXiv 2608.26418）","id":"arXiv:2608.26418","url":"https://arxiv.org/abs/2608.26418","venue":"arXiv","date":"2026-08-26","why":{"zh":"RTL、验证、固件全部由 AI 生成，运行十亿参数模型推理；尚未流片","en":"All RTL/verification/firmware AI-generated; not yet taped out"},"fieldImpact":"major","aiImpact":"major"}],"latest_papers":[{"id":"arXiv:2610.29873","title":"HAWK: Heterogeneity-Aware KV Cache Transfer for Efficient GPU-NPU Phase-Disaggregated LLM Inference","titleZh":"HAWK：面向 GPU–NPU 预填充/解码分离推理的异构感知 KV Cache 传输","summaryZh":"预填充放在 GPU、解码放在 NPU，并优化两者间的 KV Cache 传输，提高异构硬件上的大模型推理效率。","summaryEn":"Runs prefill on GPUs and decode on NPUs with heterogeneity-aware KV-cache transfer for efficient disaggregated inference.","authors":["Junhyeok Park","Jungeun Yoon","Kyungmin Kim","Hunsoon Lee"],"date":"2026-10-21","url":"https://arxiv.org/abs/2610.29873","site":"arXiv","venue":null,"category":"cs.AR","fetchedAt":"2026-10-22T10:09:00"},{"id":"arXiv:2610.29503","title":"Hard and Soft Real-Time DNN Inference Scheduling on Heterogeneous Multi-Accelerator Systems via Reinforcement Learning","titleZh":"用强化学习在异构多加速器系统上调度硬/软实时 DNN 推理","summaryZh":"在线强化学习调度器生成候选调度，配合确定性安全层保证硬实时任务截止期，同时优化软实时任务服务质量。","summaryEn":"An online RL scheduler with a deterministic safety layer schedules hard and soft real-time DNN inference on heterogeneous accelerators.","authors":["Francesco Giulio Blanco","Enrico Russo","Giuseppe Ascia","Davide Patti"],"date":"2026-10-21","url":"https://arxiv.org/abs/2610.29503","site":"arXiv","venue":null,"category":"cs.AR","fetchedAt":"2026-10-22T10:09:00"},{"id":"arXiv:2610.26432","title":"E-Layout: Layout-Stage Logic Restructuring via RL-Optimized Equality-Saturated Graph Retrieval-Augmented Generation","titleZh":"E-Layout：用强化学习优化的等式饱和图检索增强生成做版图阶段逻辑重构","summaryZh":"在版图阶段而非综合阶段重构逻辑结构，突破仅靠插缓冲和调尺寸的优化空间限制。","summaryEn":"Layout-stage logic restructuring via RL-optimised equality-saturation graph RAG.","authors":["Wenji Fang","Rongjian Liang","Yi-Chen Lu","Cunxi Yu"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.26432","site":"arXiv","venue":null,"category":"cs.AR","fetchedAt":"2026-10-22T10:09:00"},{"id":"arXiv:2610.26131","title":"A Hybrid GPU and Processing-Near-Memory Architecture with Shared DRAM for Efficient Prefill-Decode Disaggregation","titleZh":"共享 DRAM 的 GPU 与近存处理混合架构，用于高效的预填充-解码分离","summaryZh":"提出 GPU 与近存计算共享 DRAM 的架构，缓解大模型推理预填充算力型与解码访存型的不匹配。","summaryEn":"A hybrid GPU + processing-near-memory architecture with shared DRAM for prefill–decode disaggregation.","authors":["Sanghyeok Han","Inseong Choi","Byungkuk Yoon","Gyeonghwan Park"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.26131","site":"arXiv","venue":null,"category":"cs.AR","fetchedAt":"2026-10-22T10:09:00"},{"id":"arXiv:2610.26118","title":"Superposed analog memory for vector storage in neural networks","titleZh":"用于神经网络向量存储的叠加模拟存储器","summaryZh":"演示在单个模拟存储阵列中以分布式叠加方式存储多个向量，通过存内矩阵向量乘法读出，提高存储密度。","summaryEn":"Analog memory storing multiple vectors in superposition in one array, recovered by in-memory MVM.","authors":["Athanasios Vasilopoulos","Patrick Diener","Michael Hersche","Abbas Rahimi"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.26118","site":"arXiv","venue":null,"category":"cs.ET","fetchedAt":"2026-10-22T10:09:00"},{"id":"arXiv:2610.26083","title":"MX-Like Tensor Core for LLM Inference with NPU Integration via Hardware-Software Co-Design Exploration","titleZh":"面向大模型推理的类 MX 张量核：通过软硬件协同设计探索集成 NPU","summaryZh":"系统探索块缩放量化格式的设计空间，联合评估精度与芯片面积，寻找边缘 NPU 部署大模型的帕累托最优配置。","summaryEn":"Joint HW–SW exploration of MX-like block-scaled formats for edge NPU LLM inference.","authors":["Zijun Zhao","Ruichu Ao","Stef Cuyckens","Marian Verhelst"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.26083","site":"arXiv","venue":null,"category":"cs.AR","fetchedAt":"2026-10-22T10:09:00"},{"id":"arXiv:2610.26991","title":"3D-MoE: Complementarity-Guided Hardware-Software Co-Design for MoE Decode on 3D-Stacked NMP","titleZh":"3D-MoE：面向三维堆叠近存计算的 MoE 解码软硬件协同设计","summaryZh":"利用不同领域 token 路由造成的专家负载此消彼长（“专家互补性”），离线分组并做局部性感知映射，在 3D 堆叠 DRAM 近存处理上平衡 MoE 解码负载。","summaryEn":"Hardware–software co-design exploiting expert complementarity to balance MoE decode on 3D-stacked near-memory processing.","authors":["Yudong Pan","Lian Liu","Shixin Zhao","Yintao He"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.26991","site":"arXiv","venue":null,"category":"cs.AR","fetchedAt":"2026-10-22T10:09:00"},{"id":"arXiv:2610.25881","title":"Learn the Interface, Not the Implementation: Output Representation Dominates Training Regime in LLM-Based RTL Generation","titleZh":"学接口而非实现：大模型生成 RTL 时输出表示比训练方式更重要","summaryZh":"在 ECC 存储包装器生成任务上固定数据与验证器，发现输出表示层级比七种训练方式对结果影响更大。","summaryEn":"For LLM RTL generation of SECDED wrappers, output representation matters more than the training regime.","authors":["Siddharth Patel","Vikram Singh","Siddharth Savyasachi Malu","Rohit Singh"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.25881","site":"arXiv","venue":null,"category":"cs.LG","fetchedAt":"2026-10-22T10:09:00"}]},{"id":"robotics","slug":"robotics","category":"eng-computing","categoryName":{"zh":"工程与计算","en":"Engineering & computing"},"name":{"zh":"机器人与自动化","en":"Robotics & automation"},"classification":{"oecdFos":"2.2","moe":"0811 控制科学与工程"},"level":2,"levelKey":"breakthrough","levelLabel":{"zh":"局部突破","en":"Local breakthroughs"},"trend":"up","reason":{"zh":"视觉-语言-动作（VLA）大模型让机器人具备一定泛化能力（RT-2 2023、π0 2024、Gemini Robotics 2025），但真实环境可靠性和规模部署仍未解决。","en":"Vision-language-action models give robots some generalisation (RT-2 2023, π0 2024, Gemini Robotics 2025), but real-world reliability and deployment at scale remain unsolved."},"reasonSources":[{"url":"https://deepmind.google/discover/blog/rt-2-new-model-translates-vision-and-language-into-action/","linkCheck":"ok"},{"url":"https://deepmind.google/discover/blog/gemini-robotics-brings-ai-into-the-physical-world/","linkCheck":"ok"}],"directions":{"zh":["VLA 基础模型","人形机器人","仿真到现实迁移","具身数据集"],"en":["VLA foundation models","Humanoids","Sim-to-real","Embodied datasets"]},"stats":{"milestones":8,"milestonesSince2025":5,"milestones2026":4,"milestonesLast6Months":3,"news":12,"latestProgress":"2026-10-20","latestMilestone":"2026-10-13","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":0,"leaps":{"ai":0,"field":0},"latestPapers":8,"keyPapers":2},"milestones":[{"date":"2023-07-28","title":{"zh":"RT-2：视觉-语言-动作模型","en":"RT-2 vision-language-action model"},"url":"https://deepmind.google/discover/blog/rt-2-new-model-translates-vision-and-language-into-action/","source":"Google DeepMind","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"system","typeLabel":{"zh":"系统发布","en":"System"},"why":{"zh":"首个把网络规模视觉语言知识迁移到机器人动作的 VLA","en":"First VLA transferring web knowledge to actions"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首个把网络规模视觉语言知识迁移到机器人动作的 VLA","en":"First VLA transferring web knowledge to actions"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首个把网络规模视觉语言知识迁移到机器人动作的 VLA","en":"First VLA transferring web knowledge to actions"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2023-10-13","title":{"zh":"Open X-Embodiment：22 种机器人、100 万条轨迹的开放数据集与 RT-X 模型","en":"Open X-Embodiment: 22 robot types, 1M+ trajectories, RT-X models"},"url":"https://arxiv.org/abs/2310.08864","source":"arXiv","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok","datePrecision":"day","type":"dataset","typeLabel":{"zh":"数据集","en":"Dataset"},"why":{"zh":"推动跨机器人通用策略，成为 VLA 训练基础","en":"Foundation dataset for cross-embodiment policies"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"推动跨机器人通用策略，成为 VLA 训练基础","en":"Foundation dataset for cross-embodiment policies"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"推动跨机器人通用策略，成为 VLA 训练基础","en":"Foundation dataset for cross-embodiment policies"}},"impactNotes":[],"verified":{"id":"arXiv:2310.08864","title":"Open X-Embodiment: Robotic Learning Datasets and RT-X Models","venue":"arXiv","date":"2023-10-13"},"addedIn":"v2"},{"date":"2024-10-31","title":{"zh":"π0 通用机器人策略","en":"π0 generalist robot policy"},"url":"https://www.physicalintelligence.company/blog/pi0","source":"Physical Intelligence","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"system","typeLabel":{"zh":"系统发布","en":"System"},"why":{"zh":"通用机器人策略完成叠衣等长时程灵巧任务","en":"Generalist policy for long-horizon dexterous tasks"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"通用机器人策略完成叠衣等长时程灵巧任务","en":"Generalist policy for long-horizon dexterous tasks"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"通用机器人策略完成叠衣等长时程灵巧任务","en":"Generalist policy for long-horizon dexterous tasks"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2025-03-12","title":{"zh":"Gemini Robotics 发布","en":"Gemini Robotics launched"},"url":"https://deepmind.google/discover/blog/gemini-robotics-brings-ai-into-the-physical-world/","source":"Google DeepMind","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"system","typeLabel":{"zh":"系统发布","en":"System"},"why":{"zh":"Gemini 大模型进入物理世界控制","en":"Gemini enters physical control"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"Gemini 大模型进入物理世界控制","en":"Gemini enters physical control"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"Gemini 大模型进入物理世界控制","en":"Gemini enters physical control"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-04-16","title":{"zh":"π0.7：可引导的通用机器人基础模型，零样本跨本体完成叠衣等任务（arXiv 2604.15483）","en":"π0.7: steerable generalist robot foundation model with zero-shot cross-embodiment"},"url":"https://arxiv.org/abs/2604.15483","source":"arXiv / Physical Intelligence","flags":["vendor","preprint"],"flagLabels":[{"zh":"厂商自述","en":"Vendor claim"},{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok-crossref","datePrecision":"day","type":"system","typeLabel":{"zh":"系统发布","en":"System"},"why":{"zh":"在未见过的任务和机器人上开箱即用，接近专门强化学习微调模型","en":"Out-of-the-box performance near specialised RL-finetuned models"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"在未见过的任务和机器人上开箱即用，接近专门强化学习微调模型","en":"Out-of-the-box performance near specialised RL-finetuned models"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"在未见过的任务和机器人上开箱即用，接近专门强化学习微调模型","en":"Out-of-the-box performance near specialised RL-finetuned models"}},"impactNotes":[],"verified":{"id":"arXiv:2604.15483","title":"$π_{0.7}$: a Steerable Generalist Robotic Foundation Model with Emergent Capabilities","venue":"arXiv","date":"2026-04-16"},"addedIn":"v2"},{"date":"2026-07-30","title":{"zh":"Gemini Robotics 2：全身控制人形机器人、精细灵巧操作与多机协作","en":"Gemini Robotics 2: whole-body humanoid control, dexterity and multi-robot teamwork"},"url":"https://deepmind.google/blog/gemini-robotics-2-brings-whole-body-intelligence-to-robots/","source":"Google DeepMind","flags":["vendor"],"flagLabels":[{"zh":"厂商自述","en":"Vendor claim"}],"linkCheck":"ok","datePrecision":"day","type":"system","typeLabel":{"zh":"系统发布","en":"System"},"why":{"zh":"VLA 从桌面机械臂扩展到全身人形控制，可数小时适配新本体","en":"Extends VLA to whole-body humanoids"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"VLA 从桌面机械臂扩展到全身人形控制，可数小时适配新本体","en":"Extends VLA to whole-body humanoids"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"VLA 从桌面机械臂扩展到全身人形控制，可数小时适配新本体","en":"Extends VLA to whole-body humanoids"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-10-13","title":{"zh":"Physical Intelligence 开放 Pi API 公测：托管微调和推理","en":"Physical Intelligence opens Pi API beta for hosted fine-tuning and inference"},"url":"https://www.pi.website/blog/api-beta","source":"Physical Intelligence","flags":["vendor"],"flagLabels":[{"zh":"厂商自述","en":"Vendor claim"}],"linkCheck":"ok","datePrecision":"day","type":"industry","typeLabel":{"zh":"产业/采用","en":"Industry"},"why":{"zh":"机器人基础模型开始以云服务形式供给，降低部署门槛","en":"Robot foundation models offered as a service"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"机器人基础模型开始以云服务形式供给，降低部署门槛","en":"Robot foundation models offered as a service"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"机器人基础模型开始以云服务形式供给，降低部署门槛","en":"Robot foundation models offered as a service"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-10-13","title":{"zh":"韩国产业通商部监管沙盒批准 Robros 的 20 台 AI 双足人形机器人与工人同场作业，试点覆盖半导体、汽车、造船和物流四类工厂","en":"Korea’s regulatory sandbox approves 20 Robros AI humanoids to work alongside people in semiconductor, auto, shipbuilding and logistics sites"},"url":"https://en.sedaily.com/finance/2026/10/14/korea-clears-safety-rules-to-put-20-humanoid-robots-on","source":"Seoul Economic Daily","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"regulatory","typeLabel":{"zh":"监管","en":"Regulatory"},"why":{"zh":"AI：机器人需在人员共享空间自主识别环境、抓取搬运；学科：政府首次为人形机器人在真实工厂与人同场作业给出安全监管豁免，是从演示走向产线的制度节点","en":"AI: robots must perceive and manipulate in human-shared spaces. Field: first government safety exemption for humanoids working beside people in real plants"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：机器人需在人员共享空间自主识别环境、抓取搬运；学科：政府首次为人形机器人在真实工厂与人同场作业给出安全监管豁免，是从演示走向产线的制度节点","en":"AI: robots must perceive and manipulate in human-shared spaces. Field: first government safety exemption for humanoids working beside people in real plants"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：机器人需在人员共享空间自主识别环境、抓取搬运；学科：政府首次为人形机器人在真实工厂与人同场作业给出安全监管豁免，是从演示走向产线的制度节点","en":"AI: robots must perceive and manipulate in human-shared spaces. Field: first government safety exemption for humanoids working beside people in real plants"}},"impactNotes":[],"verified":null,"addedIn":"v2-20261022pm"}],"papersRepos":[{"title":"Physical-Intelligence/openpi","url":"https://github.com/Physical-Intelligence/openpi","date":null,"kind":"repo","linkCheck":"ok"},{"title":"Open X-Embodiment (arXiv 2310.08864)","url":"https://arxiv.org/abs/2310.08864","date":"2023-10","kind":"paper","linkCheck":"ok"},{"title":"$π_{0.7}$: a Steerable Generalist Robotic Foundation Model with Emergent Capabilities","url":"https://arxiv.org/abs/2604.15483","date":"2026-04-16","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"Google DeepMind","url":"https://deepmind.google"},{"name":"Physical Intelligence","url":"https://www.physicalintelligence.company"},{"name":"NVIDIA Isaac","url":"https://developer.nvidia.com/isaac"}],"news":[{"id":"5df80c445dce","date":"2026-10-20","title":"How Peristaltic Pumps Are Enabling the Next Generation of Robotic Surgery","url":"https://hitconsultant.net/2026/10/20/how-peristaltic-pumps-enable-next-generation-robotic-surgery-watson-marlow-rodd-turnquist/","source":"HIT Consultant","lang":"en","kind":"news","summaryZh":"HIT Consultant于2026-10-20发布的文章，涉及AI 医疗、AI 教育、AI 机器人，原题《How Peristaltic Pumps Are Enabling the Next Generation of Robotic Surgery》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"1d528d3cc018","date":"2026-10-20","title":"From wombat burrows to Noah's Ark: Robot joins search","url":"https://techxplore.com/news/2026-10-wombat-burrows-noah-ark-robot.html","source":"TechXplore 机器人","lang":"en","kind":"news","summaryZh":"TechXplore 机器人于2026-10-20发布的文章，涉及AI 机器人，原题《From wombat burrows to Noah's Ark: Robot joins search》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"ecea94da2879","date":"2026-10-20","title":"启元机器人新进展：北大董豪团队推出 HIL-UMI，让后训练摆脱「真机依赖」","url":"https://www.leiphone.com/category/ai/XgCxObsqu8pc44px.html","source":"雷峰网","lang":"zh","kind":"news","summaryZh":"雷峰网于2026-10-20发布的文章，涉及AI 机器人、AI 科研与调研、AI 化学材料，原题《启元机器人新进展：北大董豪团队推出 HIL-UMI，让后训练摆脱「真机依赖」》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"f55154dcc729","date":"2026-10-20","title":"具身智能，卡在哪——王兴兴的\"一两年\"到底承诺了什么","url":"https://www.tmtpost.com/8174435.html","source":"钛媒体","lang":"zh","kind":"news","summaryZh":"钛媒体于2026-10-20发布的文章，涉及AI 机器人，原题《具身智能，卡在哪——王兴兴的\"一两年\"到底承诺了什么》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"dadaf1a72477","date":"2026-10-20","title":"Gecko Robotics secures $300M contract with the U.S. Navy","url":"https://www.therobotreport.com/gecko-robotics-earns-300m-contract-with-the-u-s-navy/","source":"The Robot Report","lang":"en","kind":"news","summaryZh":"The Robot Report于2026-10-20发布的文章，涉及AI 机器人，原题《Gecko Robotics secures $300M contract with the U.S. Navy》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"be071cededa0","date":"2026-10-20","title":"AGCO and Agility Robotics Advance Smart Manufacturing with Humanoid Robot Pilot in Hesston, Kansas, Facility","url":"http://www.RoboticsTomorrow.com/news/2026/10/20/agco-and-agility-robotics-advance-smart-manufacturing-with-humanoid-robot-pilot-in-hesston-kansas-facility/27260","source":"Robotics Tomorrow","lang":"en","kind":"news","summaryZh":"Robotics Tomorrow于2026-10-20发布的文章，涉及AI 机器人，原题《AGCO and Agility Robotics Advance Smart Manufacturing with Humanoid Robot Pilot in Hesston》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"12d9e20da3d8","date":"2026-10-20","title":"Tennant Company Introduces Its First Autonomous Robotic Vacuum, the P1 VAC","url":"http://www.RoboticsTomorrow.com/news/2026/10/20/tennant-company-introduces-its-first-autonomous-robotic-vacuum-the-p1-vac/27262","source":"Robotics Tomorrow","lang":"en","kind":"news","summaryZh":"Robotics Tomorrow于2026-10-20发布的文章，涉及AI 机器人，原题《Tennant Company Introduces Its First Autonomous Robotic Vacuum, the P1 VAC》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"43a51fc3c020","date":"2026-10-20","title":"2027 TRR50 Robotics Innovation Awards open nominations","url":"https://www.therobotreport.com/2027-trr50-robotics-innovation-awards-open-nominations/","source":"The Robot Report","lang":"en","kind":"news","summaryZh":"The Robot Report于2026-10-21发布的文章，涉及AI 机器人，原题《2027 TRR50 Robotics Innovation Awards open nominations》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"9f753769042b","date":"2026-10-20","title":"优必选成立具身智能公司，注册资本2000万元","url":"https://36kr.com/newsflashes/4034030028952452?f=rss","source":"36氪","lang":"zh","kind":"news","summaryZh":"36氪于2026-10-20发布的文章，涉及AI 机器人、AI 商业融资、AI 金融，原题《优必选成立具身智能公司，注册资本2000万元》，详见原文链接。","via":"collector-run-20261020","linkCheck":"from-feed"},{"id":"62b128a7c8d4","date":"2026-10-20","title":"山东：探索“具身智能 + 旅游”落地应用，支持山岳型景区投放登山助力机器人","url":"https://www.ithome.com/1/015/388.htm","source":"IT之家","lang":"zh","kind":"news","summaryZh":"IT之家于2026-10-20发布的文章，涉及AI 机器人，原题《山东：探索“具身智能 + 旅游”落地应用，支持山岳型景区投放登山助力机器人》，详见原文链接。","via":"collector-run-20261020","linkCheck":"from-feed"},{"id":"da5ee22c5b63","date":"2026-10-13","title":"Pi API beta","url":"https://www.pi.website/blog/api-beta","source":"Physical Intelligence","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"ok"},{"id":"82c2a5cb1208","date":"2026-07-28","title":"Gemini Robotics 2 brings whole body intelligence to robots","url":"https://deepmind.google/blog/gemini-robotics-2-brings-whole-body-intelligence-to-robots/","source":"Google DeepMind Blog","lang":"en","kind":"news","summaryZh":"Google DeepMind Blog于2026-07-28发布的文章，涉及AI 机器人，原题《Gemini Robotics 2 brings whole body intelligence to robots》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"}],"review":{"lastReviewed":"2026-10-22","cycleDays":90,"nextReview":"2027-01-19"},"urls":{"zh":"/zh-hans/disciplines/robotics/","en":"/en/disciplines/robotics/"},"dimensions":{"method":{"score":3,"label":{"zh":"研究方法","en":"Research method"},"evidence":"VLA 基础模型已是机器人学习主流研究范式","source":"https://arxiv.org/abs/2410.24164","weight":0.3},"discovery":{"score":2,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"泛化能力明显提升（π0.7、Gemini Robotics 2），但真实环境可靠性未解决","source":"https://deepmind.google/blog/gemini-robotics-2-brings-whole-body-intelligence-to-robots/","weight":0.35},"adoption":{"score":2,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"工厂试点多，规模部署少","source":null,"weight":0.25},"education":{"score":1,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到课程证据","source":null,"weight":0.1}},"compositeScore":2.2,"levelCapped":false,"levelChange":{"from":2,"to":2,"direction":"same","reason":null},"biggestBreakthrough":{"date":"2026-07-30","title":{"zh":"Gemini Robotics 2：全身控制人形机器人、精细灵巧操作与多机协作","en":"Gemini Robotics 2: whole-body humanoid control, dexterity and multi-robot teamwork"},"url":"https://deepmind.google/blog/gemini-robotics-2-brings-whole-body-intelligence-to-robots/","aiImpact":"major","fieldImpact":"major","why":{"zh":"VLA 从桌面机械臂扩展到全身人形控制，可数小时适配新本体","en":"Extends VLA to whole-body humanoids"}},"keyPapers":[{"title":"Open X-Embodiment: Robotic Learning Datasets and RT-X Models","titleZh":"Open X-Embodiment：22 种机器人、100 万条轨迹的开放数据集与 RT-X 模型","id":"arXiv:2310.08864","url":"https://arxiv.org/abs/2310.08864","venue":"arXiv","date":"2023-10-13","why":{"zh":"推动跨机器人通用策略，成为 VLA 训练基础","en":"Foundation dataset for cross-embodiment policies"},"fieldImpact":"major","aiImpact":"major"},{"title":"$π_{0.7}$: a Steerable Generalist Robotic Foundation Model with Emergent Capabilities","titleZh":"π0.7：可引导的通用机器人基础模型，零样本跨本体完成叠衣等任务（arXiv 2604.15483）","id":"arXiv:2604.15483","url":"https://arxiv.org/abs/2604.15483","venue":"arXiv","date":"2026-04-16","why":{"zh":"在未见过的任务和机器人上开箱即用，接近专门强化学习微调模型","en":"Out-of-the-box performance near specialised RL-finetuned models"},"fieldImpact":"major","aiImpact":"major"}],"latest_papers":[{"id":"arXiv:2610.30136","title":"Agentic World Action Models","titleZh":"AW-0：智能体式世界动作模型","summaryZh":"把基于代码的规划推理与视频、动作联合预测结合，让机器人既能复用技能又有物理动态的预测依据。","summaryEn":"AW-0 couples code-based planning with joint video and action prediction for compositional robot manipulation.","authors":["Zhuoyang Liu","Dantong Niu","Botong Zhang","Boning Shao"],"date":"2026-10-21","url":"https://arxiv.org/abs/2610.30136","site":"arXiv","venue":null,"category":"cs.RO","fetchedAt":"2026-10-22T10:09:04"},{"id":"arXiv:2610.30097","title":"AtomicRL: Atomic Reinforcement Learning for Long-Horizon Robotic Manipulation","titleZh":"AtomicRL：面向长程机器人操作的原子化强化学习","summaryZh":"把长程任务拆成原子子任务给出奖励，缓解视觉–语言–动作模型强化学习中的稀疏奖励与子任务掌握不均。","summaryEn":"Atomic subtask-level RL for VLA policies tackles sparse, biased credit and uneven subtask mastery in long-horizon manipulation.","authors":["Likui Zhang","Zhe Zhu","Zhihao Zhan","Yibo Bai"],"date":"2026-10-21","url":"https://arxiv.org/abs/2610.30097","site":"arXiv","venue":null,"category":"cs.RO","fetchedAt":"2026-10-22T10:09:04"},{"id":"arXiv:2610.28626","title":"UniWALM: Unified World Action Language Model for Embodied Robot Control","titleZh":"UniWALM：统一的世界–动作–语言模型用于具身机器人控制","summaryZh":"在同一混合 Transformer 主干中同步去噪视频、动作和语言，生成的语言作为下一步的记忆被再次读取，无需外部规划器，并提出带校正的流匹配目标保证动作可靠。","summaryEn":"A single backbone co-generating video, action and language, using generated language as memory for robot control.","authors":["Rahul Sajnani","Diego Romeres","Mingyuan Zhang","Kai Gao"],"date":"2026-10-20","url":"https://arxiv.org/abs/2610.28626","site":"arXiv","venue":null,"category":"cs.RO","fetchedAt":"2026-10-22T10:09:04"},{"id":"arXiv:2610.28589","title":"AgenticRoboRL: Bringing Coding-Agent Post-Training to Agentic Robotics","titleZh":"AgenticRoboRL：把编程智能体式后训练带到智能体机器人","summaryZh":"用监督微调加多轮强化学习后训练机器人规划器，通过适配器兼容 VLA 技能调用与控制代码生成等不同执行方式，在 LIBERO-PRO 上提升 9B 模型成功率。","summaryEn":"Post-trains robot planners with SFT and multi-turn RL across harnesses, improving a 9B planner on LIBERO-PRO.","authors":["Xinxin Liu","Bingxin Xu","Wenrui Bao","Ming Li"],"date":"2026-10-20","url":"https://arxiv.org/abs/2610.28589","site":"arXiv","venue":null,"category":"cs.RO","fetchedAt":"2026-10-22T10:09:04"},{"id":"arXiv:2610.28530","title":"CoGaussian: Open-Vocabulary Gaussian-based Collaborative 3D Scene Understanding","titleZh":"CoGaussian：基于高斯的开放词汇协同三维场景理解","summaryZh":"用与基础模型对齐的稀疏三维高斯在多个智能体间传递并融合几何与语义信息，在 Semantic-OPV2V 上优于单智能体开放词汇方法。","summaryEn":"Foundation-model-aligned sparse 3D Gaussians for open-vocabulary collaborative occupancy across agents.","authors":["Om Shivam Verma","Shaya Farahmand","Lingjun Zhao","Lu Gan"],"date":"2026-10-20","url":"https://arxiv.org/abs/2610.28530","site":"arXiv","venue":null,"category":"cs.CV","fetchedAt":"2026-10-22T10:09:04"},{"id":"arXiv:2610.26855","title":"Ropedia W0: Scaling Loco-Manipulation World Models across Embodiments with Human Experience","titleZh":"Ropedia W0：借助人类经验跨形态扩展移动操作世界模型","summaryZh":"发布首个跨人类与多种机器人的第一视角移动-操作世界基础模型，并整理 1 万小时人类数据和 3000 小时机器人数据。","summaryEn":"First egocentric loco-manipulation world model across humans and robots, with 10k h human and 3k h robot data.","authors":["Tianqi Liu","Hanfei Zhao","Qifeng Liang","Hui En Pang"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.26855","site":"arXiv","venue":null,"category":"cs.CV","fetchedAt":"2026-10-22T10:09:04"},{"id":"arXiv:2610.26854","title":"Physical Words: A Vocabulary for Structuring Latent Reasoning in Robot Manipulation","titleZh":"物理词汇：为机器人操作的潜在推理建立结构","summaryZh":"提出 PhysWord，用原子动作动词作为语义锚点组织物理动态的连续表示，先推理再行动。","summaryEn":"PhysWord structures latent physical reasoning with atomic action verbs before acting.","authors":["Jiaming Liu","Xinghao Zhu","Yiming Zhang","Shanghang Zhang"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.26854","site":"arXiv","venue":null,"category":"cs.RO","fetchedAt":"2026-10-22T10:09:04"},{"id":"arXiv:2610.26846","title":"GPT-6 Astra Is a Native Visuo-Tactile Policy for Contact-Rich Manipulation","titleZh":"GPT-6 Astra 是原生的视觉-触觉操作策略","summaryZh":"论文称首次展示 GPT-6 Astra 能直接理解视觉式触觉图像并用于接触丰富的操作，在 19 个仿真任务上少样本闭环控制。","summaryEn":"Claims GPT-6 Astra directly interprets visuo-tactile images for few-shot contact-rich manipulation on 19 simulated tasks.","authors":["Chen Yang","Linzhe Shi","Hang Zhang","Mu Xu"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.26846","site":"arXiv","venue":null,"category":"cs.RO","fetchedAt":"2026-10-22T10:09:04"}]},{"id":"aerospace","slug":"aerospace","category":"eng-computing","categoryName":{"zh":"工程与计算","en":"Engineering & computing"},"name":{"zh":"航空航天与机械工程","en":"Aerospace & mechanical engineering"},"classification":{"oecdFos":"2.3","moe":"0825 航空宇航科学与技术"},"level":2,"levelKey":"breakthrough","levelLabel":{"zh":"局部突破","en":"Local breakthroughs"},"trend":"flat","reason":{"zh":"DARPA ACE 项目中 AI 驾驶 X-62A 与有人驾驶 F-16 进行视距内空战测试（2024 年公布），属工程实践的局部突破；设计与制造环节仍以辅助为主。","en":"Under DARPA ACE, AI flew the X-62A in within-visual-range dogfights against a crewed F-16 (announced 2024) — a local breakthrough; design and manufacturing use is still mostly assistive."},"reasonSources":[{"url":"https://www.darpa.mil/news/2024/ace-ai-aerospace","linkCheck":"ok"}],"directions":{"zh":["自主飞行","生成式设计","数字孪生","代理模型仿真"],"en":["Autonomous flight","Generative design","Digital twins","Surrogate simulation"]},"stats":{"milestones":6,"milestonesSince2025":4,"milestones2026":4,"milestonesLast6Months":1,"news":12,"latestProgress":"2026-10-19","latestMilestone":"2026-10-16","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":0,"leaps":{"ai":1,"field":0},"latestPapers":7,"keyPapers":1},"milestones":[{"date":"2023-08-30","title":{"zh":"Swift：深度强化学习无人机在竞速中击败人类世界冠军（Nature）","en":"Champion-level drone racing with deep RL (Nature)"},"url":"https://doi.org/10.1038/s41586-023-06419-4","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"competition","typeLabel":{"zh":"竞赛","en":"Competition"},"why":{"zh":"首次在实体竞技运动中由 AI 击败人类世界冠军","en":"First AI to beat human world champions in a physical sport"},"aiImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"首次在实体竞技运动中由 AI 击败人类世界冠军","en":"First AI to beat human world champions in a physical sport"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首次在实体竞技运动中由 AI 击败人类世界冠军","en":"First AI to beat human world champions in a physical sport"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-023-06419-4","title":"Champion-level drone racing using deep reinforcement learning","venue":"Nature","date":"2023-08-30"},"addedIn":"v2"},{"date":"2024-04-17","title":{"zh":"DARPA ACE：AI 控制的 X-62A（改装 F-16）首次与有人驾驶 F-16 进行真实空中格斗测试","en":"DARPA ACE: AI-flown X-62A conducts first in-air dogfight against a human-piloted F-16"},"url":"https://www.darpa.mil/news/2024/ace-ai-aerospace","source":"DARPA","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"competition","typeLabel":{"zh":"竞赛","en":"Competition"},"why":{"zh":"首次在真实战斗机上验证机器学习空战控制（试飞发生于 2023 年 9 月）","en":"First real-jet ML dogfighting (flights Sept 2023)"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首次在真实战斗机上验证机器学习空战控制（试飞发生于 2023 年 9 月）","en":"First real-jet ML dogfighting (flights Sept 2023)"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首次在真实战斗机上验证机器学习空战控制（试飞发生于 2023 年 9 月）","en":"First real-jet ML dogfighting (flights Sept 2023)"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-01-30","title":{"zh":"NASA JPL：毅力号火星车完成首次由 AI（视觉语言模型）规划路线的行驶（2025 年 12 月 8 日、10 日，分别行驶 210 米和 246 米）","en":"NASA JPL: Perseverance completes first AI-planned drives on Mars (Dec 8 & 10, 2025)"},"url":"https://www.jpl.nasa.gov/news/nasas-perseverance-rover-completes-first-ai-planned-drive-on-mars/","source":"NASA JPL","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"deployment","typeLabel":{"zh":"落地部署","en":"Deployment"},"why":{"zh":"生成式 AI 首次承担外星车辆的路线规划，经数字孪生验证后上传","en":"First generative-AI route planning on another planet, digital-twin validated"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"生成式 AI 首次承担外星车辆的路线规划，经数字孪生验证后上传","en":"First generative-AI route planning on another planet, digital-twin validated"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"生成式 AI 首次承担外星车辆的路线规划，经数字孪生验证后上传","en":"First generative-AI route planning on another planet, digital-twin validated"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-02-12","title":{"zh":"YFQ-42A 协同作战飞机首次半自主任务飞行（Collins Sidekick 自主软件，持续 4 小时以上）","en":"YFQ-42A CCA flies its first semi-autonomous mission (Collins Sidekick, 4+ hours)"},"url":"https://www.ga.com/ga-asi-achieves-new-milestone-with-semi-autonomous-cca-flight","source":"General Atomics","flags":["vendor"],"flagLabels":[{"zh":"厂商自述","en":"Vendor claim"}],"linkCheck":"ok","datePrecision":"day","type":"deployment","typeLabel":{"zh":"落地部署","en":"Deployment"},"why":{"zh":"第三方自主软件接入新型无人战机执行任务","en":"Third-party autonomy on a new uncrewed fighter"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"第三方自主软件接入新型无人战机执行任务","en":"Third-party autonomy on a new uncrewed fighter"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"第三方自主软件接入新型无人战机执行任务","en":"Third-party autonomy on a new uncrewed fighter"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-02-26","title":{"zh":"YFQ-44A 同一架次切换 Shield AI Hivemind 与 Anduril Lattice 两套自主软件","en":"YFQ-44A flies two vendors’ autonomy stacks in one sortie"},"url":"https://www.anduril.com/news/yfq-44a-flies-with-mission-autonomy-software-from-anduril-and-shield-ai","source":"Anduril","flags":["vendor"],"flagLabels":[{"zh":"厂商自述","en":"Vendor claim"}],"linkCheck":"ok","datePrecision":"day","type":"deployment","typeLabel":{"zh":"落地部署","en":"Deployment"},"why":{"zh":"验证自主软件可模块化替换（A-GRA 标准）","en":"Validates modular autonomy (A-GRA)"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"验证自主软件可模块化替换（A-GRA 标准）","en":"Validates modular autonomy (A-GRA)"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"验证自主软件可模块化替换（A-GRA 标准）","en":"Validates modular autonomy (A-GRA)"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-10-16","title":{"zh":"美国空军：F-22 飞行员在空中指挥 YFQ-44A 执行模拟作战任务","en":"USAF: an F-22 pilot directs a YFQ-44A in flight for simulated combat tasks"},"url":"https://www.af.mil/News/Article-Display/Article/4627579/f-22-directs-collaborative-combat-aircraft-in-airborne-tactical-control-demonst/","source":"U.S. Air Force","flags":[],"flagLabels":[],"linkCheck":"site-blocks-bots","datePrecision":"day","type":"deployment","typeLabel":{"zh":"落地部署","en":"Deployment"},"why":{"zh":"有人-无人协同进入实机验证，空军计划 2032 年前部署 500 架","en":"Crewed–uncrewed teaming flight-tested; 500 planned by 2032"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"有人-无人协同进入实机验证，空军计划 2032 年前部署 500 架","en":"Crewed–uncrewed teaming flight-tested; 500 planned by 2032"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"有人-无人协同进入实机验证，空军计划 2032 年前部署 500 架","en":"Crewed–uncrewed teaming flight-tested; 500 planned by 2032"}},"impactNotes":[],"verified":null,"addedIn":"v2"}],"papersRepos":[{"title":"Champion-level drone racing using deep reinforcement learning","url":"https://doi.org/10.1038/s41586-023-06419-4","date":"2023-08-30","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"DARPA","url":"https://www.darpa.mil"},{"name":"NASA","url":"https://www.nasa.gov"}],"news":[{"id":"bc8d06e34fc5","date":"2026-10-19","title":"AI that reconstructs cloud-covered ground in satellite images with photo-like clarity","url":"https://techxplore.com/news/2026-10-ai-reconstructs-cloud-ground-satellite.html","source":"Phys.org 机器学习与AI","lang":"en","kind":"news","summaryZh":"Phys.org 机器学习与AI于2026-10-19发布的文章，涉及AI 科研与调研，原题《AI that reconstructs cloud-covered ground in satellite images with photo-like clarity》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"83c630f3e4ed","date":"2026-10-16","title":"F-22 directs collaborative combat aircraft in airborne tactical control demonstration","url":"https://www.af.mil/News/Article-Display/Article/4627579/f-22-directs-collaborative-combat-aircraft-in-airborne-tactical-control-demonst/","source":"U.S. Air Force","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"site-blocks-bots"},{"id":"6424ee7702c9","date":"2026-10-16","title":"Aslan Motion Technologies Launches aslanmotion.com with NDAA Focused Drone Electronics and U.S. Manufactured Motors","url":"http://www.RoboticsTomorrow.com/news/2026/10/16/aslan-motion-technologies-launches-aslanmotioncom-with-ndaa-focused-drone-electronics-and-us-manufactured-motors/27254","source":"Robotics Tomorrow","lang":"en","kind":"news","summaryZh":"Robotics Tomorrow于2026-10-16发布的文章，涉及AI 机器人，原题《Aslan Motion Technologies Launches aslanmotion.com with NDAA Focused Drone Electronics and》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"3ddb7e4f8f34","date":"2026-10-15","title":"3D vision automates sealant application in the aviation industry","url":"http://www.RoboticsTomorrow.com/news/2026/10/15/3d-vision-automates-sealant-application-in-the-aviation-industry-/27247","source":"Robotics Tomorrow","lang":"en","kind":"news","summaryZh":"Robotics Tomorrow于2026-10-15发布的文章，涉及AI 机器人，原题《3D vision automates sealant application in the aviation industry》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"da9e81f074a3","date":"2026-10-14","title":"DJI Delivers First Drone Dock with Automated, 25-Second Battery Swaps for Always-Ready Aerial Support","url":"http://www.RoboticsTomorrow.com/news/2026/10/14/dji-delivers-first-drone-dock-with-automated-25-second-battery-swaps-for-always-ready-aerial-support/27243","source":"Robotics Tomorrow","lang":"en","kind":"news","summaryZh":"Robotics Tomorrow于2026-10-14发布的文章，涉及AI 机器人，原题《DJI Delivers First Drone Dock with Automated, 25-Second Battery Swaps for Always-Ready Aer》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"801c62dc9a9c","date":"2026-10-14","title":"AI drone outpaces former world champions using one camera and no GPS","url":"https://techxplore.com/news/2026-10-ai-drone-outpaces-world-champions.html","source":"TechXplore 机器人","lang":"en","kind":"news","summaryZh":"TechXplore 机器人于2026-10-15发布的文章，涉及AI 机器人、AI 教育，原题《AI drone outpaces former world champions using one camera and no GPS》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"be28832d1f97","date":"2026-10-07","title":"New AI method uses engineering knowledge to estimate disaster damage from incomplete satellite imagery","url":"https://techxplore.com/news/2026-10-ai-method-knowledge-disaster-incomplete.html","source":"Tech Xplore","lang":"en","kind":"news","summaryZh":"新方法把工程知识融入 AI，在卫星影像被云烟遮挡时也能快速估算城市尺度的灾后建筑损伤。","via":"research-samples-20261010","linkCheck":"from-feed"},{"id":"687bdc531f48","date":"2026-10-06","title":"Planning Automatic Drone Battery Exchange for Reliable Dispatch","url":"http://www.RoboticsTomorrow.com/news/2026/10/06/planning-automatic-drone-battery-exchange-for-reliable-dispatch/27206","source":"Robotics Tomorrow","lang":"en","kind":"news","summaryZh":"Robotics Tomorrow于2026-10-06发布的文章，涉及AI 机器人、AI 气候能源，原题《Planning Automatic Drone Battery Exchange for Reliable Dispatch》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"ef4cb492eab8","date":"2026-10-03","title":"谷歌搭载AI芯片测试卫星发射升空","url":"https://news.sciencenet.cn/htmlnews/2026/10/572462.shtm","source":"科学网（中国科学报）","lang":"zh","kind":"news","summaryZh":"谷歌发射搭载 4 枚 TPU 的原型卫星，测试 AI 芯片在太空辐射和极端温度下的表现，探索太空 AI 算力。","via":"research-samples-20261010","linkCheck":"from-feed"},{"id":"52395d7243db","date":"2026-09-30","title":"Drones, AI and the future of wildlife conservation","url":"https://physicsworld.com/a/drones-ai-and-the-future-of-wildlife-conservation/","source":"Physics World（IOP）","lang":"en","kind":"news","summaryZh":"Physics World（IOP）于2026-09-30发布的文章，涉及AI 科研与调研、AI 机器人，原题《Drones, AI and the future of wildlife conservation》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"ba148ff29ec8","date":"2026-09-28","title":"Generative AI Gives Spacecraft the Autonomy Engineers Once Feared","url":"https://spectrum.ieee.org/generative-ai-in-space-exploration","source":"IEEE Spectrum：AI","lang":"en","kind":"news","summaryZh":"IEEE Spectrum：AI于2026-09-28发布的文章，涉及AI 综合，原题《Generative AI Gives Spacecraft the Autonomy Engineers Once Feared》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"8457a73ea7a7","date":"2026-09-06","title":"Drone WMDs Don’t Need Any New Technology","url":"https://www.lesswrong.com/posts/bGoo3NWzMAzsLQceJ/drone-wmds-don-t-need-any-new-technology","source":"LessWrong (精选)","lang":"en","kind":"news","summaryZh":"LessWrong (精选)于2026-09-07发布的文章，涉及AI 安全、AI 机器人，原题《Drone WMDs Don’t Need Any New Technology》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"}],"review":{"lastReviewed":"2026-10-21","cycleDays":90,"nextReview":"2027-01-19"},"urls":{"zh":"/zh-hans/disciplines/aerospace/","en":"/en/disciplines/aerospace/"},"dimensions":{"method":{"score":2,"label":{"zh":"研究方法","en":"Research method"},"evidence":"强化学习飞控、AI 代理模型用于气动设计","source":"https://doi.org/10.1038/s41586-023-06419-4","weight":0.3},"discovery":{"score":2,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"AI 无人机竞速首次击败世界冠军（Nature 2023）；AI 驾驶 F-16 空战测试","source":"https://doi.org/10.1038/s41586-023-06419-4","weight":0.35},"adoption":{"score":2,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"美国空军 CCA 项目多型半自主无人战机试飞，计划 2032 年前 500 架","source":"https://www.af.mil/News/Article-Display/Article/4627579/f-22-directs-collaborative-combat-aircraft-in-airborne-tactical-control-demonst/","weight":0.25},"education":{"score":1,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到课程证据","source":null,"weight":0.1}},"compositeScore":1.9,"levelCapped":false,"levelChange":{"from":2,"to":2,"direction":"same","reason":null},"biggestBreakthrough":{"date":"2023-08-30","title":{"zh":"Swift：深度强化学习无人机在竞速中击败人类世界冠军（Nature）","en":"Champion-level drone racing with deep RL (Nature)"},"url":"https://doi.org/10.1038/s41586-023-06419-4","aiImpact":"leap","fieldImpact":"major","why":{"zh":"首次在实体竞技运动中由 AI 击败人类世界冠军","en":"First AI to beat human world champions in a physical sport"}},"keyPapers":[{"title":"Champion-level drone racing using deep reinforcement learning","titleZh":"Swift：深度强化学习无人机在竞速中击败人类世界冠军（Nature）","id":"10.1038/s41586-023-06419-4","url":"https://doi.org/10.1038/s41586-023-06419-4","venue":"Nature","date":"2023-08-30","why":{"zh":"首次在实体竞技运动中由 AI 击败人类世界冠军","en":"First AI to beat human world champions in a physical sport"},"fieldImpact":"major","aiImpact":"leap"}],"latest_papers":[{"id":"arXiv:2610.20433","title":"A Practical Framework for Incremental Testing and Technology Maturation of Spacecraft Autonomy","titleZh":"航天器自主能力增量测试与技术成熟的实用框架","summaryZh":"结合地面离线测试（仿真、处理器在环、硬件在环、机器人仿真）与在轨 AI 看门狗监控，缩小 AI 航天器自主的“信任差距”。","summaryEn":"Combines offline ground testing and online AI watchdogs to close the trust gap for AI spacecraft autonomy.","authors":["Kerianne Hobbs","Ben Bycroft","Pamela Wood","Mathew W. Bissonnette"],"date":"2026-10-14","url":"https://arxiv.org/abs/2610.20433","site":"arXiv","venue":null,"category":"eess.SY","fetchedAt":"2026-10-22T10:09:11"},{"id":"arXiv:2610.20252","title":"Adversarial Reinforcement Learning for Robust Planetary Landing","titleZh":"对抗强化学习实现鲁棒的行星着陆","summaryZh":"用第二个智能体生成受物理约束的相关扰动，寻找困难的非标称条件，提升着陆制导策略的鲁棒性。","summaryEn":"An adversarial agent generates correlated physical perturbations to harden RL planetary-landing policies.","authors":["Miguel Leiva-Vélez","Nicolas Gaston-Rozado","Andrea Scorsoglio","Hodei Urrutxua"],"date":"2026-10-14","url":"https://arxiv.org/abs/2610.20252","site":"arXiv","venue":null,"category":"cs.LG","fetchedAt":"2026-10-22T10:09:11"},{"id":"arXiv:2610.18425","title":"CALM: Decentralized Control of Cooperative Aerial Manipulation with Multi-Agent Reinforcement Learning","titleZh":"CALM：多智能体强化学习实现协作空中操作的分布式控制","summaryZh":"多架无人机共同吊运篮子跟踪移动地面机器人，结合多智能体 PPO 策略与状态观测器应对风和载荷扰动。","summaryEn":"Decentralised MARL control for multiple UAVs jointly carrying a basket while tracking a ground robot.","authors":["Jiaxi Huo","Petre Ricioppo","Fausto Francesco Lizzio"],"date":"2026-10-13","url":"https://arxiv.org/abs/2610.18425","site":"arXiv","venue":null,"category":"cs.RO","fetchedAt":"2026-10-22T10:09:11"},{"id":"arXiv:2610.17468","title":"From Orbit Matching to Docking: A Hierarchical Reinforcement-Learning Approach to Multiphase Rendezvous Trajectory Design","titleZh":"从轨道匹配到对接：分层强化学习设计多阶段交会轨迹","summaryZh":"用分层强化学习贯穿轨道匹配、远程接近、近距离机动到末端对接的全过程，实现航天器自主制导。","summaryEn":"Hierarchical RL spans orbit matching to terminal docking for autonomous multiphase spacecraft rendezvous.","authors":["Mariko Storey-Matsutani","Anatoly Zlotnik","Lee Holguin","Zigfried Hampel-Arias"],"date":"2026-10-12","url":"https://arxiv.org/abs/2610.17468","site":"arXiv","venue":null,"category":"cs.RO","fetchedAt":"2026-10-22T10:09:11"},{"id":"arXiv:2610.15491","title":"Phase-Gated Constrained Reinforcement Learning for End-to-End Capture and Detumbling of Non-Cooperative Targets","titleZh":"相位门控约束强化学习实现非合作目标的端到端捕获与消旋","summaryZh":"单一约束强化学习策略完成从接近、无专用夹具捕获到完全消旋的整个碎片清除任务。","summaryEn":"A single constrained RL policy performs approach, fixture-free capture and detumbling of a tumbling target.","authors":["Álvaro Belmonte-Baeza","Celia Redondo-Verdú","José L. Ramón","Miguel Cazorla"],"date":"2026-10-11","url":"https://arxiv.org/abs/2610.15491","site":"arXiv","venue":null,"category":"cs.RO","fetchedAt":"2026-10-22T10:09:11"},{"id":"arXiv:2610.10517","title":"Unsupervised Maneuver-Aware Acoustic Fault Detection for Autonomous Drones","titleZh":"面向自主无人机的无监督、机动感知声学故障检测","summaryZh":"只用正常飞行录音训练，结合去噪与机动条件重建，检测无人机故障。","summaryEn":"Unsupervised maneuver-aware acoustic fault detection for drones trained only on nominal flights.","authors":["Ali M Ali","Ziyi Tang","Nurdaulet Nazarbay","Hashim A. Hashim"],"date":"2026-10-07","url":"https://arxiv.org/abs/2610.10517","site":"arXiv","venue":null,"category":"cs.SD","fetchedAt":"2026-10-22T10:09:11"},{"id":"arXiv:2610.03924","title":"Barrier-Shaped Recurrent Reinforcement Learning for Autonomous Landing on a Heaving Ship Deck","titleZh":"屏障塑形的循环强化学习用于无人机在起伏舰船甲板上自主着陆","summaryZh":"训练时评论家可看到未来甲板高度、执行器只用机载传感器，训练出能在起伏甲板上着陆的循环策略。","summaryEn":"An asymmetric actor–critic recurrent policy lands rotorcraft UAVs on heaving ship decks using only onboard sensing.","authors":["Ritwik Shankar","Chiranjeev Prachand"," Abhishek","Soumya Ranjan Sahoo"],"date":"2026-10-02","url":"https://arxiv.org/abs/2610.03924","site":"arXiv","venue":null,"category":"eess.SY","fetchedAt":"2026-10-22T10:09:11"}]},{"id":"energy-fusion","slug":"energy-fusion","category":"eng-computing","categoryName":{"zh":"工程与计算","en":"Engineering & computing"},"name":{"zh":"能源与核聚变","en":"Energy & fusion"},"classification":{"oecdFos":"2.7","moe":"0827 核科学与技术"},"level":2,"levelKey":"breakthrough","levelLabel":{"zh":"局部突破","en":"Local breakthroughs"},"trend":"up","reason":{"zh":"深度强化学习控制托卡马克等离子体（Nature 2022）、AI 预警并避免撕裂模不稳定（Nature 2024）；DeepMind 与 CFS 用 TORAX 为 SPARC 规划运行。","en":"Deep RL controls tokamak plasmas (Nature 2022) and AI avoids tearing instabilities (Nature 2024); DeepMind and CFS use TORAX to plan SPARC operations."},"reasonSources":[{"url":"https://www.nature.com/articles/s41586-021-04301-9","linkCheck":"ok"},{"url":"https://www.nature.com/articles/s41586-024-07024-9","linkCheck":"ok"},{"url":"https://deepmind.google/blog/bringing-ai-to-the-next-generation-of-fusion-energy/","linkCheck":"ok"}],"directions":{"zh":["等离子体控制","聚变数字孪生","电网调度与预测"],"en":["Plasma control","Fusion digital twins","Grid forecasting"]},"stats":{"milestones":7,"milestonesSince2025":4,"milestones2026":2,"milestonesLast6Months":2,"news":6,"latestProgress":"2026-10-20","latestMilestone":"2026-10-08","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":0,"leaps":{"ai":1,"field":0},"latestPapers":8,"keyPapers":5},"milestones":[{"date":"2022-02-16","title":{"zh":"深度强化学习磁控托卡马克等离子体（Nature）","en":"Deep RL magnetic control of tokamak plasmas (Nature)"},"url":"https://www.nature.com/articles/s41586-021-04301-9","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"首次用深度强化学习直接控制托卡马克全部磁场线圈并塑造多种等离子体形状","en":"First deep-RL control of all tokamak coils"},"aiImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"首次用深度强化学习直接控制托卡马克全部磁场线圈并塑造多种等离子体形状","en":"First deep-RL control of all tokamak coils"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首次用深度强化学习直接控制托卡马克全部磁场线圈并塑造多种等离子体形状","en":"First deep-RL control of all tokamak coils"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-021-04301-9","title":"Magnetic control of tokamak plasmas through deep reinforcement learning","venue":"Nature","date":"2022-02-16"},"addedIn":"v1"},{"date":"2024-02-21","title":{"zh":"AI 避免聚变等离子体撕裂模不稳定（Nature）","en":"AI avoids tearing instabilities in fusion plasma (Nature)"},"url":"https://www.nature.com/articles/s41586-024-07024-9","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"AI 提前约 300 毫秒预测并避免撕裂模不稳定","en":"Predicts and avoids tearing instabilities ~300 ms ahead"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI 提前约 300 毫秒预测并避免撕裂模不稳定","en":"Predicts and avoids tearing instabilities ~300 ms ahead"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI 提前约 300 毫秒预测并避免撕裂模不稳定","en":"Predicts and avoids tearing instabilities ~300 ms ahead"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-024-07024-9","title":"Avoiding fusion plasma tearing instability with deep reinforcement learning","venue":"Nature","date":"2024-02-21"},"addedIn":"v1"},{"date":"2024-06-10","title":{"zh":"TORAX：可微分、快速的托卡马克输运模拟器（Google DeepMind，arXiv 2406.06718）","en":"TORAX: fast differentiable tokamak transport simulator"},"url":"https://arxiv.org/abs/2406.06718","source":"arXiv","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok-crossref","datePrecision":"day","type":"system","typeLabel":{"zh":"系统发布","en":"System"},"why":{"zh":"为强化学习和优化提供快速物理模拟，被 CFS 用于 SPARC 规划","en":"Fast simulator used to plan SPARC"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"为强化学习和优化提供快速物理模拟，被 CFS 用于 SPARC 规划","en":"Fast simulator used to plan SPARC"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"为强化学习和优化提供快速物理模拟，被 CFS 用于 SPARC 规划","en":"Fast simulator used to plan SPARC"}},"impactNotes":[],"verified":{"id":"arXiv:2406.06718","title":"TORAX: A Fast and Differentiable Tokamak Transport Simulator in JAX","venue":"arXiv","date":"2024-06-10"},"addedIn":"v2"},{"date":"2025-06-16","title":{"zh":"DIII-D 上无需平衡重建的深度强化学习磁控（arXiv 2506.13267）","en":"Reconstruction-free magnetic control of DIII-D plasma with deep RL"},"url":"https://arxiv.org/abs/2506.13267","source":"arXiv","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"强化学习控制在美国最大托卡马克上实验验证","en":"RL control validated on DIII-D"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"强化学习控制在美国最大托卡马克上实验验证","en":"RL control validated on DIII-D"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"强化学习控制在美国最大托卡马克上实验验证","en":"RL control validated on DIII-D"}},"impactNotes":[],"verified":{"id":"arXiv:2506.13267","title":"Reconstruction-free magnetic control of DIII-D plasma with deep reinforcement learning","venue":"arXiv","date":"2025-06-16"},"addedIn":"v2"},{"date":"2025-10-16","title":{"zh":"DeepMind 与 CFS 合作，用 TORAX 规划 SPARC 运行","en":"DeepMind & CFS partner on TORAX for SPARC"},"url":"https://deepmind.google/blog/bringing-ai-to-the-next-generation-of-fusion-energy/","source":"Google DeepMind","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"industry","typeLabel":{"zh":"产业/采用","en":"Industry"},"why":{"zh":"DeepMind 与 CFS 合作，用 AI 规划 SPARC 运行","en":"DeepMind–CFS partnership"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"DeepMind 与 CFS 合作，用 AI 规划 SPARC 运行","en":"DeepMind–CFS partnership"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"DeepMind 与 CFS 合作，用 AI 规划 SPARC 运行","en":"DeepMind–CFS partnership"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-07-02","title":{"zh":"PACMAN：DIII-D 上集成诊断、机器学习预测与安全限值的实时 AI 控制框架，约 20 毫秒决策，提前约 200 毫秒预测撕裂模（Nuclear Fusion）","en":"Enabling integrated AI control on DIII-D (PACMAN): real-time ML control with safety limits, ~20 ms cycle (Nuclear Fusion)"},"url":"https://doi.org/10.1088/1741-4326/ae7f9d","source":"Nuclear Fusion","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"system","typeLabel":{"zh":"系统发布","en":"System"},"why":{"zh":"AI：模块化框架在执行前强制安全约束，已在 5 次 DIII-D 实验中运行；学科：为在大型托卡马克上组合多个 AI 控制器提供经同行评审的工程方案（PPPL 9 月发布新闻）","en":"AI: modular framework enforces safety limits before actuation, run in five DIII-D experiments. Field: peer-reviewed engineering path for combining AI controllers on a major tokamak"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：模块化框架在执行前强制安全约束，已在 5 次 DIII-D 实验中运行；学科：为在大型托卡马克上组合多个 AI 控制器提供经同行评审的工程方案（PPPL 9 月发布新闻）","en":"AI: modular framework enforces safety limits before actuation, run in five DIII-D experiments. Field: peer-reviewed engineering path for combining AI controllers on a major tokamak"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：模块化框架在执行前强制安全约束，已在 5 次 DIII-D 实验中运行；学科：为在大型托卡马克上组合多个 AI 控制器提供经同行评审的工程方案（PPPL 9 月发布新闻）","en":"AI: modular framework enforces safety limits before actuation, run in five DIII-D experiments. Field: peer-reviewed engineering path for combining AI controllers on a major tokamak"}},"impactNotes":[],"verified":{"id":"10.1088/1741-4326/ae7f9d","title":"Enabling integrated AI control on DIII-D: a control system design with state-of-the-art experiments","venue":"Nuclear Fusion","date":"2026-07-02"},"addedIn":"v2-20261022pm"},{"date":"2026-10-08","title":{"zh":"CFS 牵头 DOE Genesis Mission 第二阶段 SPARC AI 数字孪生项目","en":"CFS leads DOE Genesis Mission phase-2 AI digital twin for SPARC"},"url":"https://www.cfs.energy/news-and-media/commonwealth-fusion-systems-to-lead-ai-enabled-digital-twin-fusion-project-in-second-phase-of-genesis-mission/","source":"Commonwealth Fusion Systems","flags":["vendor"],"flagLabels":[{"zh":"厂商自述","en":"Vendor claim"}],"linkCheck":"ok","datePrecision":"day","type":"policy","typeLabel":{"zh":"政策","en":"Policy"},"why":{"zh":"国家级计划把 AI 数字孪生纳入首个商业聚变装置的运行规划（日期依据检索结果，待页面核实）","en":"National programme puts AI twins into SPARC operations planning"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"国家级计划把 AI 数字孪生纳入首个商业聚变装置的运行规划（日期依据检索结果，待页面核实）","en":"National programme puts AI twins into SPARC operations planning"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"国家级计划把 AI 数字孪生纳入首个商业聚变装置的运行规划（日期依据检索结果，待页面核实）","en":"National programme puts AI twins into SPARC operations planning"}},"impactNotes":[],"verified":null,"addedIn":"v2"}],"papersRepos":[{"title":"google-deepmind/torax","url":"https://github.com/google-deepmind/torax","date":null,"kind":"repo","linkCheck":"ok"},{"title":"Magnetic control of tokamak plasmas through deep reinforcement learning","url":"https://doi.org/10.1038/s41586-021-04301-9","date":"2022-02-16","kind":"paper","linkCheck":"ok-crossref"},{"title":"Avoiding fusion plasma tearing instability with deep reinforcement learning","url":"https://doi.org/10.1038/s41586-024-07024-9","date":"2024-02-21","kind":"paper","linkCheck":"ok-crossref"},{"title":"TORAX: A Fast and Differentiable Tokamak Transport Simulator in JAX","url":"https://arxiv.org/abs/2406.06718","date":"2024-06-10","kind":"paper","linkCheck":"ok-crossref"},{"title":"Reconstruction-free magnetic control of DIII-D plasma with deep reinforcement learning","url":"https://arxiv.org/abs/2506.13267","date":"2025-06-16","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"Google DeepMind","url":"https://deepmind.google"},{"name":"Commonwealth Fusion Systems","url":"https://cfs.energy"},{"name":"Princeton PPPL","url":"https://www.pppl.gov"}],"news":[{"id":"44957f9878f9","date":"2026-10-20","title":"US restarts three shut-down nuclear reactors to feed AI data centers","url":"https://www.tomshardware.com/tech-industry/big-tech/us-takes-unprecedented-step-of-recommissioning-nuclear-plants-due-to-ai-power-demand-2-25-gw-currently-on-tap-from-refurbished-reactors","source":"Tom's Hardware AI","lang":"en","kind":"news","summaryZh":"Tom's Hardware AI于2026-10-20发布的文章，涉及AI 芯片硬件、AI 气候能源，原题《US restarts three shut-down nuclear reactors to feed AI data centers》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"c7e1f12893dd","date":"2026-10-19","title":"Great Sky, National Labs Explore Superconducting AI for Plasma Control","url":"https://www.hpcwire.com/off-the-wire/great-sky-national-labs-explore-superconducting-ai-for-plasma-control/","source":"HPCwire","lang":"en","kind":"news","summaryZh":"HPCwire于2026-10-20发布的文章，涉及AI 芯片硬件，原题《Great Sky, National Labs Explore Superconducting AI for Plasma Control》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"3f1c66f2ad6b","date":"2026-10-19","title":"Superconducting Neuromorphic Chip Passes Fusion Test","url":"https://www.hpcwire.com/2026/10/19/superconducting-neuromorphic-chip-passes-fusion-test/","source":"HPCwire","lang":"en","kind":"news","summaryZh":"HPCwire于2026-10-20发布的文章，涉及AI 芯片硬件，原题《Superconducting Neuromorphic Chip Passes Fusion Test》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"3ae031dc33d2","date":"2026-10-08","title":"CFS to lead AI-enabled digital twin fusion project in second phase of Genesis Mission","url":"https://www.cfs.energy/news-and-media/commonwealth-fusion-systems-to-lead-ai-enabled-digital-twin-fusion-project-in-second-phase-of-genesis-mission/","source":"Commonwealth Fusion Systems","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"ok"},{"id":"fb1e407f7e73","date":"2026-10-08","title":"The U.S. Nuclear Plant Fleet Is Leaning Into AI","url":"https://spectrum.ieee.org/ai-assistants-nuclear-power-plant","source":"IEEE Spectrum","lang":"en","kind":"news","summaryZh":"一向谨慎保守的美国核电行业，开始在电站运营中引入 AI 助手。","via":"research-samples-20261010","linkCheck":"from-feed"},{"id":"b2639a75c3ce","date":"2026-09-30","title":"Forecasting space weather risks on power grids","url":"https://www.microsoft.com/en-us/research/blog/forecasting-space-weather-risks-on-power-grids/","source":"Microsoft Research","lang":"en","kind":"news","summaryZh":"Microsoft Research于2026-10-01发布的文章，涉及AI 科研与调研、AI 气候能源、AI 安全，原题《Forecasting space weather risks on power grids》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"}],"review":{"lastReviewed":"2026-10-22","cycleDays":90,"nextReview":"2027-01-19"},"urls":{"zh":"/zh-hans/disciplines/energy-fusion/","en":"/en/disciplines/energy-fusion/"},"dimensions":{"method":{"score":2,"label":{"zh":"研究方法","en":"Research method"},"evidence":"代理模型、强化学习控制在主要装置上实验验证","source":"https://arxiv.org/abs/2506.13267","weight":0.3},"discovery":{"score":2,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"Nature 2022/2024 两项等离子体控制突破","source":"https://www.nature.com/articles/s41586-024-07024-9","weight":0.35},"adoption":{"score":1,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"仍在实验装置和规划阶段","source":null,"weight":0.25},"education":{"score":0,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到证据","source":null,"weight":0.1}},"compositeScore":1.55,"levelCapped":false,"levelChange":{"from":2,"to":2,"direction":"same","reason":null},"biggestBreakthrough":{"date":"2022-02-16","title":{"zh":"深度强化学习磁控托卡马克等离子体（Nature）","en":"Deep RL magnetic control of tokamak plasmas (Nature)"},"url":"https://www.nature.com/articles/s41586-021-04301-9","aiImpact":"leap","fieldImpact":"major","why":{"zh":"首次用深度强化学习直接控制托卡马克全部磁场线圈并塑造多种等离子体形状","en":"First deep-RL control of all tokamak coils"}},"keyPapers":[{"title":"Magnetic control of tokamak plasmas through deep reinforcement learning","titleZh":"深度强化学习磁控托卡马克等离子体（Nature）","id":"10.1038/s41586-021-04301-9","url":"https://doi.org/10.1038/s41586-021-04301-9","venue":"Nature","date":"2022-02-16","why":{"zh":"首次用深度强化学习直接控制托卡马克全部磁场线圈并塑造多种等离子体形状","en":"First deep-RL control of all tokamak coils"},"fieldImpact":"major","aiImpact":"leap"},{"title":"Avoiding fusion plasma tearing instability with deep reinforcement learning","titleZh":"AI 避免聚变等离子体撕裂模不稳定（Nature）","id":"10.1038/s41586-024-07024-9","url":"https://doi.org/10.1038/s41586-024-07024-9","venue":"Nature","date":"2024-02-21","why":{"zh":"AI 提前约 300 毫秒预测并避免撕裂模不稳定","en":"Predicts and avoids tearing instabilities ~300 ms ahead"},"fieldImpact":"major","aiImpact":"major"},{"title":"TORAX: A Fast and Differentiable Tokamak Transport Simulator in JAX","titleZh":"TORAX：可微分、快速的托卡马克输运模拟器（Google DeepMind，arXiv 2406.06718）","id":"arXiv:2406.06718","url":"https://arxiv.org/abs/2406.06718","venue":"arXiv","date":"2024-06-10","why":{"zh":"为强化学习和优化提供快速物理模拟，被 CFS 用于 SPARC 规划","en":"Fast simulator used to plan SPARC"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Reconstruction-free magnetic control of DIII-D plasma with deep reinforcement learning","titleZh":"DIII-D 上无需平衡重建的深度强化学习磁控（arXiv 2506.13267）","id":"arXiv:2506.13267","url":"https://arxiv.org/abs/2506.13267","venue":"arXiv","date":"2025-06-16","why":{"zh":"强化学习控制在美国最大托卡马克上实验验证","en":"RL control validated on DIII-D"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Enabling integrated AI control on DIII-D: a control system design with state-of-the-art experiments","titleZh":"PACMAN：DIII-D 上集成诊断、机器学习预测与安全限值的实时 AI 控制框架，约 20 毫秒决策，提前约 200 毫秒预测撕裂模（Nuclear Fusion）","id":"10.1088/1741-4326/ae7f9d","url":"https://doi.org/10.1088/1741-4326/ae7f9d","venue":"Nuclear Fusion","date":"2026-07-02","why":{"zh":"AI：模块化框架在执行前强制安全约束，已在 5 次 DIII-D 实验中运行；学科：为在大型托卡马克上组合多个 AI 控制器提供经同行评审的工程方案（PPPL 9 月发布新闻）","en":"AI: modular framework enforces safety limits before actuation, run in five DIII-D experiments. Field: peer-reviewed engineering path for combining AI controllers on a major tokamak"},"fieldImpact":"major","aiImpact":"incremental"}],"latest_papers":[{"id":"arXiv:2610.26138","title":"Reinforcement Learning Controllers for Fusion Scenario Development","titleZh":"用于聚变放电方案设计的强化学习控制器","summaryZh":"用一个预先训练、以目标为条件的强化学习控制器替代每个托卡马克放电方案的逐一控制器调参。","summaryEn":"A single goal-conditioned RL controller replaces per-scenario tuning in tokamak scenario development.","authors":["Evgeny Adishchev","Maxim Nurgaliev","Dmitry Sorokin","Alexander Granovsky"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.26138","site":"arXiv","venue":null,"category":"physics.plasm-ph","fetchedAt":"2026-10-22T10:09:14"},{"id":"arXiv:2610.24525","title":"Reconstruction of Unobserved Core Electron Density in Tokamak Plasmas with a Neural Operator Transport Surrogate and Uncertainty Quantification","titleZh":"用神经算子输运替代模型重建托卡马克未观测的芯部电子密度","summaryZh":"用傅里叶神经算子作为快速可微的输运求解器替代，在诊断受限时反推电子密度剖面并给出不确定性。","summaryEn":"A Fourier neural operator transport surrogate reconstructs unobserved core electron density in tokamaks with uncertainty.","authors":["D. Sorokin","A. Granovskiy","M. Stokolesov","E. Adishchev"],"date":"2026-10-17","url":"https://arxiv.org/abs/2610.24525","site":"arXiv","venue":null,"category":"physics.plasm-ph","fetchedAt":"2026-10-22T10:09:14"},{"id":"arXiv:2610.23844","title":"ModeCast: One-Shot Prediction of Linear MHD Eigenmodes and Growth Rates in Tokamaks","titleZh":"ModeCast：一次性预测托卡马克线性 MHD 本征模与增长率","summaryZh":"基于 4.5 万多次 JOREK 模拟训练的首个代理模型，可给出 n=1–4 磁流体不稳定性的完整线性稳定性评估。","summaryEn":"ModeCast, trained on 45,000+ JOREK runs, predicts full linear MHD stability for n=1–4 modes.","authors":["Alex Higginbottom","Zara Z. Ercan","Abetharan Antony","Guzman Sanchez Gonzalez"],"date":"2026-10-16","url":"https://arxiv.org/abs/2610.23844","site":"arXiv","venue":null,"category":"physics.plasm-ph","fetchedAt":"2026-10-22T10:09:14"},{"id":"arXiv:2610.22817","title":"Multiple Instance Learning for Interpretable Plasmoid Detection in the FLARE Device","titleZh":"用多示例学习在 FLARE 装置中可解释地检测等离子体团","summaryZh":"把多示例学习用于磁重联实验中等离子体团检测，并能指出对结果影响最大的时间段。","summaryEn":"Multiple-instance learning detects plasmoids in FLARE reconnection experiments with interpretable timing.","authors":["Charles Marsden","Adam Robbins","Hantao Ji"],"date":"2026-10-16","url":"https://arxiv.org/abs/2610.22817","site":"arXiv","venue":null,"category":"physics.plasm-ph","fetchedAt":"2026-10-22T10:09:14"},{"id":"arXiv:2610.13822","title":"Physical-state supervision for design-only Gaussian process surrogates of implosion simulations","titleZh":"物理状态监督的内爆模拟高斯过程替代模型","summaryZh":"让高斯过程替代模型从设计变量学习模拟中的物理诊断量，预测时无需额外计算即可利用这些信息。","summaryEn":"A Gaussian-process surrogate of implosion simulations supervised by physical diagnostics learned from design variables.","authors":["Chang Liu","Hao Shen","Chuanying Li"],"date":"2026-10-09","url":"https://arxiv.org/abs/2610.13822","site":"arXiv","venue":null,"category":"physics.comp-ph","fetchedAt":"2026-10-22T10:09:14"},{"id":"arXiv:2610.12395","title":"A neural characteristic mapping method: Lagrangian PINNs based on flow maps for transport-dominated problems","titleZh":"神经特征映射法：基于流映射的拉格朗日 PINN 求解输运主导问题","summaryZh":"用物理信息神经网络逼近方程的流映射而非解本身，以更好地处理大梯度和精细结构。","summaryEn":"Lagrangian PINNs approximating flow maps for transport-dominated problems.","authors":["Martin Campos Pinto","Victor Fournet","Emmanuel Franck","Philipp Krah"],"date":"2026-10-08","url":"https://arxiv.org/abs/2610.12395","site":"arXiv","venue":null,"category":"math.NA","fetchedAt":"2026-10-22T10:09:14"},{"id":"arXiv:2610.11871","title":"WEST advanced wall protection achievements toward long pulse operation","titleZh":"WEST 先进壁保护助力长脉冲运行","summaryZh":"WEST 装置用多套诊断监测 10 个主要面向等离子体部件，部分高级处理采用机器学习，以避免壁热点。","summaryEn":"WEST monitors plasma-facing components with indicators including ML-based processes to avoid hot spots.","authors":["R. Mitteau","M. H. Aumeunier","L. Dubus","J. Gerardin"],"date":"2026-10-08","url":"https://arxiv.org/abs/2610.11871","site":"arXiv","venue":null,"category":"physics.plasm-ph","fetchedAt":"2026-10-22T10:09:14"},{"id":"arXiv:2610.01485","title":"Quantitative schlieren imaging of a laser-ionized plasma channel in atomic vapor using symbolic regression","titleZh":"用符号回归定量解析原子蒸气中激光电离等离子体通道的纹影图像","summaryZh":"为 AWAKE 等离子体尾场加速实验，从模拟数据中用符号回归得到连接纹影信号与通道半径、鞘层宽度的简洁公式，并重新分析既有测量。","summaryEn":"Symbolic regression yields compact formulas linking schlieren signals to plasma-channel parameters in the AWAKE experiment.","authors":["Gabor Demeter"],"date":"2026-10-01","url":"https://arxiv.org/abs/2610.01485","site":"arXiv","venue":null,"category":"physics.plasm-ph","fetchedAt":"2026-10-22T10:09:14"}]},{"id":"civil-architecture","slug":"civil-architecture","category":"eng-computing","categoryName":{"zh":"工程与计算","en":"Engineering & computing"},"name":{"zh":"建筑与土木工程","en":"Architecture & civil engineering"},"classification":{"oecdFos":"2.1","moe":"0813 建筑学 / 0814 土木工程"},"level":1,"levelKey":"assistive","levelLabel":{"zh":"辅助起步","en":"Assistive"},"trend":"flat","reason":{"zh":"使用面广但深度浅：RIBA 2026 报告称 74% 的建筑事务所在部分项目中使用 AI，主要是通用生成式工具；结构工程界强调 AI 嵌入既有软件并保留工程师判断，尚无公认的 AI 原创设计突破。","en":"Broad but shallow: RIBA's 2026 report says 74% of practices use AI on some projects, mostly general generative tools; structural engineers keep judgment central. No recognised AI-originated design breakthrough."},"reasonSources":[{"url":"https://www.riba.org/media/fecm2pl5/riba-ai-report-2026-final_v3.pdf","linkCheck":"ok"},{"url":"https://www.ncsea.com/news-post/structural-engineering-report-explores-the-future-of-ai-adoption-workforce-and-teams/","linkCheck":"ok"}],"directions":{"zh":["生成式方案设计","BIM 自动化","结构优化"],"en":["Generative concept design","BIM automation","Structural optimisation"]},"stats":{"milestones":7,"milestonesSince2025":6,"milestones2026":6,"milestonesLast6Months":5,"news":4,"latestProgress":"2026-10-20","latestMilestone":"2026-10-09","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":0,"leaps":{"ai":0,"field":0},"latestPapers":4,"keyPapers":2},"milestones":[{"date":"2017-05","title":{"zh":"Cha 等：用卷积神经网络检测混凝土裂缝（Computer-Aided Civil and Infrastructure Engineering）","en":"Cha et al.: deep-learning crack damage detection using CNNs (CACIE)"},"url":"https://doi.org/10.1111/mice.12263","source":"Computer-Aided Civil and Infrastructure Engineering","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"month","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"土木结构健康监测中深度学习视觉检测的奠基论文，被引数千次","en":"Foundational deep-learning visual inspection paper in SHM"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"土木结构健康监测中深度学习视觉检测的奠基论文，被引数千次","en":"Foundational deep-learning visual inspection paper in SHM"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"土木结构健康监测中深度学习视觉检测的奠基论文，被引数千次","en":"Foundational deep-learning visual inspection paper in SHM"}},"impactNotes":[],"verified":{"id":"10.1111/mice.12263","title":"Deep Learning‐Based Crack Damage Detection Using Convolutional Neural Networks","venue":"Computer-Aided Civil and Infrastructure Engineering","date":"2017-05"},"addedIn":"v2"},{"date":"2026-03-09","title":{"zh":"丹佛市议会以 10:1 批准 459 万美元、为期 5 年的 ComplyAI（CivCheck）AI 辅助规划审查合同","en":"Denver City Council approves a 5-year, $4.59M ComplyAI (CivCheck) AI-guided plan-review contract (10–1)"},"url":"https://www.coloradopolitics.com/2026/03/09/denver-approves-4-6m-ai-powered-permit-review-contract/","source":"Colorado Politics","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"deployment","typeLabel":{"zh":"落地部署","en":"Deployment"},"why":{"zh":"美国大城市在建筑审批中采购 AI 合规审查，人工复核仍必需；有议员质疑可靠性","en":"Major US city buys AI code-compliance review; human review still required"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"美国大城市在建筑审批中采购 AI 合规审查，人工复核仍必需；有议员质疑可靠性","en":"Major US city buys AI code-compliance review; human review still required"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"美国大城市在建筑审批中采购 AI 合规审查，人工复核仍必需；有议员质疑可靠性","en":"Major US city buys AI code-compliance review; human review still required"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-04-27","title":{"zh":"加拿大伯灵顿市上线 eCheck 辅助 AI 工具，对开发申请预检《安大略建筑规范》和分区条例","en":"Burlington, Canada launches eCheck assistive AI to pre-check development applications against building code and zoning"},"url":"https://www.burlington.ca/en/news/burlington-launches-new-assistive-ai-tool-to-power-an-improved-development-review-experience.aspx","source":"City of Burlington","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"deployment","typeLabel":{"zh":"落地部署","en":"Deployment"},"why":{"zh":"AI 进入建筑审批流程（最终决定仍由工作人员做出）","en":"AI enters permit review; staff retain decisions"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI 进入建筑审批流程（最终决定仍由工作人员做出）","en":"AI enters permit review; staff retain decisions"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI 进入建筑审批流程（最终决定仍由工作人员做出）","en":"AI enters permit review; staff retain decisions"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-05","title":{"zh":"系统综述：建筑全生命周期自动合规检查（Automation in Construction）","en":"Systematic review: automated compliance checking across the building lifecycle (Automation in Construction)"},"url":"https://doi.org/10.1016/j.autcon.2026.106859","source":"Automation in Construction","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"month","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"合规检查是 AI 进入结构与建筑设计的主要入口","en":"Compliance checking is AI’s main entry point"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"合规检查是 AI 进入结构与建筑设计的主要入口","en":"Compliance checking is AI’s main entry point"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"合规检查是 AI 进入结构与建筑设计的主要入口","en":"Compliance checking is AI’s main entry point"}},"impactNotes":[],"verified":{"id":"10.1016/j.autcon.2026.106859","title":"Automated compliance checking across the building lifecycle: Systematic and semantic review integrating PRISMA and deep search","venue":"Automation in Construction","date":"2026-05"},"addedIn":"v2"},{"date":"2026-07-22","title":{"zh":"RIBA《2026 AI 报告》：74% 英国建筑事务所在部分项目使用 AI，40% 用于合规检查","en":"RIBA AI Report 2026: 74% of UK practices use AI on some projects; 40% for compliance checks"},"url":"https://www.riba.org/news/ai-adoption-reaches-tipping-point-across-uk-architecture/","source":"RIBA","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"industry","typeLabel":{"zh":"产业/采用","en":"Industry"},"why":{"zh":"行业协会大样本调查显示采用度一年内从 59% 升到 74%（原“待核实”日期已确认 2026-07-22）","en":"Adoption rose from 59% to 74% in a year"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"行业协会大样本调查显示采用度一年内从 59% 升到 74%（原“待核实”日期已确认 2026-07-22）","en":"Adoption rose from 59% to 74% in a year"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"行业协会大样本调查显示采用度一年内从 59% 升到 74%（原“待核实”日期已确认 2026-07-22）","en":"Adoption rose from 59% to 74% in a year"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-08-27","title":{"zh":"多伦多市上线 AI 建筑许可申请预检工具（试点），申请人提交前自查问题","en":"City of Toronto launches AI Building Permit Application Pre-Check pilot"},"url":"https://www.toronto.ca/news/city-of-toronto-launches-building-permit-application-pre-check-to-support-faster-reviews/","source":"City of Toronto","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"deployment","typeLabel":{"zh":"落地部署","en":"Deployment"},"why":{"zh":"北美大城市把 AI 合规预检纳入审批","en":"Major city adds AI compliance pre-check"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"北美大城市把 AI 合规预检纳入审批","en":"Major city adds AI compliance pre-check"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"北美大城市把 AI 合规预检纳入审批","en":"Major city adds AI compliance pre-check"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-10-09","title":{"zh":"美国斯波坎市为山火灾后重建开放 Permittable AI 住宅许可申请完整性预审，一年内免费，市政仍保留正式审查权","en":"Spokane offers free Permittable AI intake-completeness pre-screening for wildfire-rebuild residential permits"},"url":"https://my.spokanecity.org/news/stories/2026/10/09/rebuilding-permit-application-prescreening-with-permittable-ai/","source":"City of Spokane","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"deployment","typeLabel":{"zh":"落地部署","en":"Deployment"},"why":{"zh":"AI：按该市现行提交要求校准，检查缺失文件和图纸；学科：继多伦多、伯灵顿之后又一城市把 AI 预审用于建筑许可，但仅限完整性检查、不替代官方审查","en":"AI: calibrated to the city’s submittal rules to flag missing documents and sheets. Field: another city adopts AI permit pre-checks, limited to completeness"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：按该市现行提交要求校准，检查缺失文件和图纸；学科：继多伦多、伯灵顿之后又一城市把 AI 预审用于建筑许可，但仅限完整性检查、不替代官方审查","en":"AI: calibrated to the city’s submittal rules to flag missing documents and sheets. Field: another city adopts AI permit pre-checks, limited to completeness"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：按该市现行提交要求校准，检查缺失文件和图纸；学科：继多伦多、伯灵顿之后又一城市把 AI 预审用于建筑许可，但仅限完整性检查、不替代官方审查","en":"AI: calibrated to the city’s submittal rules to flag missing documents and sheets. Field: another city adopts AI permit pre-checks, limited to completeness"}},"impactNotes":[],"verified":null,"addedIn":"v2-20261022pm"}],"papersRepos":[{"title":"Deep Learning‐Based Crack Damage Detection Using Convolutional Neural Networks","url":"https://doi.org/10.1111/mice.12263","date":"2017-05","kind":"paper","linkCheck":"ok-crossref"},{"title":"Automated compliance checking across the building lifecycle: Systematic and semantic review integrating PRISMA and deep search","url":"https://doi.org/10.1016/j.autcon.2026.106859","date":"2026-05","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"RIBA","url":"https://www.riba.org"},{"name":"NCSEA","url":"https://www.ncsea.com"}],"news":[{"id":"c93bd11ac406","date":"2026-10-20","title":"스카이폴AI, CAD·BIM 3D 설계 자동화 월드모델 ‘스카이-001’ 공개","url":"https://www.aitimes.com/news/articleView.html?idxno=216450","source":"AI Times Korea","lang":"ko","kind":"news","summaryZh":"美国 Skyfall AI 发布 SKY-001“世界模型”，可从想法或 2D 草图自动生成 CAD/BIM 三维设计（厂商自述）。","via":"pool-v2-20261021","linkCheck":"from-feed"},{"id":"b2d4bbc91ee6","date":"2026-10-20","title":"Autodesk Pushes AI Beyond Individual Design Tools with a Connected Assistant","url":"https://engtechnica.com/autodesk-pushes-ai-beyond-individual-design-tools-with-a-connected-assistant/","source":"ENGtechnica","lang":"en","kind":"news","summaryZh":"Autodesk 在 AU 2026 提出 Project Intelligence：由互联的 Autodesk Assistant 理解意图和项目上下文，跨 Revit、Forma、Civil 3D 调度工具与智能体（厂商规划）。","via":"manual-websearch-20261022","linkCheck":"ok"},{"id":"1728b9d73434","date":"2026-10-13","title":"Bluebeam Report Brings Construction’s AI Paradox to Light: Firms Running AI in Production Report Real Savings, but Only 28% Have Gotten That Far","url":"https://press.bluebeam.com/2026/10/bluebeam-report-brings-constructions-ai-paradox-to-light-firms-running-ai-in-production-report-real-savings-but-only-28-have-gotten-that-far/","source":"Bluebeam","lang":"en","kind":"news","summaryZh":"Bluebeam 行业报告：已在生产中用 AI 的建筑企业报告了实际节省，但只有 28% 的企业走到这一步。","via":"manual-websearch-20261021","linkCheck":"ok"},{"id":"6ecbcc7a7d94","date":"2026-10-12","title":"This AI startup says it can generate permit-ready blueprints in minutes","url":"https://techcrunch.com/2026/10/12/this-ai-startup-says-it-can-generate-permit-ready-blueprints-in-minutes/","source":"TechCrunch","lang":"en","kind":"news","summaryZh":"TechCrunch：Blueprints AI 声称能从 2D 草图或 PDF 几分钟内生成符合规范、可报批的建筑施工图（厂商自述）。","via":"manual-websearch-20261021","linkCheck":"ok"}],"review":{"lastReviewed":"2026-10-22","cycleDays":90,"nextReview":"2027-01-19"},"urls":{"zh":"/zh-hans/disciplines/civil-architecture/","en":"/en/disciplines/civil-architecture/"},"dimensions":{"method":{"score":1,"label":{"zh":"研究方法","en":"Research method"},"evidence":"结构工程 AI 研究多为合规检查、代理模型等辅助","source":"https://doi.org/10.1016/j.autcon.2026.106859","weight":0.3},"discovery":{"score":1,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"没有公认的 AI 原创设计突破","source":null,"weight":0.35},"adoption":{"score":2,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"RIBA 2026：74% 英国建筑事务所在部分项目使用 AI（2025 年 59%）","source":"https://www.riba.org/news/ai-adoption-reaches-tipping-point-across-uk-architecture/","weight":0.25},"education":{"score":1,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到课程证据","source":null,"weight":0.1}},"compositeScore":1.25,"levelCapped":false,"levelChange":{"from":1,"to":1,"direction":"same","reason":null},"biggestBreakthrough":{"date":"2017-05","title":{"zh":"Cha 等：用卷积神经网络检测混凝土裂缝（Computer-Aided Civil and Infrastructure Engineering）","en":"Cha et al.: deep-learning crack damage detection using CNNs (CACIE)"},"url":"https://doi.org/10.1111/mice.12263","aiImpact":"major","fieldImpact":"major","why":{"zh":"土木结构健康监测中深度学习视觉检测的奠基论文，被引数千次","en":"Foundational deep-learning visual inspection paper in SHM"}},"keyPapers":[{"title":"Deep Learning‐Based Crack Damage Detection Using Convolutional Neural Networks","titleZh":"Cha 等：用卷积神经网络检测混凝土裂缝（Computer-Aided Civil and Infrastructure Engineering）","id":"10.1111/mice.12263","url":"https://doi.org/10.1111/mice.12263","venue":"Computer-Aided Civil and Infrastructure Engineering","date":"2017-05","why":{"zh":"土木结构健康监测中深度学习视觉检测的奠基论文，被引数千次","en":"Foundational deep-learning visual inspection paper in SHM"},"fieldImpact":"major","aiImpact":"major"},{"title":"Automated compliance checking across the building lifecycle: Systematic and semantic review integrating PRISMA and deep search","titleZh":"系统综述：建筑全生命周期自动合规检查（Automation in Construction）","id":"10.1016/j.autcon.2026.106859","url":"https://doi.org/10.1016/j.autcon.2026.106859","venue":"Automation in Construction","date":"2026-05","why":{"zh":"合规检查是 AI 进入结构与建筑设计的主要入口","en":"Compliance checking is AI’s main entry point"},"fieldImpact":"incremental","aiImpact":"incremental"}],"latest_papers":[{"id":"doi:10.1017/asjcl.2026.10049","title":"AI-Generated Deepfake Financial Scams: A Missing Liability Regime for Consumer Protection Frameworks","titleZh":"AI 深度伪造金融诈骗：消费者保护框架中缺失的责任制度","summaryZh":"论证现有消费者保护框架缺少针对深度伪造诈骗的责任制度，未明确约束生成式 AI 开发者和社交平台。","summaryEn":"Argues consumer-protection law lacks a liability regime for GenAI deepfake financial scams.","authors":["A Lui","A Miglionico"],"date":"2026-10-19","url":"https://doi.org/10.1017/asjcl.2026.10049","site":"OpenAlex","venue":"Asian Journal of Comparative Law","category":null,"fetchedAt":"2026-10-22T10:09:19"},{"id":"doi:10.1007/s12273-026-1504-2","title":"Indoor temperature prediction methods for urban overheating: A systematic review","titleZh":"面向城市过热的室内温度预测方法：系统综述","summaryZh":"综述 46 项研究：物理模拟占 32.6%，统计 19.6%，机器学习 17.4%，深度学习 8.7%，混合方法 21.7%，并评估其城市尺度适用性。","summaryEn":"Review of 46 studies on indoor temperature prediction methods (ML 17.4%, DL 8.7%) for city-scale overheating.","authors":["Sepideh Motamedpooya","Rodney A. Stewart","Shannon Rutherford","Aaron J. E. Bach"],"date":"2026-10-19","url":"https://doi.org/10.1007/s12273-026-1504-2","site":"OpenAlex","venue":"Building Simulation","category":null,"fetchedAt":"2026-10-22T10:09:19"},{"id":"doi:10.1007/s44163-026-02462-y","title":"A multi‑modal recognition method for building structural materials integrating structured text description and adaptive decision weighting","titleZh":"融合结构化文本描述与自适应决策加权的建筑结构材料多模态识别","summaryZh":"结合图像和结构化文本描述识别建筑结构材料，克服光照、遮挡等对纯视觉方法的限制，服务结构安全评估。","summaryEn":"Multimodal recognition of building structural materials combining images and structured text.","authors":["Min Ma"],"date":"2026-10-17","url":"https://doi.org/10.1007/s44163-026-02462-y","site":"OpenAlex","venue":"Discover Artificial Intelligence","category":null,"fetchedAt":"2026-10-22T10:09:19"},{"id":"arXiv:2610.22436","title":"Generative adversarial networks with Monte Carlo dropout uncertainty quantification for structural health monitoring: Fibre Bragg grating data augmentation comparison in Hawk T1A aircraft","titleZh":"带蒙特卡洛 dropout 不确定性量化的生成对抗网络用于结构健康监测数据增强","summaryZh":"用条件 GAN 学习光纤布拉格光栅测得的结构响应特征，生成增强数据并量化不确定性。","summaryEn":"Conditional GANs with MC-dropout UQ augment fibre-Bragg-grating structural health monitoring data.","authors":["Alvaro Peiro","Marios Impraimakis"],"date":"2026-10-15","url":"https://arxiv.org/abs/2610.22436","site":"arXiv","venue":null,"category":"cs.LG","fetchedAt":"2026-10-22T10:09:19"}]},{"id":"weather-climate","slug":"weather-climate","category":"earth-env","categoryName":{"zh":"地球与环境","en":"Earth & environment"},"name":{"zh":"气象与气候科学","en":"Meteorology & climate science"},"classification":{"oecdFos":"1.5","moe":"0706 大气科学"},"level":3,"levelKey":"transforming","levelLabel":{"zh":"系统改造","en":"Systemic transformation"},"trend":"up","reason":{"zh":"AI 天气模型已进入业务预报：ECMWF 的 AIFS 于 2025-02 转为业务运行、2026-05-12 升级到 v2；NOAA 部署 AIGFS 等 AI 全球模型。GraphCast、GenCast、盘古等同行评审成果奠定了基础。","en":"AI weather models are operational: ECMWF's AIFS went operational in Feb 2025 and v2 on 12 May 2026; NOAA runs AI global models (AIGFS etc.). GraphCast, GenCast and Pangu laid the peer-reviewed groundwork."},"reasonSources":[{"url":"https://www.ecmwf.int/en/about/media-centre/news/2025/ecmwfs-ai-forecasts-become-operational","linkCheck":"ok"},{"url":"https://www.ecmwf.int/en/newsletter/187/news/implementation-aifs-v2","linkCheck":"ok"},{"url":"https://www.noaa.gov/news-release/noaa-deploys-new-generation-of-ai-driven-global-weather-models","linkCheck":"ok"}],"directions":{"zh":["数据驱动中期预报","集合概率预报","物理-AI 混合气候模型","极端天气"],"en":["Data-driven medium-range forecasting","Probabilistic ensembles","Hybrid physics-AI climate models","Extreme weather"]},"stats":{"milestones":12,"milestonesSince2025":8,"milestones2026":4,"milestonesLast6Months":4,"news":11,"latestProgress":"2026-10-20","latestMilestone":"2026-10-19","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":0,"leaps":{"ai":1,"field":4},"latestPapers":8,"keyPapers":7},"milestones":[{"date":"2023-07-05","title":{"zh":"盘古气象大模型（Nature）","en":"Pangu-Weather (Nature)"},"url":"https://www.nature.com/articles/s41586-023-06185-3","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"首个在中期预报精度上超过欧洲中心业务数值模式的 AI 模型","en":"First AI model to beat ECMWF’s operational NWP on medium-range skill"},"aiImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"首个在中期预报精度上超过欧洲中心业务数值模式的 AI 模型","en":"First AI model to beat ECMWF’s operational NWP on medium-range skill"}},"fieldImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"首个在中期预报精度上超过欧洲中心业务数值模式的 AI 模型","en":"First AI model to beat ECMWF’s operational NWP on medium-range skill"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-023-06185-3","title":"Accurate medium-range global weather forecasting with 3D neural networks","venue":"Nature","date":"2023-07-05"},"addedIn":"v1"},{"date":"2023-11-14","title":{"zh":"GraphCast 中期天气预报（Science）","en":"GraphCast medium-range forecasting (Science)"},"url":"https://www.science.org/doi/10.1126/science.adi2336","source":"Science","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"90% 指标优于 HRES，1 分钟出 10 天预报","en":"Beats HRES on 90% of targets, 10-day forecast in a minute"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"90% 指标优于 HRES，1 分钟出 10 天预报","en":"Beats HRES on 90% of targets, 10-day forecast in a minute"}},"fieldImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"90% 指标优于 HRES，1 分钟出 10 天预报","en":"Beats HRES on 90% of targets, 10-day forecast in a minute"}},"impactNotes":[],"verified":{"id":"10.1126/science.adi2336","title":"Learning skillful medium-range global weather forecasting","venue":"Science","date":"2023-12-22"},"addedIn":"v1"},{"date":"2024-07-22","title":{"zh":"NeuralGCM 混合大气模型（Nature）","en":"NeuralGCM hybrid atmospheric model (Nature)"},"url":"https://www.nature.com/articles/s41586-024-07744-y","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"AI 与物理混合的大气环流模型，可做气候模拟","en":"Hybrid ML–physics GCM usable for climate"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI 与物理混合的大气环流模型，可做气候模拟","en":"Hybrid ML–physics GCM usable for climate"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI 与物理混合的大气环流模型，可做气候模拟","en":"Hybrid ML–physics GCM usable for climate"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-024-07744-y","title":"Neural general circulation models for weather and climate","venue":"Nature","date":"2024-07-22"},"addedIn":"v1"},{"date":"2024-12-04","title":{"zh":"GenCast 概率集合预报（Nature）","en":"GenCast probabilistic ensemble forecasting (Nature)"},"url":"https://www.nature.com/articles/s41586-024-08252-9","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"概率集合预报在 97% 指标上超过 ECMWF ENS","en":"Beats ECMWF ENS on 97% of targets"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"概率集合预报在 97% 指标上超过 ECMWF ENS","en":"Beats ECMWF ENS on 97% of targets"}},"fieldImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"概率集合预报在 97% 指标上超过 ECMWF ENS","en":"Beats ECMWF ENS on 97% of targets"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-024-08252-9","title":"Probabilistic weather forecasting with machine learning","venue":"Nature","date":"2024-12-04"},"addedIn":"v1"},{"date":"2025-02-25","title":{"zh":"ECMWF AIFS 转为业务运行","en":"ECMWF AIFS becomes operational"},"url":"https://www.ecmwf.int/en/about/media-centre/news/2025/ecmwfs-ai-forecasts-become-operational","source":"ECMWF","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"deployment","typeLabel":{"zh":"落地部署","en":"Deployment"},"why":{"zh":"全球顶级预报中心首次把 AI 模型转为业务运行","en":"First operational AI model at a top forecasting centre"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"全球顶级预报中心首次把 AI 模型转为业务运行","en":"First operational AI model at a top forecasting centre"}},"fieldImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"全球顶级预报中心首次把 AI 模型转为业务运行","en":"First operational AI model at a top forecasting centre"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2025-03-20","title":{"zh":"Aardvark Weather：端到端数据驱动天气预报，直接从观测出发（Nature）","en":"Aardvark Weather: end-to-end data-driven forecasting from observations (Nature)"},"url":"https://doi.org/10.1038/s41586-025-08897-0","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"绕过传统数据同化，首次用单一模型从原始观测出预报","en":"Bypasses data assimilation, forecasting straight from observations"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"绕过传统数据同化，首次用单一模型从原始观测出预报","en":"Bypasses data assimilation, forecasting straight from observations"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"绕过传统数据同化，首次用单一模型从原始观测出预报","en":"Bypasses data assimilation, forecasting straight from observations"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-025-08897-0","title":"End-to-end data-driven weather prediction","venue":"Nature","date":"2025-03-20"},"addedIn":"v2"},{"date":"2025-05-21","title":{"zh":"Aurora 地球系统基础模型（Nature）","en":"Aurora Earth-system foundation model (Nature)"},"url":"https://www.nature.com/articles/s41586-025-09005-y","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"地球系统基础模型覆盖空气质量、海浪、热带气旋","en":"Earth-system foundation model"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"地球系统基础模型覆盖空气质量、海浪、热带气旋","en":"Earth-system foundation model"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"地球系统基础模型覆盖空气质量、海浪、热带气旋","en":"Earth-system foundation model"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-025-09005-y","title":"A foundation model for the Earth system","venue":"Nature","date":"2025-05-21"},"addedIn":"v1"},{"date":"2025-12-17","title":{"zh":"NOAA 部署新一代 AI 全球天气模型（AIGFS 等）","en":"NOAA deploys new AI-driven global weather models (AIGFS etc.)"},"url":"https://www.noaa.gov/news-release/noaa-deploys-new-generation-of-ai-driven-global-weather-models","source":"NOAA","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"deployment","typeLabel":{"zh":"落地部署","en":"Deployment"},"why":{"zh":"美国国家天气业务部署 AI 全球模型","en":"NOAA deploys AI global models"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"美国国家天气业务部署 AI 全球模型","en":"NOAA deploys AI global models"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"美国国家天气业务部署 AI 全球模型","en":"NOAA deploys AI global models"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-05-12","title":{"zh":"ECMWF 升级 AIFS Single / ENS v2","en":"ECMWF upgrades to AIFS Single/ENS v2"},"url":"https://www.ecmwf.int/en/newsletter/187/news/implementation-aifs-v2","source":"ECMWF","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"deployment","typeLabel":{"zh":"落地部署","en":"Deployment"},"why":{"zh":"AIFS v2 实施说明","en":"AIFS v2 implementation"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AIFS v2 实施说明","en":"AIFS v2 implementation"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AIFS v2 实施说明","en":"AIFS v2 implementation"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-05-12","title":{"zh":"ECMWF IFS 50r1 与 AIFS v2 同步上线：AIFS ENS 新增海浪和积雪预报","en":"ECMWF IFS 50r1 and AIFS v2 go live; AIFS ENS adds waves and snow cover"},"url":"https://www.ecmwf.int/en/about/media-centre/news/2026/ifs-cycle-50r1-aifsv2-live","source":"ECMWF","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"deployment","typeLabel":{"zh":"落地部署","en":"Deployment"},"why":{"zh":"全球领先预报中心的 AI 模型进入第二代业务版本","en":"Second operational generation of AI forecasts at ECMWF"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"全球领先预报中心的 AI 模型进入第二代业务版本","en":"Second operational generation of AI forecasts at ECMWF"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"全球领先预报中心的 AI 模型进入第二代业务版本","en":"Second operational generation of AI forecasts at ECMWF"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-10-14","title":{"zh":"微软 PBC：机器学习概率偏差校正推进次季节（2–6 周）预报，使 ECMWF AIFS 技巧翻倍（Nature）","en":"Advancing subseasonal forecasting with machine learning — Probabilistic Bias Correction (Nature)"},"url":"https://www.nature.com/articles/s41586-026-11126-x","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"AI：轻量模型学习任意预报系统的系统性误差，并在 ECMWF AI Weather Quest 中夺冠；学科：改进了 ECMWF 业务次季节预报 98% 的降水目标，填补 2–6 周预报空档","en":"AI: a light model learns any system’s biases and won ECMWF’s AI Weather Quest; Field: improves ECMWF subseasonal forecasts for 98% of precipitation targets"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：轻量模型学习任意预报系统的系统性误差，并在 ECMWF AI Weather Quest 中夺冠；学科：改进了 ECMWF 业务次季节预报 98% 的降水目标，填补 2–6 周预报空档","en":"AI: a light model learns any system’s biases and won ECMWF’s AI Weather Quest; Field: improves ECMWF subseasonal forecasts for 98% of precipitation targets"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：轻量模型学习任意预报系统的系统性误差，并在 ECMWF AI Weather Quest 中夺冠；学科：改进了 ECMWF 业务次季节预报 98% 的降水目标，填补 2–6 周预报空档","en":"AI: a light model learns any system’s biases and won ECMWF’s AI Weather Quest; Field: improves ECMWF subseasonal forecasts for 98% of precipitation targets"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-026-11126-x","title":"Advancing subseasonal forecasting with machine learning","venue":"Nature","date":"2026-10-14"},"addedIn":"v2"},{"date":"2026-10-19","title":{"zh":"中国气象局“风清”AI 全球短中期集合预报系统 V1.0 通过业务化评审并投入运行，有效预报时效 11 天","en":"CMA’s Fengqing AI global ensemble forecast system V1.0 passes operational review and goes live (11-day skill)"},"url":"https://www.chinanews.com/sh/2026/10-19/10717398.shtml","source":"中国新闻网","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"deployment","typeLabel":{"zh":"落地部署","en":"Deployment"},"why":{"zh":"AI：完全国产数据、模型、算力训练的 AI 集合预报进入国家业务（10 月 17 日起已有报道）；学科：500 hPa 集合平均有效时效 11.0 天，部分指标优于 ECMWF 数值与 AI 集合系统（研发方检验）","en":"AI: a fully domestic AI ensemble enters national operations; Field: 11.0-day skill at 500 hPa, some scores above ECMWF systems (developer evaluation)"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：完全国产数据、模型、算力训练的 AI 集合预报进入国家业务（10 月 17 日起已有报道）；学科：500 hPa 集合平均有效时效 11.0 天，部分指标优于 ECMWF 数值与 AI 集合系统（研发方检验）","en":"AI: a fully domestic AI ensemble enters national operations; Field: 11.0-day skill at 500 hPa, some scores above ECMWF systems (developer evaluation)"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：完全国产数据、模型、算力训练的 AI 集合预报进入国家业务（10 月 17 日起已有报道）；学科：500 hPa 集合平均有效时效 11.0 天，部分指标优于 ECMWF 数值与 AI 集合系统（研发方检验）","en":"AI: a fully domestic AI ensemble enters national operations; Field: 11.0-day skill at 500 hPa, some scores above ECMWF systems (developer evaluation)"}},"impactNotes":[],"verified":null,"addedIn":"v2"}],"papersRepos":[{"title":"google-deepmind/graphcast","url":"https://github.com/google-deepmind/graphcast","date":null,"kind":"repo","linkCheck":"ok"},{"title":"ecmwf/anemoi-core","url":"https://github.com/ecmwf/anemoi-core","date":null,"kind":"repo","linkCheck":"ok"},{"title":"google-research/neuralgcm","url":"https://github.com/google-research/neuralgcm","date":null,"kind":"repo","linkCheck":"ok"},{"title":"microsoft/aurora","url":"https://github.com/microsoft/aurora","date":null,"kind":"repo","linkCheck":"ok"},{"title":"Accurate medium-range global weather forecasting with 3D neural networks","url":"https://doi.org/10.1038/s41586-023-06185-3","date":"2023-07-05","kind":"paper","linkCheck":"ok-crossref"},{"title":"Learning skillful medium-range global weather forecasting","url":"https://doi.org/10.1126/science.adi2336","date":"2023-12-22","kind":"paper","linkCheck":"ok-crossref"},{"title":"Probabilistic weather forecasting with machine learning","url":"https://doi.org/10.1038/s41586-024-08252-9","date":"2024-12-04","kind":"paper","linkCheck":"ok-crossref"},{"title":"Neural general circulation models for weather and climate","url":"https://doi.org/10.1038/s41586-024-07744-y","date":"2024-07-22","kind":"paper","linkCheck":"ok-crossref"},{"title":"End-to-end data-driven weather prediction","url":"https://doi.org/10.1038/s41586-025-08897-0","date":"2025-03-20","kind":"paper","linkCheck":"ok-crossref"},{"title":"A foundation model for the Earth system","url":"https://doi.org/10.1038/s41586-025-09005-y","date":"2025-05-21","kind":"paper","linkCheck":"ok-crossref"},{"title":"Advancing subseasonal forecasting with machine learning","url":"https://doi.org/10.1038/s41586-026-11126-x","date":"2026-10-14","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"ECMWF","url":"https://www.ecmwf.int"},{"name":"NOAA","url":"https://www.noaa.gov"},{"name":"Google DeepMind","url":"https://deepmind.google"},{"name":"Microsoft Research (Aurora)","url":"https://www.microsoft.com/en-us/research/project/aurora-forecasting/"}],"news":[{"id":"9d1ab2ad9d18","date":"2026-10-20","title":"AI reveals how a warming climate will intensify NZ’s most extreme downpours","url":"https://theconversation.com/ai-reveals-how-a-warming-climate-will-intensify-nzs-most-extreme-downpours-294147","source":"The Conversation · AI","lang":"en","kind":"news","summaryZh":"The Conversation · AI于2026-10-20发布的文章，涉及AI 科研与调研、AI 气候能源，原题《AI reveals how a warming climate will intensify NZ’s most extreme downpours》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"436ff237cff2","date":"2026-10-20","title":"Interpretability of AI weather models via intermediate decoding","url":"https://arxiv.org/abs/2610.24110","source":"arXiv physics.ao-ph","lang":"en","kind":"paper","summaryZh":"arXiv physics.ao-ph于2026-10-20发布的论文，涉及AI 气候能源、AI 安全、AI 软件开发，原题《Interpretability of AI weather models via intermediate decoding》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"6ac667bf6bdb","date":"2026-10-14","title":"HiRO-ACEv2: Simulating extreme weather at 3 km resolution for a fraction of the cost","url":"https://allenai.org/blog/hiro-acev2","source":"Ai2 Blog","lang":"en","kind":"news","summaryZh":"Ai2 Blog于2026-10-14发布的文章，涉及AI 科研与调研、AI 气候能源，原题《HiRO-ACEv2: Simulating extreme weather at 3 km resolution for a fraction of the cost》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"378737ffe80d","date":"2026-10-14","title":"Researchers improve weather forecasts looking two to six weeks out","url":"https://news.mit.edu/2026/researchers-improve-weather-forecasts-looking-two-six-weeks-out-1014","source":"MIT News：Artificial intelligence","lang":"en","kind":"news","summaryZh":"MIT News：Artificial intelligence于2026-10-14发布的文章，涉及AI 科研与调研、AI 气候能源，原题《Researchers improve weather forecasts looking two to six weeks out》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"8ac7ca7542e1","date":"2026-10-14","title":"AI and climate: Hero or villain?","url":"https://www.latitudemedia.com/news/open-circuit-ai-and-climate-hero-or-villain/","source":"Latitude Media","lang":"en","kind":"news","summaryZh":"Latitude Media于2026-10-15发布的文章，涉及AI 气候能源，原题《AI and climate: Hero or villain?》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"82cfa5584ea7","date":"2026-10-12","title":"Will AI be good or bad for the climate? Here’s what the latest evidence on energy use suggests","url":"https://theconversation.com/will-ai-be-good-or-bad-for-the-climate-heres-what-the-latest-evidence-on-energy-use-suggests-293889","source":"The Conversation · AI","lang":"en","kind":"news","summaryZh":"The Conversation · AI于2026-10-12发布的文章，涉及AI 科研与调研、AI 气候能源、AI 芯片硬件，原题《Will AI be good or bad for the climate? Here’s what the latest evidence on energy use sugg》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"b2639a75c3ce","date":"2026-09-30","title":"Forecasting space weather risks on power grids","url":"https://www.microsoft.com/en-us/research/blog/forecasting-space-weather-risks-on-power-grids/","source":"Microsoft Research","lang":"en","kind":"news","summaryZh":"Microsoft Research于2026-10-01发布的文章，涉及AI 科研与调研、AI 气候能源、AI 安全，原题《Forecasting space weather risks on power grids》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"1c820979b120","date":"2026-09-09","title":"Using AI to plan for climate variability in renewable energy systems","url":"https://physicsworld.com/a/using-ai-to-plan-for-climate-variability-in-renewable-energy-systems/","source":"Physics World（IOP）","lang":"en","kind":"news","summaryZh":"Physics World（IOP）于2026-09-09发布的文章，涉及AI 科研与调研、AI 气候能源，原题《Using AI to plan for climate variability in renewable energy systems》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"d75554c78e0f","date":"2026-09-03","title":"Introducing WeatherNext 3, our most advanced and accurate global weather AI model","url":"https://deepmind.google/blog/introducing-weathernext-3-our-most-advanced-and-accurate-global-weather-ai-model/","source":"Google DeepMind Blog","lang":"en","kind":"news","summaryZh":"Google DeepMind Blog于2026-09-03发布的文章，涉及AI 气候能源，原题《Introducing WeatherNext 3, our most advanced and accurate global weather AI model》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"f5f908b6ffe4","date":"2026-08-06","title":"WeatherNext: AI model achieves breakthrough in forecasting cyclones","url":"https://deepmind.google/blog/weathernext-ai-model-achieves-breakthrough-in-forecasting-cyclones/","source":"Google DeepMind Blog","lang":"en","kind":"news","summaryZh":"Google DeepMind Blog于2026-08-06发布的文章，涉及AI 气候能源，原题《WeatherNext: AI model achieves breakthrough in forecasting cyclones》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"8088398c680e","date":"2026-05-12","title":"IFS Cycle 50r1 and AIFS v2 go live","url":"https://www.ecmwf.int/en/about/media-centre/news/2026/ifs-cycle-50r1-aifsv2-live","source":"ECMWF","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"ok"}],"review":{"lastReviewed":"2026-10-21","cycleDays":60,"nextReview":"2026-12-20"},"urls":{"zh":"/zh-hans/disciplines/weather-climate/","en":"/en/disciplines/weather-climate/"},"dimensions":{"method":{"score":4,"label":{"zh":"研究方法","en":"Research method"},"evidence":"数据驱动全球预报模型已成为主流研究方向","source":"https://www.science.org/doi/10.1126/science.adi2336","weight":0.3},"discovery":{"score":3,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"GraphCast、GenCast 等在多数指标上超过物理数值预报","source":"https://www.nature.com/articles/s41586-024-08252-9","weight":0.35},"adoption":{"score":4,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"ECMWF AIFS 2025-02 业务化、2026-05 升级 v2；NOAA 2025-12 部署 AI 全球模型","source":"https://www.ecmwf.int/en/about/media-centre/news/2026/ifs-cycle-50r1-aifsv2-live","weight":0.25},"education":{"score":1,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到课程证据","source":null,"weight":0.1}},"compositeScore":3.35,"levelCapped":false,"levelChange":{"from":3,"to":3,"direction":"same","reason":null},"biggestBreakthrough":{"date":"2023-07-05","title":{"zh":"盘古气象大模型（Nature）","en":"Pangu-Weather (Nature)"},"url":"https://www.nature.com/articles/s41586-023-06185-3","aiImpact":"leap","fieldImpact":"leap","why":{"zh":"首个在中期预报精度上超过欧洲中心业务数值模式的 AI 模型","en":"First AI model to beat ECMWF’s operational NWP on medium-range skill"}},"keyPapers":[{"title":"Accurate medium-range global weather forecasting with 3D neural networks","titleZh":"盘古气象大模型（Nature）","id":"10.1038/s41586-023-06185-3","url":"https://doi.org/10.1038/s41586-023-06185-3","venue":"Nature","date":"2023-07-05","why":{"zh":"首个在中期预报精度上超过欧洲中心业务数值模式的 AI 模型","en":"First AI model to beat ECMWF’s operational NWP on medium-range skill"},"fieldImpact":"leap","aiImpact":"leap"},{"title":"Learning skillful medium-range global weather forecasting","titleZh":"GraphCast 中期天气预报（Science）","id":"10.1126/science.adi2336","url":"https://doi.org/10.1126/science.adi2336","venue":"Science","date":"2023-12-22","why":{"zh":"90% 指标优于 HRES，1 分钟出 10 天预报","en":"Beats HRES on 90% of targets, 10-day forecast in a minute"},"fieldImpact":"leap","aiImpact":"major"},{"title":"Probabilistic weather forecasting with machine learning","titleZh":"GenCast 概率集合预报（Nature）","id":"10.1038/s41586-024-08252-9","url":"https://doi.org/10.1038/s41586-024-08252-9","venue":"Nature","date":"2024-12-04","why":{"zh":"概率集合预报在 97% 指标上超过 ECMWF ENS","en":"Beats ECMWF ENS on 97% of targets"},"fieldImpact":"leap","aiImpact":"major"},{"title":"Neural general circulation models for weather and climate","titleZh":"NeuralGCM 混合大气模型（Nature）","id":"10.1038/s41586-024-07744-y","url":"https://doi.org/10.1038/s41586-024-07744-y","venue":"Nature","date":"2024-07-22","why":{"zh":"AI 与物理混合的大气环流模型，可做气候模拟","en":"Hybrid ML–physics GCM usable for climate"},"fieldImpact":"major","aiImpact":"major"},{"title":"End-to-end data-driven weather prediction","titleZh":"Aardvark Weather：端到端数据驱动天气预报，直接从观测出发（Nature）","id":"10.1038/s41586-025-08897-0","url":"https://doi.org/10.1038/s41586-025-08897-0","venue":"Nature","date":"2025-03-20","why":{"zh":"绕过传统数据同化，首次用单一模型从原始观测出预报","en":"Bypasses data assimilation, forecasting straight from observations"},"fieldImpact":"major","aiImpact":"major"},{"title":"A foundation model for the Earth system","titleZh":"Aurora 地球系统基础模型（Nature）","id":"10.1038/s41586-025-09005-y","url":"https://doi.org/10.1038/s41586-025-09005-y","venue":"Nature","date":"2025-05-21","why":{"zh":"地球系统基础模型覆盖空气质量、海浪、热带气旋","en":"Earth-system foundation model"},"fieldImpact":"major","aiImpact":"major"},{"title":"Advancing subseasonal forecasting with machine learning","titleZh":"微软 PBC：机器学习概率偏差校正推进次季节（2–6 周）预报，使 ECMWF AIFS 技巧翻倍（Nature）","id":"10.1038/s41586-026-11126-x","url":"https://doi.org/10.1038/s41586-026-11126-x","venue":"Nature","date":"2026-10-14","why":{"zh":"AI：轻量模型学习任意预报系统的系统性误差，并在 ECMWF AI Weather Quest 中夺冠；学科：改进了 ECMWF 业务次季节预报 98% 的降水目标，填补 2–6 周预报空档","en":"AI: a light model learns any system’s biases and won ECMWF’s AI Weather Quest; Field: improves ECMWF subseasonal forecasts for 98% of precipitation targets"},"fieldImpact":"major","aiImpact":"major"}],"latest_papers":[{"id":"arXiv:2610.29630","title":"Space weather dependence of decameter-scale ionospheric irregularities spatial distribution","titleZh":"空间天气对十米级电离层不规则体空间分布的影响","summaryZh":"基于 SuperDARN 等雷达数据和空间天气参数，用自学习模式识别构建高纬电离层不规则体的现报与预报模型。","summaryEn":"Data-driven self-learning pattern recognition for nowcasting and forecasting high-latitude ionospheric irregularities.","authors":["Oleg I. Berngardt","Ivan A. Lavygin"],"date":"2026-10-21","url":"https://arxiv.org/abs/2610.29630","site":"arXiv","venue":null,"category":"physics.ao-ph","fetchedAt":"2026-10-22T10:09:25"},{"id":"arXiv:2610.28219","title":"Disentangling forced and internal components in observed climate evolution using spatiotemporal deep learning","titleZh":"用时空深度学习分离观测气候演变中的强迫与内部变率","summaryZh":"训练 ConvLSTM 区分外强迫响应与内部变率，表现优于现有方法；应用于观测发现内部变率主导了赤道太平洋类拉尼娜增暖格局等争议特征。","summaryEn":"A ConvLSTM separates forced response from internal variability, finding internal variability dominated the La Niña-like Pacific pattern.","authors":["Parvathi Kooloth","Wenyu Zhou","L. Ruby Leung","Shang-Ping Xie"],"date":"2026-10-20","url":"https://arxiv.org/abs/2610.28219","site":"arXiv","venue":null,"category":"physics.ao-ph","fetchedAt":"2026-10-22T10:09:25"},{"id":"arXiv:2610.25797","title":"AirMPA: A Meteorology-to-Pollution Adapter for Global Air Quality Forecasting","titleZh":"AirMPA：面向全球空气质量预报的气象-污染适配器","summaryZh":"提出解耦的自回归框架，先预报气象再映射到污染物，提高全球空气质量预报的灵活性。","summaryEn":"AirMPA, a decoupled meteorology-to-pollution adapter for global air-quality forecasting.","authors":["Yiheng Wang","Kai Zheng","Yuetan Lin","Fanglu Fan"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.25797","site":"arXiv","venue":null,"category":"physics.ao-ph","fetchedAt":"2026-10-22T10:09:25"},{"id":"arXiv:2610.24708","title":"Spherical Physics-informed Neural Operator with Multi-scale Coupling for Meteorological Downscaling","titleZh":"多尺度耦合的球面物理信息神经算子用于气象降尺度","summaryZh":"在球面上处理几何畸变，并引入大气物理约束和多尺度耦合，提升区域高分辨率气象降尺度。","summaryEn":"A spherical physics-informed neural operator with multi-scale coupling for meteorological downscaling.","authors":["Yiqiang Ye","Yichi Wang","Haodi Zhang","Jiawei Wen"],"date":"2026-10-17","url":"https://arxiv.org/abs/2610.24708","site":"arXiv","venue":null,"category":"cs.AI","fetchedAt":"2026-10-22T10:09:25"},{"id":"arXiv:2610.21259","title":"Energy Flow Shapes the Decomposition Dynamics of a Reactive Atmospheric Intermediate","titleZh":"能量流动塑造大气活性中间体的分解动力学","summaryZh":"在神经网络势能面上做分子动力学模拟，研究 Criegee 中间体 anti-CH3CHOO 在不同激发方式下的分解。","summaryEn":"Neural-network potential MD simulations of Criegee intermediate decomposition under different excitations.","authors":["Mitradip Das","Cangtao Yin","M. Meuwly"],"date":"2026-10-15","url":"https://arxiv.org/abs/2610.21259","site":"arXiv","venue":null,"category":"physics.chem-ph","fetchedAt":"2026-10-22T10:09:25"},{"id":"arXiv:2610.20947","title":"Ranked Right, Sized Wrong: Ensemble Spread Fails Silently at Streamflow Extremes","titleZh":"排序对、大小错：集合离散度在径流极值下悄然失效","summaryZh":"评估业务化 LSTM 径流预报的集合离散度在高流量下是否仍能反映误差，发现其低估极值误差。","summaryEn":"LSTM streamflow ensemble spread under-represents errors at extreme flows.","authors":["Justin Gallant","Tyler Wilson"],"date":"2026-10-14","url":"https://arxiv.org/abs/2610.20947","site":"arXiv","venue":null,"category":"stat.ML","fetchedAt":"2026-10-22T10:09:25"},{"id":"arXiv:2610.20405","title":"Improved Precipitation Variability in the Community Atmosphere Model through Implementation of a Neural-Network Deep Convection Scheme","titleZh":"在 CAM 大气模型中用神经网络深对流方案改善降水变率","summaryZh":"在 CAM6 中用以风暴分辨率全球模型输出训练的深度神经网络替换 Zhang-McFarlane 深对流方案，改善降水变率。","summaryEn":"A DNN trained on storm-resolving output replaces CAM6’s deep convection scheme, improving precipitation variability.","authors":["Xavier J. Levine","Griffin Mooers","Judith Berner","Jack W. Atkinson"],"date":"2026-10-14","url":"https://arxiv.org/abs/2610.20405","site":"arXiv","venue":null,"category":"physics.ao-ph","fetchedAt":"2026-10-22T10:09:25"},{"id":"arXiv:2610.19885","title":"Generative modelling for multivariate spatio-temporal downscaling of monthly climate fields","titleZh":"用生成模型对月尺度气候场做多变量时空降尺度","summaryZh":"用生成模型从粗分辨率月平均气候场联合推断高分辨率的日尺度场。","summaryEn":"Generative modelling for joint spatio-temporal downscaling of monthly climate fields.","authors":["Yu Huang","Shangshang Yang","Sebastian Bathiany","Michael Aich"],"date":"2026-10-14","url":"https://arxiv.org/abs/2610.19885","site":"arXiv","venue":null,"category":"physics.ao-ph","fetchedAt":"2026-10-22T10:09:25"}]},{"id":"earth-sciences","slug":"earth-sciences","category":"earth-env","categoryName":{"zh":"地球与环境","en":"Earth & environment"},"name":{"zh":"地球科学（地震/地质/海洋）","en":"Earth sciences (seismology, geology, oceans)"},"classification":{"oecdFos":"1.5","moe":"0708 地球物理学 / 0709 地质学 / 0707 海洋科学"},"level":2,"levelKey":"breakthrough","levelLabel":{"zh":"局部突破","en":"Local breakthroughs"},"trend":"flat","reason":{"zh":"深度学习震相拾取（PhaseNet、EQTransformer）已是构建地震目录的常用工具；2026 年 AI 可在数秒内给出余震风险，但对单次大地震的预测仍不可靠。","en":"Deep-learning phase picking (PhaseNet, EQTransformer) is standard for building catalogues; in 2026 AI gives aftershock risk in seconds, but individual large-quake prediction remains unreliable."},"reasonSources":[{"url":"https://doi.org/10.1093/gji/ggy423","linkCheck":"ok-crossref"},{"url":"https://www.nature.com/articles/s41467-020-17591-w","linkCheck":"ok"},{"url":"https://www.bgs.ac.uk/news/new-research-shows-ai-earthquake-tools-forecast-aftershock-risk-in-seconds/","linkCheck":"ok"}],"directions":{"zh":["震相拾取与地震目录","余震预测","地震动生成模拟","海洋预报"],"en":["Phase picking & catalogues","Aftershock forecasting","Generative ground motion","Ocean forecasting"]},"stats":{"milestones":8,"milestonesSince2025":5,"milestones2026":4,"milestonesLast6Months":4,"news":3,"latestProgress":"2026-10-20","latestMilestone":"2026-10-01","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":0,"leaps":{"ai":0,"field":0},"latestPapers":8,"keyPapers":6},"milestones":[{"date":"2018-08-29","title":{"zh":"深度学习预测大地震后的余震分布，优于库仑应力变化（Nature）","en":"Deep learning of aftershock patterns following large earthquakes (Nature)"},"url":"https://doi.org/10.1038/s41586-018-0438-y","source":"Nature","flags":["contested"],"flagLabels":[{"zh":"有争议","en":"Contested"}],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"引发广泛关注，但后续研究指出简单基线即可达到同等效果","en":"High-profile, but later shown matched by simple baselines"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"引发广泛关注，但后续研究指出简单基线即可达到同等效果","en":"High-profile, but later shown matched by simple baselines"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"引发广泛关注，但后续研究指出简单基线即可达到同等效果","en":"High-profile, but later shown matched by simple baselines"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-018-0438-y","title":"Deep learning of aftershock patterns following large earthquakes","venue":"Nature","date":"2018-08-29"},"addedIn":"v2"},{"date":"2018-10-13","title":{"zh":"PhaseNet 深度学习震相拾取（Geophys. J. Int.）","en":"PhaseNet deep-learning phase picking (GJI)"},"url":"https://doi.org/10.1093/gji/ggy423","source":"Geophysical Journal International","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"system","typeLabel":{"zh":"系统发布","en":"System"},"why":{"zh":"震相拾取精度超过人工规则方法，被广泛用于地震目录","en":"Picks beat rule-based methods; widely used"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"震相拾取精度超过人工规则方法，被广泛用于地震目录","en":"Picks beat rule-based methods; widely used"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"震相拾取精度超过人工规则方法，被广泛用于地震目录","en":"Picks beat rule-based methods; widely used"}},"impactNotes":[],"verified":{"id":"10.1093/gji/ggy423","title":"PhaseNet: A Deep-Neural-Network-Based Seismic Arrival Time Picking Method","venue":"Geophysical Journal International","date":"2018-10-13"},"addedIn":"v1"},{"date":"2020-08-07","title":{"zh":"EQTransformer 地震检测与震相拾取（Nat. Commun.）","en":"EQTransformer detection & picking (Nat. Commun.)"},"url":"https://www.nature.com/articles/s41467-020-17591-w","source":"Nature Communications","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"system","typeLabel":{"zh":"系统发布","en":"System"},"why":{"zh":"同时检测地震和拾取震相，成为标准工具","en":"Joint detection and picking; standard tool"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"同时检测地震和拾取震相，成为标准工具","en":"Joint detection and picking; standard tool"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"同时检测地震和拾取震相，成为标准工具","en":"Joint detection and picking; standard tool"}},"impactNotes":[],"verified":{"id":"10.1038/s41467-020-17591-w","title":"Earthquake transformer—an attentive deep-learning model for simultaneous earthquake detection and phase picking","venue":"Nature Communications","date":"2020-08-07"},"addedIn":"v1"},{"date":"2025-11-25","title":{"zh":"BGS：AI 工具数秒内预测余震风险","en":"BGS: AI tools forecast aftershock risk in seconds"},"url":"https://www.bgs.ac.uk/news/new-research-shows-ai-earthquake-tools-forecast-aftershock-risk-in-seconds/","source":"British Geological Survey","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"ML 余震预测与 ETAS 相当但数秒可得","en":"Comparable to ETAS, in seconds"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"ML 余震预测与 ETAS 相当但数秒可得","en":"Comparable to ETAS, in seconds"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"ML 余震预测与 ETAS 相当但数秒可得","en":"Comparable to ETAS, in seconds"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-05-28","title":{"zh":"深度学习在南极东部之下识别出约千次板内中源地震，部分深达 70–150 千米，以岩石圈弯曲应力解释其成因（Science）","en":"Deep learning reveals upper-mantle intraplate earthquakes beneath East Antarctica (Science)"},"url":"https://doi.org/10.1126/science.aea9895","source":"Science","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"discovery","typeLabel":{"zh":"科学发现","en":"Discovery"},"why":{"zh":"AI：迁移增强的深度学习从区域台网数据中检出传统方法漏掉的微震；学科：板内中源地震难以用经典板块构造解释，结果支持岩石圈–软流圈相互作用模型，并提示此类地震可能比已知更普遍","en":"AI: transfer-enhanced deep learning detects events missed by conventional methods. Field: intraplate intermediate-depth quakes support lithosphere–asthenosphere models and may be more common than recognised"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：迁移增强的深度学习从区域台网数据中检出传统方法漏掉的微震；学科：板内中源地震难以用经典板块构造解释，结果支持岩石圈–软流圈相互作用模型，并提示此类地震可能比已知更普遍","en":"AI: transfer-enhanced deep learning detects events missed by conventional methods. Field: intraplate intermediate-depth quakes support lithosphere–asthenosphere models and may be more common than recognised"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：迁移增强的深度学习从区域台网数据中检出传统方法漏掉的微震；学科：板内中源地震难以用经典板块构造解释，结果支持岩石圈–软流圈相互作用模型，并提示此类地震可能比已知更普遍","en":"AI: transfer-enhanced deep learning detects events missed by conventional methods. Field: intraplate intermediate-depth quakes support lithosphere–asthenosphere models and may be more common than recognised"}},"impactNotes":[],"verified":{"id":"10.1126/science.aea9895","title":"Upper-mantle earthquakes beneath East Antarctica","venue":"Science","date":"2026-05-28"},"addedIn":"v2-20261022pm"},{"date":"2026-07-27","title":{"zh":"QuakeGen：条件生成模型做余震预测，空间预测优于 USGS 业务模型（arXiv 2607.24109）","en":"QuakeGen: conditional generative aftershock forecasting beats USGS operational spatial forecasts"},"url":"https://arxiv.org/abs/2607.24109","source":"arXiv","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"在 2024–25 事件上空间预测优于业务模型，区域日预报与 ETAS 相当","en":"Beats operational spatial forecasts on 2024–25 events"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"在 2024–25 事件上空间预测优于业务模型，区域日预报与 ETAS 相当","en":"Beats operational spatial forecasts on 2024–25 events"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"在 2024–25 事件上空间预测优于业务模型，区域日预报与 ETAS 相当","en":"Beats operational spatial forecasts on 2024–25 events"}},"impactNotes":[],"verified":{"id":"arXiv:2607.24109","title":"Earthquake Aftershock Forecasting using Conditional Generative Models","venue":"arXiv","date":"2026-07-27"},"addedIn":"v2"},{"date":"2026-08-12","title":{"zh":"Berkeley Lab CGM-GM：条件生成模型模拟宽频地震动","en":"Berkeley Lab CGM-GM generative ground-motion simulation"},"url":"https://cs.lbl.gov/news-and-events/news/2026/cgm-gm/","source":"Berkeley Lab","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"生成模型模拟宽频地震动","en":"Generative ground-motion simulation"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"生成模型模拟宽频地震动","en":"Generative ground-motion simulation"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"生成模型模拟宽频地震动","en":"Generative ground-motion simulation"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-10-01","title":{"zh":"机器学习全球海山普查：识别的海山数量是既有资料的 1.8–2.7 倍，其中高出周围海底 1 千米以上的达 28,434 座（EPSL）","en":"Global seamount detection using machine learning: 1.8–2.7× more seamounts than prior ground truth, 28,434 rising ≥1 km (EPSL)"},"url":"https://doi.org/10.1016/j.epsl.2026.120368","source":"Earth and Planetary Science Letters","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"AI：以船测与重力梯度识别的海山为标注、GEBCO 2025 网格为输入训练模型并后处理；学科：更新全球海山普查及高度、面积、体积估计，关系到海洋环流、生态与板内火山研究（EPSL 2026 年 12 月卷，10-01 在线）","en":"AI: trained on ship/gravity-derived seamounts with GEBCO 2025 input. Field: updates the global seamount census with height/area/volume, relevant to ocean circulation, ecology and intraplate volcanism"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：以船测与重力梯度识别的海山为标注、GEBCO 2025 网格为输入训练模型并后处理；学科：更新全球海山普查及高度、面积、体积估计，关系到海洋环流、生态与板内火山研究（EPSL 2026 年 12 月卷，10-01 在线）","en":"AI: trained on ship/gravity-derived seamounts with GEBCO 2025 input. Field: updates the global seamount census with height/area/volume, relevant to ocean circulation, ecology and intraplate volcanism"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：以船测与重力梯度识别的海山为标注、GEBCO 2025 网格为输入训练模型并后处理；学科：更新全球海山普查及高度、面积、体积估计，关系到海洋环流、生态与板内火山研究（EPSL 2026 年 12 月卷，10-01 在线）","en":"AI: trained on ship/gravity-derived seamounts with GEBCO 2025 input. Field: updates the global seamount census with height/area/volume, relevant to ocean circulation, ecology and intraplate volcanism"}},"impactNotes":[],"verified":{"id":"10.1016/j.epsl.2026.120368","title":"Global seamount detection using machine learning","venue":"Earth and Planetary Science Letters","date":"2026-10-01"},"addedIn":"v2-20261022pm"}],"papersRepos":[{"title":"AI4EPS/PhaseNet","url":"https://github.com/AI4EPS/PhaseNet","date":null,"kind":"repo","linkCheck":"ok"},{"title":"smousavi05/EQTransformer","url":"https://github.com/smousavi05/EQTransformer","date":null,"kind":"repo","linkCheck":"ok"},{"title":"Deep learning of aftershock patterns following large earthquakes","url":"https://doi.org/10.1038/s41586-018-0438-y","date":"2018-08-29","kind":"paper","linkCheck":"ok-crossref"},{"title":"PhaseNet: A Deep-Neural-Network-Based Seismic Arrival Time Picking Method","url":"https://doi.org/10.1093/gji/ggy423","date":"2018-10-13","kind":"paper","linkCheck":"ok-crossref"},{"title":"Earthquake transformer—an attentive deep-learning model for simultaneous earthquake detection and phase picking","url":"https://doi.org/10.1038/s41467-020-17591-w","date":"2020-08-07","kind":"paper","linkCheck":"ok-crossref"},{"title":"Earthquake Aftershock Forecasting using Conditional Generative Models","url":"https://arxiv.org/abs/2607.24109","date":"2026-07-27","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"USGS","url":"https://www.usgs.gov"},{"name":"British Geological Survey","url":"https://www.bgs.ac.uk"},{"name":"Berkeley Lab","url":"https://cs.lbl.gov"}],"news":[{"id":"61140efd018b","date":"2026-10-20","title":"A hybrid data-centric machine learning for lost circulation prediction using imbalanced drilling and geological datasets","url":"https://www.nature.com/articles/s41598-026-75043-9","source":"Nature 学科·数学","lang":"en","kind":"news","summaryZh":"Nature 学科·数学于2026-10-20发布的文章，涉及AI 数学，原题《A hybrid data-centric machine learning for lost circulation prediction using imbalanced dr》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"6a9f3e662d1f","date":"2026-10-20","title":"BridgeCast: Bridging Ocean Wave Forecasts to Reanalysis via Flow Matching with Exogenous Variables","url":"https://arxiv.org/abs/2610.03759","source":"arXiv physics.ao-ph","lang":"en","kind":"paper","summaryZh":"arXiv physics.ao-ph于2026-10-20发布的论文，涉及AI 气候能源、AI 安全、AI 软件开发，原题《BridgeCast: Bridging Ocean Wave Forecasts to Reanalysis via Flow Matching with Exogenous V》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"3ec0a51fe38a","date":"2026-10","title":"New forecasting research could help millions living in earthquake-prone regions","url":"https://www.bristol.ac.uk/news/2026/october/new-forecasting-research-could-help-millions-living-in-earthquake-prone-regions-.html","source":"University of Bristol","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"ok"}],"review":{"lastReviewed":"2026-10-22","cycleDays":90,"nextReview":"2027-01-19"},"urls":{"zh":"/zh-hans/disciplines/earth-sciences/","en":"/en/disciplines/earth-sciences/"},"dimensions":{"method":{"score":3,"label":{"zh":"研究方法","en":"Research method"},"evidence":"深度学习震相拾取已是构建地震目录的常规方法","source":"https://doi.org/10.1093/gji/ggy423","weight":0.3},"discovery":{"score":2,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"ML 地震目录揭示更多小震和断层结构；余震预测与传统模型相当","source":"https://www.bgs.ac.uk/news/new-research-shows-ai-earthquake-tools-forecast-aftershock-risk-in-seconds/","weight":0.35},"adoption":{"score":2,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"研究机构普遍使用，业务预警仍以传统方法为主","source":null,"weight":0.25},"education":{"score":1,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到课程证据","source":null,"weight":0.1}},"compositeScore":2.2,"levelCapped":false,"levelChange":{"from":2,"to":2,"direction":"same","reason":null},"biggestBreakthrough":{"date":"2020-08-07","title":{"zh":"EQTransformer 地震检测与震相拾取（Nat. Commun.）","en":"EQTransformer detection & picking (Nat. Commun.)"},"url":"https://www.nature.com/articles/s41467-020-17591-w","aiImpact":"major","fieldImpact":"major","why":{"zh":"同时检测地震和拾取震相，成为标准工具","en":"Joint detection and picking; standard tool"}},"keyPapers":[{"title":"Deep learning of aftershock patterns following large earthquakes","titleZh":"深度学习预测大地震后的余震分布，优于库仑应力变化（Nature）","id":"10.1038/s41586-018-0438-y","url":"https://doi.org/10.1038/s41586-018-0438-y","venue":"Nature","date":"2018-08-29","why":{"zh":"引发广泛关注，但后续研究指出简单基线即可达到同等效果","en":"High-profile, but later shown matched by simple baselines"},"fieldImpact":"incremental","aiImpact":"major"},{"title":"PhaseNet: A Deep-Neural-Network-Based Seismic Arrival Time Picking Method","titleZh":"PhaseNet 深度学习震相拾取（Geophys. J. Int.）","id":"10.1093/gji/ggy423","url":"https://doi.org/10.1093/gji/ggy423","venue":"Geophysical Journal International","date":"2018-10-13","why":{"zh":"震相拾取精度超过人工规则方法，被广泛用于地震目录","en":"Picks beat rule-based methods; widely used"},"fieldImpact":"major","aiImpact":"major"},{"title":"Earthquake transformer—an attentive deep-learning model for simultaneous earthquake detection and phase picking","titleZh":"EQTransformer 地震检测与震相拾取（Nat. Commun.）","id":"10.1038/s41467-020-17591-w","url":"https://doi.org/10.1038/s41467-020-17591-w","venue":"Nature Communications","date":"2020-08-07","why":{"zh":"同时检测地震和拾取震相，成为标准工具","en":"Joint detection and picking; standard tool"},"fieldImpact":"major","aiImpact":"major"},{"title":"Upper-mantle earthquakes beneath East Antarctica","titleZh":"深度学习在南极东部之下识别出约千次板内中源地震，部分深达 70–150 千米，以岩石圈弯曲应力解释其成因（Science）","id":"10.1126/science.aea9895","url":"https://doi.org/10.1126/science.aea9895","venue":"Science","date":"2026-05-28","why":{"zh":"AI：迁移增强的深度学习从区域台网数据中检出传统方法漏掉的微震；学科：板内中源地震难以用经典板块构造解释，结果支持岩石圈–软流圈相互作用模型，并提示此类地震可能比已知更普遍","en":"AI: transfer-enhanced deep learning detects events missed by conventional methods. Field: intraplate intermediate-depth quakes support lithosphere–asthenosphere models and may be more common than recognised"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Earthquake Aftershock Forecasting using Conditional Generative Models","titleZh":"QuakeGen：条件生成模型做余震预测，空间预测优于 USGS 业务模型（arXiv 2607.24109）","id":"arXiv:2607.24109","url":"https://arxiv.org/abs/2607.24109","venue":"arXiv","date":"2026-07-27","why":{"zh":"在 2024–25 事件上空间预测优于业务模型，区域日预报与 ETAS 相当","en":"Beats operational spatial forecasts on 2024–25 events"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Global seamount detection using machine learning","titleZh":"机器学习全球海山普查：识别的海山数量是既有资料的 1.8–2.7 倍，其中高出周围海底 1 千米以上的达 28,434 座（EPSL）","id":"10.1016/j.epsl.2026.120368","url":"https://doi.org/10.1016/j.epsl.2026.120368","venue":"Earth and Planetary Science Letters","date":"2026-10-01","why":{"zh":"AI：以船测与重力梯度识别的海山为标注、GEBCO 2025 网格为输入训练模型并后处理；学科：更新全球海山普查及高度、面积、体积估计，关系到海洋环流、生态与板内火山研究（EPSL 2026 年 12 月卷，10-01 在线）","en":"AI: trained on ship/gravity-derived seamounts with GEBCO 2025 input. Field: updates the global seamount census with height/area/volume, relevant to ocean circulation, ecology and intraplate volcanism"},"fieldImpact":"major","aiImpact":"incremental"}],"latest_papers":[{"id":"arXiv:2610.19147","title":"Physics-based deep learning for non-stationary adaptive multiple suppression","titleZh":"基于物理的深度学习实现非平稳自适应多次波压制","summaryZh":"用纯数据驱动、基于物理的深度学习算法改进海洋地震勘探中需要主观人工控制的多次波自适应相减步骤。","summaryEn":"Physics-based deep learning improves adaptive subtraction of surface-related multiples in marine seismic processing.","authors":["Chao Li","Yangkang Chen"],"date":"2026-10-13","url":"https://arxiv.org/abs/2610.19147","site":"arXiv","venue":null,"category":"physics.geo-ph","fetchedAt":"2026-10-22T10:09:29"},{"id":"arXiv:2610.17404","title":"Event-Centric Earthquake Monitoring with a Multi-Station Transformer","titleZh":"基于多台站 Transformer 的以事件为中心的地震监测","summaryZh":"提出 EQNet2，一个考虑台网几何的多台站 Transformer，同时预测震相到时、初动极性、震源位置、震级和事件关联。","summaryEn":"EQNet2, a multi-station transformer, jointly predicts arrivals, polarities, location, magnitude and event association.","authors":["Haoyu Wang","Weiqiang Zhu"],"date":"2026-10-12","url":"https://arxiv.org/abs/2610.17404","site":"arXiv","venue":null,"category":"physics.geo-ph","fetchedAt":"2026-10-22T10:09:29"},{"id":"arXiv:2610.10788","title":"Fault-conditioned Seismic Image Generation using Denoising Diffusion Probabilistic Modeling and Neural Style Transfer","titleZh":"用去噪扩散模型与神经风格迁移生成断层条件地震图像","summaryZh":"直接从野外数据学习地震图像分布，生成带断层的逼真样本，用于训练断层识别深度学习模型。","summaryEn":"A DDPM framework generates realistic fault-conditioned seismic images to augment training data.","authors":["Tolulope Agbaje","Nam Pham","Altay Sansal","Yangkang Chen"],"date":"2026-10-07","url":"https://arxiv.org/abs/2610.10788","site":"arXiv","venue":null,"category":"physics.geo-ph","fetchedAt":"2026-10-22T10:09:29"},{"id":"arXiv:2610.09871","title":"An AI-assisted conditioning and geological interpretation workflow for usage in implicit geological modeling","titleZh":"面向隐式地质建模的 AI 辅助数据调理与地质解释流程","summaryZh":"在欧盟 Horizon 项目中实现机器学习工具包，加速 300–3500 米深度陆上地震资料解释，为隐式地质建模提供输入。","summaryEn":"An ML toolkit speeds interpretation of onshore seismic data (300–3,500 m) for implicit geological modelling.","authors":["Stefan Carpentier","Jan Diederik van Wees","Eva de Boever","Jan Niederau"],"date":"2026-10-07","url":"https://arxiv.org/abs/2610.09871","site":"arXiv","venue":null,"category":"physics.geo-ph","fetchedAt":"2026-10-22T10:09:29"},{"id":"arXiv:2610.09663","title":"Real-world application of deep learning in large-scale seismic interference attenuation: A case study in the Camie field of Angola","titleZh":"深度学习在大规模地震干扰压制中的实际应用：安哥拉 Camie 油田案例","summaryZh":"把基于深度神经网络的地震干扰压制流程应用于安哥拉 345 平方公里以上的海上地震区块。","summaryEn":"Field case applying a DNN workflow to attenuate seismic interference over 345+ km² offshore Angola.","authors":["Jing Sun","Song Hou"],"date":"2026-10-07","url":"https://arxiv.org/abs/2610.09663","site":"arXiv","venue":null,"category":"physics.geo-ph","fetchedAt":"2026-10-22T10:09:29"},{"id":"arXiv:2610.02337","title":"Multi-Modal Surface Wave Inversion using Physics Guided Variational Autoencoders","titleZh":"用物理引导变分自编码器做多模态面波反演","summaryZh":"提出物理引导的多模态变分自编码器，处理反演非唯一性，实现可扩展的近地表速度结构刻画。","summaryEn":"A physics-guided multimodal VAE for scalable near-surface surface-wave inversion.","authors":["Sujith Swaminadhan","Andrey Bakulin","Sergey Fomel","Yangkang Chen"],"date":"2026-10-01","url":"https://arxiv.org/abs/2610.02337","site":"arXiv","venue":null,"category":"physics.geo-ph","fetchedAt":"2026-10-22T10:09:29"},{"id":"arXiv:2610.01033","title":"GeoFWI3D: Large-scale 3D Velocity Model Dataset for Deep Learning-assisted Seismic Imaging","titleZh":"GeoFWI3D：深度学习辅助地震成像的大规模三维速度模型数据集","summaryZh":"开源大量地质上合理的三维地下模型，用于训练和评测深度学习辅助的地震成像与全波形反演。","summaryEn":"An open benchmark of geologically plausible 3D velocity models for deep-learning-assisted seismic imaging and FWI.","authors":["Sujith Swaminadhan","Yiran Shen","Chao Li","Kai Gao"],"date":"2026-10-01","url":"https://arxiv.org/abs/2610.01033","site":"arXiv","venue":null,"category":"physics.geo-ph","fetchedAt":"2026-10-22T10:09:29"},{"id":"arXiv:2609.39423","title":"Elasticity of polycrystalline davemaoite constrains its grain size in the lower mantle","titleZh":"多晶钙钙钛矿弹性约束其在下地幔中的晶粒尺寸","summaryZh":"用百万原子机器学习分子动力学模拟，解释钙钙钛矿剪切模量实验与理论的长期差异，并约束其在下地幔的晶粒尺寸。","summaryEn":"Million-atom machine-learning MD reconciles davemaoite’s shear modulus and constrains its grain size in the lower mantle.","authors":["Peiyu Zhang","Liang Yuan","Rong Huang","Yu Ye"],"date":"2026-09-30","url":"https://arxiv.org/abs/2609.39423","site":"arXiv","venue":null,"category":"physics.chem-ph","fetchedAt":"2026-10-22T10:09:29"}]},{"id":"geography-remote-sensing","slug":"geography-remote-sensing","category":"earth-env","categoryName":{"zh":"地球与环境","en":"Earth & environment"},"name":{"zh":"地理与遥感","en":"Geography & remote sensing"},"classification":{"oecdFos":"1.5","moe":"0705 地理学 / 0816 测绘科学与技术"},"level":2,"levelKey":"breakthrough","levelLabel":{"zh":"局部突破","en":"Local breakthroughs"},"trend":"up","reason":{"zh":"地理空间基础模型开始替代逐任务建模：NASA 与 IBM 开源 Prithvi，Google DeepMind 发布 AlphaEarth Foundations 全球地表嵌入。","en":"Geospatial foundation models are replacing per-task models: NASA/IBM open-sourced Prithvi; DeepMind released AlphaEarth Foundations global embeddings."},"reasonSources":[{"url":"https://huggingface.co/ibm-nasa-geospatial","linkCheck":"ok"},{"url":"https://deepmind.google/discover/blog/alphaearth-foundations-helps-map-our-planet-in-unprecedented-detail/","linkCheck":"ok"}],"directions":{"zh":["地理空间基础模型","土地覆盖与变化检测","灾害遥感"],"en":["Geospatial foundation models","Land-cover change","Disaster remote sensing"]},"stats":{"milestones":10,"milestonesSince2025":5,"milestones2026":3,"milestonesLast6Months":3,"news":5,"latestProgress":"2026-10-20","latestMilestone":"2026-10-20","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":1,"leaps":{"ai":0,"field":0},"latestPapers":8,"keyPapers":4},"milestones":[{"date":"2021-07-26","title":{"zh":"Google Open Buildings：用深度学习从卫星影像检测非洲大陆级建筑轮廓（arXiv 2107.12283）","en":"Open Buildings: continental-scale building detection from satellite imagery (arXiv 2107.12283)"},"url":"https://arxiv.org/abs/2107.12283","source":"arXiv","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok-crossref","datePrecision":"day","type":"dataset","typeLabel":{"zh":"数据集","en":"Dataset"},"why":{"zh":"为缺乏测绘的地区提供建筑底图，被人道与规划机构广泛使用","en":"Building maps for under-mapped regions"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"为缺乏测绘的地区提供建筑底图，被人道与规划机构广泛使用","en":"Building maps for under-mapped regions"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"为缺乏测绘的地区提供建筑底图，被人道与规划机构广泛使用","en":"Building maps for under-mapped regions"}},"impactNotes":[],"verified":{"id":"arXiv:2107.12283","title":"Continental-Scale Building Detection from High Resolution Satellite Imagery","venue":"arXiv","date":"2021-07-26"},"addedIn":"v2"},{"date":"2022-06-09","title":{"zh":"Google/WRI Dynamic World：深度学习生成近实时全球 10 米土地覆盖（Scientific Data）","en":"Dynamic World: near-real-time global 10 m land cover with deep learning (Scientific Data)"},"url":"https://doi.org/10.1038/s41597-022-01307-4","source":"Scientific Data","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"dataset","typeLabel":{"zh":"数据集","en":"Dataset"},"why":{"zh":"土地覆盖制图从年度产品变为随卫星过境更新","en":"Land-cover mapping moves from annual to per-overpass"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"土地覆盖制图从年度产品变为随卫星过境更新","en":"Land-cover mapping moves from annual to per-overpass"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"土地覆盖制图从年度产品变为随卫星过境更新","en":"Land-cover mapping moves from annual to per-overpass"}},"impactNotes":[],"verified":{"id":"10.1038/s41597-022-01307-4","title":"Dynamic World, Near real-time global 10 m land use land cover mapping","venue":"Scientific Data","date":"2022-06-09"},"addedIn":"v2"},{"date":"2022-07-17","title":{"zh":"SatMAE：多光谱、多时相卫星影像的自监督预训练（NeurIPS 2022）","en":"SatMAE: pre-training transformers for temporal and multi-spectral satellite imagery"},"url":"https://arxiv.org/abs/2207.08051","source":"arXiv","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"遥感基础模型路线的早期代表","en":"Early remote-sensing foundation model"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"遥感基础模型路线的早期代表","en":"Early remote-sensing foundation model"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"遥感基础模型路线的早期代表","en":"Early remote-sensing foundation model"}},"impactNotes":[],"verified":{"id":"arXiv:2207.08051","title":"SatMAE: Pre-training Transformers for Temporal and Multi-Spectral Satellite Imagery","venue":"arXiv","date":"2022-07-17"},"addedIn":"v2"},{"date":"2023-08-03","title":{"zh":"NASA 与 IBM 开源 Prithvi 地理空间基础模型","en":"NASA & IBM open-source Prithvi geospatial model"},"url":"https://huggingface.co/ibm-nasa-geospatial","source":"Hugging Face","flags":["tbv"],"flagLabels":[{"zh":"日期或细节待核实","en":"Date/details to be verified"}],"linkCheck":"ok","datePrecision":"day","type":"system","typeLabel":{"zh":"系统发布","en":"System"},"why":{"zh":"NASA 与 IBM 开源最大地理空间基础模型（2023-08-03 官方新闻稿已确认）","en":"Largest open geospatial foundation model (date confirmed by IBM newsroom)"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"NASA 与 IBM 开源最大地理空间基础模型（2023-08-03 官方新闻稿已确认）","en":"Largest open geospatial foundation model (date confirmed by IBM newsroom)"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"NASA 与 IBM 开源最大地理空间基础模型（2023-08-03 官方新闻稿已确认）","en":"Largest open geospatial foundation model (date confirmed by IBM newsroom)"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2024-12-04","title":{"zh":"IBM 与 NASA 发布全球版 Prithvi-EO-2.0（6 亿参数）","en":"IBM and NASA release global Prithvi-EO-2.0 (600M parameters)"},"url":"https://research.ibm.com/blog/prithvi2-geospatial","source":"IBM Research","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"system","typeLabel":{"zh":"系统发布","en":"System"},"why":{"zh":"开源地理空间基础模型扩展到全球数据","en":"Open geospatial model goes global"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"开源地理空间基础模型扩展到全球数据","en":"Open geospatial model goes global"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"开源地理空间基础模型扩展到全球数据","en":"Open geospatial model goes global"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2025-04-15","title":{"zh":"ESA 与 IBM 发布 TerraMind：融合 9 种模态的开源地球观测生成式基础模型（ICCV 2025）","en":"ESA & IBM TerraMind: open any-to-any generative EO foundation model (ICCV 2025)"},"url":"https://arxiv.org/abs/2504.11171","source":"arXiv / ESA","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok-crossref","datePrecision":"day","type":"model","typeLabel":{"zh":"模型发布","en":"Model"},"why":{"zh":"遥感基础模型从单模态走向多模态生成","en":"EO foundation models go multimodal-generative"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"遥感基础模型从单模态走向多模态生成","en":"EO foundation models go multimodal-generative"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"遥感基础模型从单模态走向多模态生成","en":"EO foundation models go multimodal-generative"}},"impactNotes":[],"verified":{"id":"arXiv:2504.11171","title":"TerraMind: Large-Scale Generative Multimodality for Earth Observation","venue":"arXiv","date":"2025-04-15"},"addedIn":"v2"},{"date":"2025-07-30","title":{"zh":"AlphaEarth Foundations 全球地表嵌入","en":"AlphaEarth Foundations global embeddings"},"url":"https://deepmind.google/discover/blog/alphaearth-foundations-helps-map-our-planet-in-unprecedented-detail/","source":"Google DeepMind","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"dataset","typeLabel":{"zh":"数据集","en":"Dataset"},"why":{"zh":"10 米分辨率全球年度地表嵌入，误差比其他方法低约 24%","en":"10 m global annual embeddings"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"10 米分辨率全球年度地表嵌入，误差比其他方法低约 24%","en":"10 m global annual embeddings"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"10 米分辨率全球年度地表嵌入，误差比其他方法低约 24%","en":"10 m global annual embeddings"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-05-07","title":{"zh":"NASA：Prithvi 成为首个在轨运行的 AI 地理空间基础模型，在卫星上演示洪水与云检测","en":"NASA: Prithvi becomes the first AI geospatial foundation model in orbit"},"url":"https://science.nasa.gov/science-research/ai-foundation-model-in-orbit/","source":"NASA Science","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"deployment","typeLabel":{"zh":"落地部署","en":"Deployment"},"why":{"zh":"遥感基础模型从地面处理走向星上实时推理","en":"EO foundation models move to on-board inference"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"遥感基础模型从地面处理走向星上实时推理","en":"EO foundation models move to on-board inference"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"遥感基础模型从地面处理走向星上实时推理","en":"EO foundation models move to on-board inference"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-06-04","title":{"zh":"剑桥大学 Tessera 开源地球观测基础模型经 CVPR 2026 同行评审发表，ESA 宣布广泛开放 2017–2025 年全球 10 米嵌入","en":"Tessera open EO foundation model peer-reviewed at CVPR 2026; ESA announces global 10 m embeddings 2017–2025"},"url":"https://www.esa.int/Applications/Observing_the_Earth/Copernicus/Tessera_AI_model_offers_accessible_way_to_view_Earth","source":"ESA","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"model","typeLabel":{"zh":"模型发布","en":"Model"},"why":{"zh":"AI：融合 Sentinel-1 雷达与 Sentinel-2 光学时序的开源嵌入，可替代封闭的 AlphaEarth；学科：非 AI 专家用笔记本即可做全球尺度制图，已用于英国保护地生境监测","en":"AI: open fused SAR/optical time-series embeddings, an open alternative to AlphaEarth; Field: global-scale mapping on a laptop, already used for UK habitat monitoring"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：融合 Sentinel-1 雷达与 Sentinel-2 光学时序的开源嵌入，可替代封闭的 AlphaEarth；学科：非 AI 专家用笔记本即可做全球尺度制图，已用于英国保护地生境监测","en":"AI: open fused SAR/optical time-series embeddings, an open alternative to AlphaEarth; Field: global-scale mapping on a laptop, already used for UK habitat monitoring"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：融合 Sentinel-1 雷达与 Sentinel-2 光学时序的开源嵌入，可替代封闭的 AlphaEarth；学科：非 AI 专家用笔记本即可做全球尺度制图，已用于英国保护地生境监测","en":"AI: open fused SAR/optical time-series embeddings, an open alternative to AlphaEarth; Field: global-scale mapping on a laptop, already used for UK habitat monitoring"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-10-20","title":{"zh":"KAIST 发布 GeoCore-9B：90 亿参数、位置与尺度感知的卫星影像生成基础模型（NeurIPS 2026 录用）","en":"KAIST GeoCore-9B: 9B-parameter geo-aware generative EO foundation model (NeurIPS 2026)"},"url":"https://www.eurekalert.org/news-releases/1148233","source":"KAIST / EurekAlert!","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok","datePrecision":"day","type":"model","typeLabel":{"zh":"模型发布","en":"Model"},"why":{"zh":"AI：在 1000 万对卫星图文上从头训练，引入经纬度与地面分辨率条件；学科：去云和雷达转光学任务超过专用模型（arXiv 2608.01896）","en":"AI: trained from scratch on 10M satellite image–text pairs with location and GSD conditioning; Field: beats task-specific models on cloud removal and SAR-to-optical"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：在 1000 万对卫星图文上从头训练，引入经纬度与地面分辨率条件；学科：去云和雷达转光学任务超过专用模型（arXiv 2608.01896）","en":"AI: trained from scratch on 10M satellite image–text pairs with location and GSD conditioning; Field: beats task-specific models on cloud removal and SAR-to-optical"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：在 1000 万对卫星图文上从头训练，引入经纬度与地面分辨率条件；学科：去云和雷达转光学任务超过专用模型（arXiv 2608.01896）","en":"AI: trained from scratch on 10M satellite image–text pairs with location and GSD conditioning; Field: beats task-specific models on cloud removal and SAR-to-optical"}},"impactNotes":[],"verified":null,"addedIn":"v2"}],"papersRepos":[{"title":"ibm-nasa-geospatial (Hugging Face)","url":"https://huggingface.co/ibm-nasa-geospatial","date":null,"kind":"repo","linkCheck":"ok"},{"title":"Continental-Scale Building Detection from High Resolution Satellite Imagery","url":"https://arxiv.org/abs/2107.12283","date":"2021-07-26","kind":"paper","linkCheck":"ok-crossref"},{"title":"Dynamic World, Near real-time global 10 m land use land cover mapping","url":"https://doi.org/10.1038/s41597-022-01307-4","date":"2022-06-09","kind":"paper","linkCheck":"ok-crossref"},{"title":"SatMAE: Pre-training Transformers for Temporal and Multi-Spectral Satellite Imagery","url":"https://arxiv.org/abs/2207.08051","date":"2022-07-17","kind":"paper","linkCheck":"ok-crossref"},{"title":"TerraMind: Large-Scale Generative Multimodality for Earth Observation","url":"https://arxiv.org/abs/2504.11171","date":"2025-04-15","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"NASA Earthdata","url":"https://www.earthdata.nasa.gov"},{"name":"IBM Research","url":"https://research.ibm.com"},{"name":"Google Earth Engine","url":"https://earthengine.google.com"}],"news":[{"id":"bc8d06e34fc5","date":"2026-10-19","title":"AI that reconstructs cloud-covered ground in satellite images with photo-like clarity","url":"https://techxplore.com/news/2026-10-ai-reconstructs-cloud-ground-satellite.html","source":"Phys.org 机器学习与AI","lang":"en","kind":"news","summaryZh":"Phys.org 机器学习与AI于2026-10-19发布的文章，涉及AI 科研与调研，原题《AI that reconstructs cloud-covered ground in satellite images with photo-like clarity》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"09a39f2629eb","date":"2026-10-14","title":"Putting Earth AI to work: Geospatial agents for our planet's biggest challenges","url":"https://research.google/blog/putting-earth-ai-to-work-geospatial-agents-for-our-planets-biggest-challenges/","source":"Google Research Blog","lang":"en","kind":"news","summaryZh":"Google Research Blog于2026-10-15发布的文章，涉及AI 科研与调研，原题《Putting Earth AI to work: Geospatial agents for our planet's biggest challenges》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"be28832d1f97","date":"2026-10-07","title":"New AI method uses engineering knowledge to estimate disaster damage from incomplete satellite imagery","url":"https://techxplore.com/news/2026-10-ai-method-knowledge-disaster-incomplete.html","source":"Tech Xplore","lang":"en","kind":"news","summaryZh":"新方法把工程知识融入 AI，在卫星影像被云烟遮挡时也能快速估算城市尺度的灾后建筑损伤。","via":"research-samples-20261010","linkCheck":"from-feed"},{"id":"2e0aae10ded8","date":"2026-10-06","title":"Unlocking Earth AI’s planetary geospatial foundation models for global public health","url":"https://research.google/blog/earth-ais-planetary-geospatial-foundation-models-for-global-public-health/","source":"Google Research Blog","lang":"en","kind":"news","summaryZh":"谷歌用 Earth AI 人口动态基础模型（PDFM）弥补流行病学数据的缺口和滞后，发布五个合作案例。","via":"research-samples-20261010","linkCheck":"from-feed"},{"id":"801593cdd075","date":"2026-07-28","title":"The OlmoEarth Platform: Geospatial inference at planetary scale","url":"https://allenai.org/blog/olmoearth-infrastructure","source":"Ai2 Blog","lang":"en","kind":"news","summaryZh":"Ai2 Blog于2026-07-28发布的文章，涉及AI 科研与调研，原题《The OlmoEarth Platform: Geospatial inference at planetary scale》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"}],"review":{"lastReviewed":"2026-10-21","cycleDays":90,"nextReview":"2027-01-19"},"urls":{"zh":"/zh-hans/disciplines/geography-remote-sensing/","en":"/en/disciplines/geography-remote-sensing/"},"dimensions":{"method":{"score":3,"label":{"zh":"研究方法","en":"Research method"},"evidence":"地理空间基础模型开始替代逐任务建模","source":"https://arxiv.org/abs/2310.18660","weight":0.3},"discovery":{"score":1,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"主要是制图效率提升","source":null,"weight":0.35},"adoption":{"score":3,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"AlphaEarth 嵌入数据集进入 Google Earth Engine；Prithvi 开源","source":"https://deepmind.google/discover/blog/alphaearth-foundations-helps-map-our-planet-in-unprecedented-detail/","weight":0.25},"education":{"score":1,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到课程证据","source":null,"weight":0.1}},"compositeScore":2.1,"levelCapped":false,"levelChange":{"from":2,"to":2,"direction":"same","reason":null},"biggestBreakthrough":{"date":"2025-07-30","title":{"zh":"AlphaEarth Foundations 全球地表嵌入","en":"AlphaEarth Foundations global embeddings"},"url":"https://deepmind.google/discover/blog/alphaearth-foundations-helps-map-our-planet-in-unprecedented-detail/","aiImpact":"major","fieldImpact":"major","why":{"zh":"10 米分辨率全球年度地表嵌入，误差比其他方法低约 24%","en":"10 m global annual embeddings"}},"keyPapers":[{"title":"Continental-Scale Building Detection from High Resolution Satellite Imagery","titleZh":"Google Open Buildings：用深度学习从卫星影像检测非洲大陆级建筑轮廓（arXiv 2107.12283）","id":"arXiv:2107.12283","url":"https://arxiv.org/abs/2107.12283","venue":"arXiv","date":"2021-07-26","why":{"zh":"为缺乏测绘的地区提供建筑底图，被人道与规划机构广泛使用","en":"Building maps for under-mapped regions"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Dynamic World, Near real-time global 10 m land use land cover mapping","titleZh":"Google/WRI Dynamic World：深度学习生成近实时全球 10 米土地覆盖（Scientific Data）","id":"10.1038/s41597-022-01307-4","url":"https://doi.org/10.1038/s41597-022-01307-4","venue":"Scientific Data","date":"2022-06-09","why":{"zh":"土地覆盖制图从年度产品变为随卫星过境更新","en":"Land-cover mapping moves from annual to per-overpass"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"SatMAE: Pre-training Transformers for Temporal and Multi-Spectral Satellite Imagery","titleZh":"SatMAE：多光谱、多时相卫星影像的自监督预训练（NeurIPS 2022）","id":"arXiv:2207.08051","url":"https://arxiv.org/abs/2207.08051","venue":"arXiv","date":"2022-07-17","why":{"zh":"遥感基础模型路线的早期代表","en":"Early remote-sensing foundation model"},"fieldImpact":"incremental","aiImpact":"major"},{"title":"TerraMind: Large-Scale Generative Multimodality for Earth Observation","titleZh":"ESA 与 IBM 发布 TerraMind：融合 9 种模态的开源地球观测生成式基础模型（ICCV 2025）","id":"arXiv:2504.11171","url":"https://arxiv.org/abs/2504.11171","venue":"arXiv","date":"2025-04-15","why":{"zh":"遥感基础模型从单模态走向多模态生成","en":"EO foundation models go multimodal-generative"},"fieldImpact":"incremental","aiImpact":"incremental"}],"latest_papers":[{"id":"arXiv:2610.28313","title":"Towards Consistent Geospatial Representations: Aligning One Encoder to Many Foundation Models","titleZh":"迈向一致的地理空间表征：让一个编码器对齐多个基础模型","summaryZh":"发布 Pangaea-Latents（5 个地理空间基础模型在 11 个数据集上的嵌入），发现各模型表征互补；提出 GeoMixture 对齐多个模型，平均排名与最佳模型持平且更稳定。","summaryEn":"Shows geospatial foundation models are complementary and aligns one encoder to many, matching the best average rank.","authors":["Arthur Ouaknine","David Rolnick"],"date":"2026-10-20","url":"https://arxiv.org/abs/2610.28313","site":"arXiv","venue":null,"category":"cs.CV","fetchedAt":"2026-10-22T10:09:33"},{"id":"arXiv:2610.27110","title":"MatchCityLoc: Aerial Localization in LoD City Models via Surface Normal Matching","titleZh":"MatchCityLoc：通过表面法向匹配在 LoD 城市模型中做无人机定位","summaryZh":"用零样本基础模型从图像提取表面法向，与渲染的无纹理城市模型法向直接匹配求位姿，用于 GNSS 不可靠的城市环境。","summaryEn":"Matches surface normals from a zero-shot model to LoD city models for UAV localisation without GNSS.","authors":["Oussema Dhaouadi","Zuria Bauer","Johannes Michael Meier","Marc Pollefeys"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.27110","site":"arXiv","venue":null,"category":"cs.CV","fetchedAt":"2026-10-22T10:09:33"},{"id":"arXiv:2610.20973","title":"PiCo: Physics-Informed Concept Probing for Geo-Foundation Models","titleZh":"PiCo：面向地理基础模型的物理信息概念探测","summaryZh":"用三维辐射传输模型和游戏引擎生成合成卫星观测，在地理基础模型嵌入空间中找出物理概念向量，解释模型预测。","summaryEn":"Uses synthetic radiative-transfer imagery to find physical concept vectors that explain geo-foundation-model predictions.","authors":["Yihang She","Andrew Blake","Jianbo Qi","Srinivasan Keshav"],"date":"2026-10-14","url":"https://arxiv.org/abs/2610.20973","site":"arXiv","venue":null,"category":"cs.CV","fetchedAt":"2026-10-22T10:09:33"},{"id":"arXiv:2610.13857","title":"Revisiting the Task of Building Damage Assessment by Satellite Imaging","titleZh":"重新审视卫星影像建筑损毁评估任务","summaryZh":"回顾十年来卫星影像建筑损毁评估研究，指出领域缺乏共同基础，并重新梳理任务设定。","summaryEn":"Revisits satellite building-damage assessment, arguing the field lacks a shared foundation despite a decade of work.","authors":["Anthony Cioppa","Virgile Blua","Sébastien Piérard","Pietro Ceccato"],"date":"2026-10-09","url":"https://arxiv.org/abs/2610.13857","site":"arXiv","venue":null,"category":"cs.CV","fetchedAt":"2026-10-22T10:09:33"},{"id":"arXiv:2610.13766","title":"Shortcut Learning in Geospatial Foundation Model Embeddings for Cross-Disaster Damage Assessment","titleZh":"跨灾害损毁评估中地理基础模型嵌入的捷径学习","summaryZh":"留出整场灾害做测试时，TerraMind、Prithvi-EO-2.0 等模型的平均平衡准确率在 xView2 上从 0.807 降至 0.638，在 BRIGHT 上从 0.818 降至 0.568。","summaryEn":"Holding out whole disasters cuts balanced accuracy of geo-foundation-model embeddings from 0.807 to 0.638 (xView2) and 0.818 to 0.568 (BRIGHT).","authors":["Ali Shibli","Andrea Nascetti","Yifang Ban"],"date":"2026-10-09","url":"https://arxiv.org/abs/2610.13766","site":"arXiv","venue":null,"category":"cs.CV","fetchedAt":"2026-10-22T10:09:33"},{"id":"arXiv:2610.05790","title":"FairRSFM: A Biome-Aware Benchmark and Debiasing Framework for Remote Sensing Foundation Models","titleZh":"FairRSFM：考虑生物群系的遥感基础模型基准与去偏框架","summaryZh":"把样本按 14 类陆地生物群系分为 6 组，评估遥感基础模型在不同生态区的表现差异。","summaryEn":"A biome-aware benchmark exposing performance disparities of remote-sensing foundation models across ecological regions.","authors":["Md Aminur Hossain","Omkumar Vaghasiya","Rajeev Ranjan Dwivedi","Vinod Kurmi"],"date":"2026-10-05","url":"https://arxiv.org/abs/2610.05790","site":"arXiv","venue":null,"category":"cs.CV","fetchedAt":"2026-10-22T10:09:33"},{"id":"arXiv:2610.05731","title":"T-JEPA: A Temporal Joint-Embedding Predictive Architecture for Learning Better Remote Sensing Representations","titleZh":"T-JEPA：学习更好遥感表示的时间联合嵌入预测架构","summaryZh":"利用对地观测序列中稀疏不均匀的时间间隔作为监督信号，学习以时间间隔为条件的潜在状态转变。","summaryEn":"T-JEPA learns time-gap-conditioned latent transitions from irregular Earth-observation sequences.","authors":["Bowen Peng","Li Liu","Yongxiang Liu","Weijie Li"],"date":"2026-10-05","url":"https://arxiv.org/abs/2610.05731","site":"arXiv","venue":null,"category":"cs.CV","fetchedAt":"2026-10-22T10:09:33"},{"id":"arXiv:2609.37340","title":"HyperSAM: A Promptable Foundation Model for Hyperspectral Remote Sensing","titleZh":"HyperSAM：可提示的高光谱遥感基础模型","summaryZh":"面向高光谱遥感构建可提示分割的通用基础模型，缓解高分辨率密集标注稀缺和从零训练的问题。","summaryEn":"A promptable foundation model for hyperspectral remote sensing.","authors":["Li Pang","Xinqiao Wu","Jing Yao","Pedram Ghamisi"],"date":"2026-09-29","url":"https://arxiv.org/abs/2609.37340","site":"arXiv","venue":null,"category":"cs.CV","fetchedAt":"2026-10-22T10:09:33"}]},{"id":"economics","slug":"economics","category":"social","categoryName":{"zh":"社会科学","en":"Social sciences"},"name":{"zh":"经济学","en":"Economics"},"classification":{"oecdFos":"5.2","moe":"0201 理论经济学 / 0202 应用经济学"},"level":1,"levelKey":"assistive","levelLabel":{"zh":"辅助起步","en":"Assistive"},"trend":"flat","reason":{"zh":"保持 L1（接近 L2）：机器学习因果推断与文本方法已常用，NBER 2026 年出现用大模型工作流复现并扩展已发表论文的研究；但 2026 年《自然·人类行为》跨 74 国研究表明大模型模拟被试只是人类经济偏好的弱代理，学科尚无以 AI 为关键的公认新发现。","en":"Stays L1 (close to L2): ML causal inference and text methods are common and a 2026 NBER paper uses an LLM workflow to reproduce and extend published work, but a 74-country Nature Human Behaviour study shows simulated LLM respondents are weak proxies for human preferences; no recognised AI-key discovery yet."},"reasonSources":[{"url":"https://www.nber.org/papers/w35782","linkCheck":"ok"},{"url":"https://www.nature.com/articles/s41562-026-02611-4","linkCheck":"ok"}],"directions":{"zh":["大模型模拟经济主体","文本即数据","AI 对劳动市场影响（研究对象）"],"en":["LLM-simulated agents","Text as data","AI labour-market impact (as subject)"]},"stats":{"milestones":7,"milestonesSince2025":4,"milestones2026":3,"milestonesLast6Months":3,"news":10,"latestProgress":"2026-10-20","latestMilestone":"2026-10-12","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":1,"leaps":{"ai":0,"field":0},"latestPapers":8,"keyPapers":5},"milestones":[{"date":"2023-07-14","title":{"zh":"Noy、Zhang：ChatGPT 使专业写作任务耗时减少约 40%、质量提升 18%（Science 随机实验）","en":"Noy & Zhang: experimental evidence on generative AI productivity in writing tasks (Science)"},"url":"https://doi.org/10.1126/science.adh2586","source":"Science","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"首个顶刊随机实验测量生成式 AI 对知识工作的生产率影响","en":"First top-journal RCT of generative AI productivity in knowledge work"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"首个顶刊随机实验测量生成式 AI 对知识工作的生产率影响","en":"First top-journal RCT of generative AI productivity in knowledge work"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首个顶刊随机实验测量生成式 AI 对知识工作的生产率影响","en":"First top-journal RCT of generative AI productivity in knowledge work"}},"impactNotes":[],"verified":{"id":"10.1126/science.adh2586","title":"Experimental evidence on the productivity effects of generative artificial intelligence","venue":"Science","date":"2023-07-14"},"addedIn":"v2"},{"date":"2023-12-01","title":{"zh":"Korinek：经济研究中的生成式 AI 用例与启示（Journal of Economic Literature）","en":"Korinek: Generative AI for economic research (JEL)"},"url":"https://doi.org/10.1257/jel.20231736","source":"Journal of Economic Literature","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"经济学权威综述期刊系统讨论大模型改变研究流程","en":"Authoritative review of LLMs in economic research"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"经济学权威综述期刊系统讨论大模型改变研究流程","en":"Authoritative review of LLMs in economic research"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"经济学权威综述期刊系统讨论大模型改变研究流程","en":"Authoritative review of LLMs in economic research"}},"impactNotes":[],"verified":{"id":"10.1257/jel.20231736","title":"Generative AI for Economic Research: Use Cases and Implications for Economists","venue":"Journal of Economic Literature","date":"2023-12-01"},"addedIn":"v2"},{"date":"2024-07","title":{"zh":"“Homo Silicus”：大模型作为模拟经济主体（ACM EC 2024）","en":"“Homo Silicus”: LLMs as simulated economic agents (ACM EC 2024)"},"url":"https://doi.org/10.1145/3670865.3673513","source":"ACM EC","flags":["tbv"],"flagLabels":[{"zh":"日期或细节待核实","en":"Date/details to be verified"}],"linkCheck":"ok-crossref","datePrecision":"month","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"提出用大模型模拟经济主体做实验","en":"LLMs as simulated economic agents"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"提出用大模型模拟经济主体做实验","en":"LLMs as simulated economic agents"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"提出用大模型模拟经济主体做实验","en":"LLMs as simulated economic agents"}},"impactNotes":[],"verified":{"id":"10.1145/3670865.3673513","title":"Large Language Models as Simulated Economic Agents: What Can We Learn from Homo Silicus?","venue":"Proceedings of the 25th ACM Conference on Economics and Computation","date":"2024-12-17"},"addedIn":"v1"},{"date":"2025-02-04","title":{"zh":"Brynjolfsson 等：生成式 AI 在客服工作中提升 14% 生产率，新手提升最大（QJE）","en":"Generative AI at Work: 14% productivity gain, largest for novices (QJE)"},"url":"https://doi.org/10.1093/qje/qjae044","source":"Quarterly Journal of Economics","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"首批顶刊实证测量生成式 AI 对劳动生产率的影响","en":"Among first top-journal estimates of generative-AI productivity"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"首批顶刊实证测量生成式 AI 对劳动生产率的影响","en":"Among first top-journal estimates of generative-AI productivity"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首批顶刊实证测量生成式 AI 对劳动生产率的影响","en":"Among first top-journal estimates of generative-AI productivity"}},"impactNotes":[],"verified":{"id":"10.1093/qje/qjae044","title":"Generative AI at Work","venue":"The Quarterly Journal of Economics","date":"2025-02-04"},"addedIn":"v2"},{"date":"2026-09-28","title":{"zh":"NBER：一个能复现、改进并扩展已发表经济学研究的大模型工作流（w35782）","en":"NBER w35782: an LLM workflow that reproduces, improves and extends published economics research"},"url":"https://www.nber.org/papers/w35782","source":"NBER","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"AI 开始承担经济学研究的复现与扩展环节","en":"AI takes on replication and extension"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI 开始承担经济学研究的复现与扩展环节","en":"AI takes on replication and extension"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI 开始承担经济学研究的复现与扩展环节","en":"AI takes on replication and extension"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-09-28","title":{"zh":"NBER：首次用 2026 年 6–8 月 CPS 数据估计 AI 对应届大学毕业生就业的影响，失业率未见显著上升（w35796）","en":"NBER: The Early Impacts of AI on Employment among Recent College Graduates — no significant rise in summer 2026 (w35796)"},"url":"https://www.nber.org/papers/w35796","source":"NBER","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"AI：针对被认为最易被 AI 替代的群体；学科：双重差分与事件研究均未发现相对失业率上升，加入“想工作但未求职”的扩展口径后结论不变，为“AI 冲击入门岗位”之争提供反证（工作论文）","en":"AI: targets the group thought most exposed. Field: DiD and event-study find no relative rise, even with an expanded unemployment measure (working paper)"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：针对被认为最易被 AI 替代的群体；学科：双重差分与事件研究均未发现相对失业率上升，加入“想工作但未求职”的扩展口径后结论不变，为“AI 冲击入门岗位”之争提供反证（工作论文）","en":"AI: targets the group thought most exposed. Field: DiD and event-study find no relative rise, even with an expanded unemployment measure (working paper)"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：针对被认为最易被 AI 替代的群体；学科：双重差分与事件研究均未发现相对失业率上升，加入“想工作但未求职”的扩展口径后结论不变，为“AI 冲击入门岗位”之争提供反证（工作论文）","en":"AI: targets the group thought most exposed. Field: DiD and event-study find no relative rise, even with an expanded unemployment measure (working paper)"}},"impactNotes":[],"verified":null,"addedIn":"v2-20261022pm"},{"date":"2026-10-12","title":{"zh":"跨 74 国研究：大模型模拟回答是人类经济偏好的弱代理（Nat. Hum. Behav.）","en":"74-country study: LLM answers are weak proxies for human economic preferences"},"url":"https://www.nature.com/articles/s41562-026-02611-4","source":"Nature Human Behaviour","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"74 国研究表明模拟回答只是弱代理，是重要反证","en":"74-country evidence that simulations are weak proxies"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"74 国研究表明模拟回答只是弱代理，是重要反证","en":"74-country evidence that simulations are weak proxies"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"74 国研究表明模拟回答只是弱代理，是重要反证","en":"74-country evidence that simulations are weak proxies"}},"impactNotes":[],"verified":{"id":"10.1038/s41562-026-02611-4","title":"Simulated responses from large language models are weak proxies for human economic preferences across 74 countries","venue":"Nature Human Behaviour","date":"2026-10-12"},"addedIn":"v1"}],"papersRepos":[{"title":"Experimental evidence on the productivity effects of generative artificial intelligence","url":"https://doi.org/10.1126/science.adh2586","date":"2023-07-14","kind":"paper","linkCheck":"ok-crossref"},{"title":"Generative AI for Economic Research: Use Cases and Implications for Economists","url":"https://doi.org/10.1257/jel.20231736","date":"2023-12-01","kind":"paper","linkCheck":"ok-crossref"},{"title":"Large Language Models as Simulated Economic Agents: What Can We Learn from Homo Silicus?","url":"https://doi.org/10.1145/3670865.3673513","date":"2024-12-17","kind":"paper","linkCheck":"ok-crossref"},{"title":"Generative AI at Work","url":"https://doi.org/10.1093/qje/qjae044","date":"2025-02-04","kind":"paper","linkCheck":"ok-crossref"},{"title":"Simulated responses from large language models are weak proxies for human economic preferences across 74 countries","url":"https://doi.org/10.1038/s41562-026-02611-4","date":"2026-10-12","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"NBER","url":"https://www.nber.org"},{"name":"Stanford Digital Economy Lab","url":"https://digitaleconomy.stanford.edu"}],"news":[{"id":"eb477756520c","date":"2026-10-20","title":"How 3 Java Champions are using GitHub Copilot for increased productivity","url":"https://github.blog/ai-and-ml/github-copilot/how-3-java-champions-are-using-github-copilot-for-increased-productivity/","source":"The GitHub Blog","lang":"en","kind":"news","summaryZh":"The GitHub Blog于2026-10-21发布的文章，涉及AI 软件开发，原题《How 3 Java Champions are using GitHub Copilot for increased productivity》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"ebc7ad18c4ce","date":"2026-10-20","title":"The new economics of enterprise AI","url":"https://www.technologyreview.com/2026/10/20/1146937/the-new-economics-of-enterprise-ai/","source":"MIT Technology Review - AI","lang":"en","kind":"news","summaryZh":"MIT Technology Review - AI于2026-10-20发布的文章，涉及AI 综合、AI 气候能源，原题《The new economics of enterprise AI》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"a4348ade7db4","date":"2026-10-18","title":"Global economy set for AI-driven surge, FT survey suggests","url":"https://www.ft.com/content/c6aeb3a1-1874-4947-8ffc-4d993301244a?syn-25a6b1a6=1","source":"Financial Times AI 栏目","lang":"en","kind":"news","summaryZh":"Financial Times AI 栏目于2026-10-18发布的文章，涉及AI 金融，原题《Global economy set for AI-driven surge, FT survey suggests》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"bc7f44c7eb34","date":"2026-10-13","title":"The Humanoid Economy by Modar Alaoui Published Today by Wiley","url":"http://www.RoboticsTomorrow.com/news/2026/10/13/the-humanoid-economy-by-modar-alaoui-published-today-by-wiley/27235","source":"Robotics Tomorrow","lang":"en","kind":"news","summaryZh":"Robotics Tomorrow于2026-10-13发布的文章，涉及AI 机器人，原题《The Humanoid Economy by Modar Alaoui Published Today by Wiley》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"cdeb54f0a306","date":"2026-10-12","title":"Nature Human Behaviour: 74-country study of generative AI use","url":"https://www.nature.com/articles/s41562-026-02611-4","source":"Nature Human Behaviour","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"ok"},{"id":"3b307531cba8","date":"2026-10-08","title":"Robots at Work: How Intelligent Automation Is Reshaping the Global Economy","url":"http://www.RoboticsTomorrow.com/news/2026/10/08/robots-at-work-how-intelligent-automation-is-reshaping-the-global-economy/27221","source":"Robotics Tomorrow","lang":"en","kind":"news","summaryZh":"Robotics Tomorrow于2026-10-08发布的文章，涉及AI 机器人、AI 安全，原题《Robots at Work: How Intelligent Automation Is Reshaping the Global Economy》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"2a01f0d16ca8","date":"2026-10-05","title":"Import AI 475: Swarm scaling; Google DeepMind watermarks biology; and the AI science economy","url":"https://importai.substack.com/p/import-ai-475-swarm-scaling-google","source":"Import AI (Jack Clark)","lang":"en","kind":"news","summaryZh":"Import AI 周刊第 475 期：群体扩展、DeepMind 给生物序列加水印、AI 科研经济。","via":"research-samples-20261010","linkCheck":"from-feed"},{"id":"80187c39ac57","date":"2026-10-05","title":"Too big to pause: Could an AI slowdown crash the economy?","url":"https://cointelegraph.com/magazine/too-big-to-pause-the-economic-stakes-of-an-ai-slowdown?utm_source=rss_feed&utm_medium=rss_tag_ai&utm_campaign=rss_partner_inbound","source":"Cointelegraph AI","lang":"en","kind":"news","summaryZh":"Cointelegraph AI于2026-10-05发布的文章，涉及AI 金融、AI 商业融资，原题《Too big to pause: Could an AI slowdown crash the economy?》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"bbe49531d638","date":"2026-09-28","title":"NBER w35782","url":"https://www.nber.org/papers/w35782","source":"NBER","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"ok"},{"id":"35c189191415","date":"2026-09-23","title":"Report: AI's Productivity Gains in Science Tempered by Time Spent Validating Outputs","url":"https://thejournal.com/articles/2026/09/23/report-ais-productivity-gains-in-science-tempered-by-time-spent-validating-outputs.aspx","source":"THE Journal","lang":"en","kind":"news","summaryZh":"THE Journal于2026-09-24发布的文章，涉及AI 教育，原题《Report: AI's Productivity Gains in Science Tempered by Time Spent Validating Outputs》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"}],"review":{"lastReviewed":"2026-10-22","cycleDays":90,"nextReview":"2027-01-19"},"urls":{"zh":"/zh-hans/disciplines/economics/","en":"/en/disciplines/economics/"},"dimensions":{"method":{"score":2,"label":{"zh":"研究方法","en":"Research method"},"evidence":"机器学习因果推断和文本数据方法已进入主流实证研究；LLM 工作流开始用于复现论文（NBER 2026）","source":"https://www.nber.org/papers/w35782","weight":0.3},"discovery":{"score":1,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"没有以 AI 为关键的公认经济学新发现；大模型被试被证明只是人类偏好的弱代理","source":"https://www.nature.com/articles/s41562-026-02611-4","weight":0.35},"adoption":{"score":1,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"缺少经济学者使用率调查","source":null,"weight":0.25},"education":{"score":1,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到课程证据","source":null,"weight":0.1}},"compositeScore":1.3,"levelCapped":false,"levelChange":{"from":1,"to":1,"direction":"same","reason":null},"biggestBreakthrough":{"date":"2026-09-28","title":{"zh":"NBER：一个能复现、改进并扩展已发表经济学研究的大模型工作流（w35782）","en":"NBER w35782: an LLM workflow that reproduces, improves and extends published economics research"},"url":"https://www.nber.org/papers/w35782","aiImpact":"major","fieldImpact":"major","why":{"zh":"AI 开始承担经济学研究的复现与扩展环节","en":"AI takes on replication and extension"}},"keyPapers":[{"title":"Experimental evidence on the productivity effects of generative artificial intelligence","titleZh":"Noy、Zhang：ChatGPT 使专业写作任务耗时减少约 40%、质量提升 18%（Science 随机实验）","id":"10.1126/science.adh2586","url":"https://doi.org/10.1126/science.adh2586","venue":"Science","date":"2023-07-14","why":{"zh":"首个顶刊随机实验测量生成式 AI 对知识工作的生产率影响","en":"First top-journal RCT of generative AI productivity in knowledge work"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Generative AI for Economic Research: Use Cases and Implications for Economists","titleZh":"Korinek：经济研究中的生成式 AI 用例与启示（Journal of Economic Literature）","id":"10.1257/jel.20231736","url":"https://doi.org/10.1257/jel.20231736","venue":"Journal of Economic Literature","date":"2023-12-01","why":{"zh":"经济学权威综述期刊系统讨论大模型改变研究流程","en":"Authoritative review of LLMs in economic research"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Large Language Models as Simulated Economic Agents: What Can We Learn from Homo Silicus?","titleZh":"“Homo Silicus”：大模型作为模拟经济主体（ACM EC 2024）","id":"10.1145/3670865.3673513","url":"https://doi.org/10.1145/3670865.3673513","venue":"Proceedings of the 25th ACM Conference on Economics and Computation","date":"2024-12-17","why":{"zh":"提出用大模型模拟经济主体做实验","en":"LLMs as simulated economic agents"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Generative AI at Work","titleZh":"Brynjolfsson 等：生成式 AI 在客服工作中提升 14% 生产率，新手提升最大（QJE）","id":"10.1093/qje/qjae044","url":"https://doi.org/10.1093/qje/qjae044","venue":"The Quarterly Journal of Economics","date":"2025-02-04","why":{"zh":"首批顶刊实证测量生成式 AI 对劳动生产率的影响","en":"Among first top-journal estimates of generative-AI productivity"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Simulated responses from large language models are weak proxies for human economic preferences across 74 countries","titleZh":"跨 74 国研究：大模型模拟回答是人类经济偏好的弱代理（Nat. Hum. Behav.）","id":"10.1038/s41562-026-02611-4","url":"https://doi.org/10.1038/s41562-026-02611-4","venue":"Nature Human Behaviour","date":"2026-10-12","why":{"zh":"74 国研究表明模拟回答只是弱代理，是重要反证","en":"74-country evidence that simulations are weak proxies"},"fieldImpact":"major","aiImpact":"incremental"}],"latest_papers":[{"id":"arXiv:2610.27296","title":"Stronger artificial intelligence breakthroughs hasten their own exhaustion","titleZh":"越强的 AI 突破越快耗尽自身","summaryZh":"分析计算机科学 36 年 720 万篇论文、7.3 万份评审和 370 万研究者，发现 Transformer 之后注意力急剧集中，每次突破提高了“思想承载量”，随后却更快耗尽；结构生物学在冷冻电镜和 AlphaFold 后也出现同样周期。","summaryEn":"7.2M papers show each AI breakthrough concentrates attention and is followed by faster exhaustion of ideas.","authors":["Beichen Lu","Honglin Bao","Qi Huang","Qingguo Meng"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.27296","site":"arXiv","venue":null,"category":"cs.AI","fetchedAt":"2026-10-22T10:09:36"},{"id":"arXiv:2610.24267","title":"How exposed is work to robotics? Occupational, industry, and geographic measures","titleZh":"工作有多大程度暴露于机器人？职业、行业与地区测度","summaryZh":"构建首个基于能力的机器人与物理 AI 暴露指数，覆盖职业、行业和地区，含 8 个领域 16 项机器人能力。","summaryEn":"First abilities-based exposure measures for robotics and physical AI across occupations, industries and regions.","authors":["Sebastian Sartor","Robert Seamans","Neil Thompson"],"date":"2026-10-16","url":"https://arxiv.org/abs/2610.24267","site":"arXiv","venue":null,"category":"cs.RO","fetchedAt":"2026-10-22T10:09:36"},{"id":"arXiv:2610.24035","title":"TRACE: Governing AI-Assisted Research with Principles from Computer Science","titleZh":"TRACE：借鉴计算机科学原则治理 AI 辅助研究","summaryZh":"从软件工程和安全研究中提炼五条原则（记录、对照来源审查、独立接受、控制转换、评估更正），用于组织 AI 辅助的研究工作。","summaryEn":"Five principles (TRACE) from CS practice for governing AI-assisted research.","authors":["Amirhossein Zohrehvand","Aldona Kapačinskaitė"],"date":"2026-10-16","url":"https://arxiv.org/abs/2610.24035","site":"arXiv","venue":null,"category":"econ.GN","fetchedAt":"2026-10-22T10:09:36"},{"id":"arXiv:2610.21711","title":"How language model agents reshape price competition","titleZh":"大模型智能体如何重塑价格竞争","summaryZh":"在重复伯特兰竞争中，大模型定价智能体可以摆脱接近边际成本的纳什结果，具体取决于部署方式。","summaryEn":"LLM pricing agents can escape near-marginal-cost Bertrand outcomes depending on deployment choices.","authors":["Martin Bichler","Philip Hierhager","Kassian Köck","Fabian Pieroth"],"date":"2026-10-15","url":"https://arxiv.org/abs/2610.21711","site":"arXiv","venue":null,"category":"cs.GT","fetchedAt":"2026-10-22T10:09:36"},{"id":"arXiv:2610.17674","title":"AI Information Synthesis and Content Polarization","titleZh":"AI 信息综合与内容极化","summaryZh":"用博弈论模型说明：AI 综合信息本可减少极化，但内容生产者的策略性反应可能抵消这一效果。","summaryEn":"A game-theoretic model shows content producers’ strategic responses can undermine AI’s potential to reduce polarisation.","authors":["Chang Liu","Yuting Zhu"],"date":"2026-10-13","url":"https://arxiv.org/abs/2610.17674","site":"arXiv","venue":null,"category":"econ.TH","fetchedAt":"2026-10-22T10:09:36"},{"id":"arXiv:2610.23934","title":"Organizing LLM Inference for Aggregate Predictions from Behavioral Data","titleZh":"组织大模型推理以从行为数据做总体预测","summaryZh":"在电商场景中研究用大模型模拟用户群体时，先汇总会话还是逐会话预测再平均，两者的偏差-方差权衡。","summaryEn":"Studies bias–variance trade-offs in aggregating behavioural histories when LLMs simulate user populations.","authors":["Varun Gupta","Vijay Kamble"],"date":"2026-10-07","url":"https://arxiv.org/abs/2610.23934","site":"arXiv","venue":null,"category":"cs.LG","fetchedAt":"2026-10-22T10:09:36"},{"id":"arXiv:2610.10506","title":"Validity Without Ground Truth: What Stated-Preference Economics Offers the Evaluation of Language Models","titleZh":"无真值时的效度：陈述偏好经济学能为大模型评估提供什么","summaryZh":"主张把陈述偏好经济学中的内容、构念、效标效度和激励相容等框架用于评估无标准答案的大模型输出。","summaryEn":"Argues stated-preference validity concepts offer a general method for evaluating LLMs without ground truth.","authors":["Daniel Robert Kling Alexander","Catherine Louise Kling"],"date":"2026-10-07","url":"https://arxiv.org/abs/2610.10506","site":"arXiv","venue":null,"category":"cs.AI","fetchedAt":"2026-10-22T10:09:36"},{"id":"arXiv:2610.08264","title":"Understanding Interfirm AI Talent Flow Networks through Online Professional Profiles","titleZh":"从在线职业档案看企业间 AI 人才流动网络","summaryZh":"基于 58 个国家约 5.35 亿条就业记录重建 2010–2022 年 AI 人才流动网络，发现人才流入日益集中于少数领先企业。","summaryEn":"Reconstructs 2010–2022 inter-firm AI talent flows from ~535M employment records, finding growing concentration in a few leading firms.","authors":["Donghang Li","Yunhan Zheng","Alok Prakash","Shenhao Wang"],"date":"2026-10-06","url":"https://arxiv.org/abs/2610.08264","site":"arXiv","venue":null,"category":"cs.SI","fetchedAt":"2026-10-22T10:09:36"}]},{"id":"finance","slug":"finance","category":"social","categoryName":{"zh":"社会科学","en":"Social sciences"},"name":{"zh":"金融学","en":"Finance"},"classification":{"oecdFos":"5.2","moe":"020204 金融学"},"level":2,"levelKey":"breakthrough","levelLabel":{"zh":"局部突破","en":"Local breakthroughs"},"trend":"flat","reason":{"zh":"机器学习资产定价（Gu、Kelly、Xiu，RFS 2020）已成为实证金融的主流方法之一，量化与风控广泛使用；大模型在研究与投研中仍以辅助为主。","en":"ML asset pricing (Gu, Kelly & Xiu, RFS 2020) is a mainstream empirical method; quant and risk use is widespread, while LLMs remain assistive."},"reasonSources":[{"url":"https://doi.org/10.1093/rfs/hhaa009","linkCheck":"ok-crossref"}],"directions":{"zh":["机器学习资产定价","另类数据","金融大模型"],"en":["ML asset pricing","Alternative data","Financial LLMs"]},"stats":{"milestones":6,"milestonesSince2025":1,"milestones2026":1,"milestonesLast6Months":1,"news":12,"latestProgress":"2026-10-20","latestMilestone":"2026-10-16","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":0,"leaps":{"ai":0,"field":0},"latestPapers":8,"keyPapers":5},"milestones":[{"date":"2020-02-26","title":{"zh":"Gu、Kelly、Xiu：机器学习实证资产定价（RFS）","en":"Gu, Kelly & Xiu: Empirical asset pricing via ML (RFS)"},"url":"https://doi.org/10.1093/rfs/hhaa009","source":"Review of Financial Studies","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"机器学习使收益预测的夏普比率约翻倍，成为该领域被引最多的论文之一；属方法改进而非新理论","en":"ML roughly doubles Sharpe ratios; a method advance, not a new theory"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"机器学习使收益预测的夏普比率约翻倍，成为该领域被引最多的论文之一；属方法改进而非新理论","en":"ML roughly doubles Sharpe ratios; a method advance, not a new theory"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"机器学习使收益预测的夏普比率约翻倍，成为该领域被引最多的论文之一；属方法改进而非新理论","en":"ML roughly doubles Sharpe ratios; a method advance, not a new theory"}},"impactNotes":[],"verified":{"id":"10.1093/rfs/hhaa009","title":"Empirical Asset Pricing via Machine Learning","venue":"The Review of Financial Studies","date":"2020-02-26"},"addedIn":"v1"},{"date":"2023-03-30","title":{"zh":"BloombergGPT 金融大模型","en":"BloombergGPT financial LLM"},"url":"https://arxiv.org/abs/2303.17564","source":"arXiv","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok","datePrecision":"day","type":"system","typeLabel":{"zh":"系统发布","en":"System"},"why":{"zh":"首个大型金融专用语言模型","en":"First large finance-specific LM"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首个大型金融专用语言模型","en":"First large finance-specific LM"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"首个大型金融专用语言模型","en":"First large finance-specific LM"}},"impactNotes":[],"verified":{"id":"arXiv:2303.17564","title":"BloombergGPT: A Large Language Model for Finance","venue":"arXiv","date":"2023-03-30"},"addedIn":"v1"},{"date":"2023-04-15","title":{"zh":"ChatGPT 能否预测股价变动？新闻标题情绪与次日收益（arXiv 2304.07619）","en":"Can ChatGPT forecast stock price movements? (arXiv 2304.07619)"},"url":"https://arxiv.org/abs/2304.07619","source":"arXiv","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"大模型新闻解读产生显著收益可预测性，引发大量跟进","en":"LLM headline reading predicts returns; heavily followed up"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"大模型新闻解读产生显著收益可预测性，引发大量跟进","en":"LLM headline reading predicts returns; heavily followed up"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"大模型新闻解读产生显著收益可预测性，引发大量跟进","en":"LLM headline reading predicts returns; heavily followed up"}},"impactNotes":[],"verified":{"id":"arXiv:2304.07619","title":"Can ChatGPT Forecast Stock Price Movements? Return Predictability and Large Language Models","venue":"arXiv","date":"2023-04-15"},"addedIn":"v2"},{"date":"2023-12-21","title":{"zh":"Kelly、Malamud、Zhou：“复杂性的优点”——参数远多于样本的机器学习模型反而提升收益预测（Journal of Finance）","en":"Kelly, Malamud & Zhou: The Virtue of Complexity in Return Prediction (J. Finance)"},"url":"https://doi.org/10.1111/jofi.13298","source":"The Journal of Finance","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"为“大模型式”高复杂度预测在金融中的有效性给出理论与实证依据，改变了对过拟合的传统看法","en":"Theory and evidence that highly over-parameterised models improve return prediction"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"为“大模型式”高复杂度预测在金融中的有效性给出理论与实证依据，改变了对过拟合的传统看法","en":"Theory and evidence that highly over-parameterised models improve return prediction"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"为“大模型式”高复杂度预测在金融中的有效性给出理论与实证依据，改变了对过拟合的传统看法","en":"Theory and evidence that highly over-parameterised models improve return prediction"}},"impactNotes":[],"verified":{"id":"10.1111/jofi.13298","title":"The Virtue of Complexity in Return Prediction","venue":"The Journal of Finance","date":"2023-12-21"},"addedIn":"v2"},{"date":"2024-02","title":{"zh":"Chen、Pelger、Zhu：用深度学习估计满足无套利条件的资产定价模型（Management Science）","en":"Chen, Pelger & Zhu: Deep Learning in Asset Pricing (Management Science)"},"url":"https://doi.org/10.1287/mnsc.2023.4695","source":"Management Science","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"month","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"把经济约束（无套利）嵌入深度网络，是机器学习资产定价的代表作","en":"Embeds no-arbitrage in deep nets; representative ML asset-pricing work"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"把经济约束（无套利）嵌入深度网络，是机器学习资产定价的代表作","en":"Embeds no-arbitrage in deep nets; representative ML asset-pricing work"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"把经济约束（无套利）嵌入深度网络，是机器学习资产定价的代表作","en":"Embeds no-arbitrage in deep nets; representative ML asset-pricing work"}},"impactNotes":[],"verified":{"id":"10.1287/mnsc.2023.4695","title":"Deep Learning in Asset Pricing","venue":"Management Science","date":"2024-02"},"addedIn":"v2"},{"date":"2026-10-16","title":{"zh":"英国金融行为监管局（FCA）：与英格兰银行联合调查显示 86% 的金融机构已在使用 AI（2024 年为 75%）","en":"FCA: joint survey with the Bank of England shows 86% of financial firms now use AI (75% in 2024)"},"url":"https://www.fca.org.uk/news/speeches/building-future-finance-age-ai","source":"FCA","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"industry","typeLabel":{"zh":"产业/采用","en":"Industry"},"why":{"zh":"监管机构口径的行业采用率，说明 AI 已成为金融业常规工具；完整报告尚未发布","en":"Regulator-measured adoption; full report pending"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"监管机构口径的行业采用率，说明 AI 已成为金融业常规工具；完整报告尚未发布","en":"Regulator-measured adoption; full report pending"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"监管机构口径的行业采用率，说明 AI 已成为金融业常规工具；完整报告尚未发布","en":"Regulator-measured adoption; full report pending"}},"impactNotes":[],"verified":null,"addedIn":"v2"}],"papersRepos":[{"title":"BloombergGPT (arXiv 2303.17564)","url":"https://arxiv.org/abs/2303.17564","date":"2023-03","kind":"paper","linkCheck":"ok"},{"title":"Empirical Asset Pricing via Machine Learning","url":"https://doi.org/10.1093/rfs/hhaa009","date":"2020-02-26","kind":"paper","linkCheck":"ok-crossref"},{"title":"Can ChatGPT Forecast Stock Price Movements? Return Predictability and Large Language Models","url":"https://arxiv.org/abs/2304.07619","date":"2023-04-15","kind":"paper","linkCheck":"ok-crossref"},{"title":"The Virtue of Complexity in Return Prediction","url":"https://doi.org/10.1111/jofi.13298","date":"2023-12-21","kind":"paper","linkCheck":"ok-crossref"},{"title":"Deep Learning in Asset Pricing","url":"https://doi.org/10.1287/mnsc.2023.4695","date":"2024-02","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"BIS","url":"https://www.bis.org"}],"news":[{"id":"4e5dafb4986f","date":"2026-10-20","title":"AI debt worries spark surge in default insurance trading","url":"https://www.ft.com/content/6dd31584-064c-4015-bf5d-e5747799ccd5?syn-25a6b1a6=1","source":"Financial Times AI 栏目","lang":"en","kind":"news","summaryZh":"Financial Times AI 栏目于2026-10-20发布的文章，涉及AI 金融、AI 商业融资，原题《AI debt worries spark surge in default insurance trading》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"135424ceddb1","date":"2026-10-20","title":"卢比年内跌逾5%，印度央行向10亿美元级海外并购下发核查清单：不设新限额，但银行须独立核价","url":"https://wallstreetcn.com/articles/3784016","source":"华尔街见闻","lang":"zh","kind":"news","summaryZh":"华尔街见闻于2026-10-20发布的文章，涉及AI 金融、AI 法律政策、AI 商业融资，原题《卢比年内跌逾5%，印度央行向10亿美元级海外并购下发核查清单：不设新限额，但银行须独立核价》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"d620e8d03f6a","date":"2026-10-20","title":"周小川：应开展研究并做好预案 防范AI领域投资失败引发的金融风险","url":"https://www.panewslab.com/zh/articles/01a1585e-d140-76cf-86ff-f44337274ec4","source":"PANews","lang":"zh","kind":"news","summaryZh":"PANews于2026-10-20发布的文章，涉及AI 金融、AI 法律政策、AI 安全，原题《周小川：应开展研究并做好预案 防范AI领域投资失败引发的金融风险》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"c9fe1612a387","date":"2026-10-20","title":"‘The true barrier to entry is scale’: Inside the new frontier of European fintech’s AI arms race","url":"https://sifted.eu/articles/transactional-foundation-models/","source":"Sifted","lang":"en","kind":"news","summaryZh":"Sifted于2026-10-20发布的文章，涉及AI 金融、AI 商业融资，原题《‘The true barrier to entry is scale’: Inside the new frontier of European fintech’s AI arm》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"637d0f756386","date":"2026-10-20","title":"Is CoreWeave the Silicon Valley Bank of 2026?","url":"https://www.ft.com/content/4fb42dd7-a72c-4e20-9e7c-cdc9d3215159","source":"Financial Times AI 栏目","lang":"en","kind":"news","summaryZh":"Financial Times AI 栏目于2026-10-20发布的文章，涉及AI 金融，原题《Is CoreWeave the Silicon Valley Bank of 2026?》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"9a6468f18ac8","date":"2026-10-20","title":"Claude 用 53 小时逼着比特币和银行提速","url":"https://www.panewslab.com/zh/articles/01a158c8-dd51-759d-9e98-9c7810d638b6","source":"PANews","lang":"zh","kind":"news","summaryZh":"PANews于2026-10-20发布的文章，涉及AI 金融，原题《Claude 用 53 小时逼着比特币和银行提速》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"0a8f9fe31317","date":"2026-10-20","title":"Bank of England Considers Stress Testing AI’s Effect on the Financial Sector","url":"https://www.insurancejournal.com/news/international/2026/10/20/889641.htm","source":"Insurance Journal","lang":"en","kind":"news","summaryZh":"Insurance Journal于2026-10-20发布的文章，涉及AI 金融、AI 法律政策，原题《Bank of England Considers Stress Testing AI’s Effect on the Financial Sector》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"d5fc1db58cb7","date":"2026-10-20","title":"AI Tool That Hacked Korean Banks Was Used in Japan: Security Firm","url":"https://www.insurancejournal.com/news/international/2026/10/20/889658.htm","source":"Insurance Journal","lang":"en","kind":"news","summaryZh":"Insurance Journal于2026-10-20发布的文章，涉及AI 金融、AI 安全，原题《AI Tool That Hacked Korean Banks Was Used in Japan: Security Firm》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"c464cbc1d981","date":"2026-10-20","title":"CoreWeave会是2026年的硅谷银行吗？","url":"https://wallstreetcn.com/articles/3784046","source":"华尔街见闻","lang":"zh","kind":"news","summaryZh":"华尔街见闻于2026-10-20发布的文章，涉及AI 金融、AI 芯片硬件、AI 安全，原题《CoreWeave会是2026年的硅谷银行吗？》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"3e183bdf64d9","date":"2026-10-20","title":"AI Tool That Hacked Korean Banks Was Used in Japan, GMO Says","url":"https://www.bloomberg.com/news/articles/2026-10-20/ai-tool-that-hacked-korea-banks-were-used-in-japan-gmo-says","source":"Bloomberg Technology","lang":"en","kind":"news","summaryZh":"Bloomberg Technology于2026-10-20发布的文章，涉及AI 金融、AI 安全，原题《AI Tool That Hacked Korean Banks Was Used in Japan, GMO Says》，详见原文链接。","via":"collector-run-20261020","linkCheck":"from-feed"},{"id":"dd60fbdb28af","date":"2026-10-20","title":"From SEC Filings to Financial Signals: Evaluating LLMs on Stock Market Reactions and Analyst Forecasts","url":"https://arxiv.org/abs/2610.24268","source":"arXiv cs.AI","lang":"en","kind":"paper","summaryZh":"arXiv cs.AI于2026-10-20发布的论文，涉及AI 科研与调研、AI 金融、AI 商业融资，原题《From SEC Filings to Financial Signals: Evaluating LLMs on Stock Market Reactions and Analy》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"e82fcea8f7d1","date":"2026-10-20","title":"The Control Layer Behind Rule-Following Agents Trading Across Market Conditions","url":"https://arxiv.org/abs/2610.26324","source":"arXiv q-fin","lang":"en","kind":"paper","summaryZh":"arXiv q-fin于2026-10-20发布的论文，涉及AI 金融、AI 安全、AI 商业融资，原题《The Control Layer Behind Rule-Following Agents Trading Across Market Conditions》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"}],"review":{"lastReviewed":"2026-10-21","cycleDays":90,"nextReview":"2027-01-19"},"urls":{"zh":"/zh-hans/disciplines/finance/","en":"/en/disciplines/finance/"},"dimensions":{"method":{"score":3,"label":{"zh":"研究方法","en":"Research method"},"evidence":"机器学习资产定价已是实证金融主流方法之一","source":"https://doi.org/10.1093/rfs/hhaa009","weight":0.3},"discovery":{"score":2,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"机器学习显著提升收益预测（RFS 2020），但无颠覆性新理论","source":null,"weight":0.35},"adoption":{"score":3,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"FCA/英格兰银行 2026 年调查：86% 的英国金融机构使用 AI","source":"https://www.fca.org.uk/news/speeches/building-future-finance-age-ai","weight":0.25},"education":{"score":1,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到课程证据","source":null,"weight":0.1}},"compositeScore":2.45,"levelCapped":false,"levelChange":{"from":2,"to":2,"direction":"same","reason":null},"biggestBreakthrough":{"date":"2020-02-26","title":{"zh":"Gu、Kelly、Xiu：机器学习实证资产定价（RFS）","en":"Gu, Kelly & Xiu: Empirical asset pricing via ML (RFS)"},"url":"https://doi.org/10.1093/rfs/hhaa009","aiImpact":"major","fieldImpact":"major","why":{"zh":"机器学习使收益预测的夏普比率约翻倍，成为该领域被引最多的论文之一；属方法改进而非新理论","en":"ML roughly doubles Sharpe ratios; a method advance, not a new theory"}},"keyPapers":[{"title":"Empirical Asset Pricing via Machine Learning","titleZh":"Gu、Kelly、Xiu：机器学习实证资产定价（RFS）","id":"10.1093/rfs/hhaa009","url":"https://doi.org/10.1093/rfs/hhaa009","venue":"The Review of Financial Studies","date":"2020-02-26","why":{"zh":"机器学习使收益预测的夏普比率约翻倍，成为该领域被引最多的论文之一；属方法改进而非新理论","en":"ML roughly doubles Sharpe ratios; a method advance, not a new theory"},"fieldImpact":"major","aiImpact":"major"},{"title":"BloombergGPT: A Large Language Model for Finance","titleZh":"BloombergGPT 金融大模型","id":"arXiv:2303.17564","url":"https://arxiv.org/abs/2303.17564","venue":"arXiv","date":"2023-03-30","why":{"zh":"首个大型金融专用语言模型","en":"First large finance-specific LM"},"fieldImpact":"incremental","aiImpact":"major"},{"title":"Can ChatGPT Forecast Stock Price Movements? Return Predictability and Large Language Models","titleZh":"ChatGPT 能否预测股价变动？新闻标题情绪与次日收益（arXiv 2304.07619）","id":"arXiv:2304.07619","url":"https://arxiv.org/abs/2304.07619","venue":"arXiv","date":"2023-04-15","why":{"zh":"大模型新闻解读产生显著收益可预测性，引发大量跟进","en":"LLM headline reading predicts returns; heavily followed up"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"The Virtue of Complexity in Return Prediction","titleZh":"Kelly、Malamud、Zhou：“复杂性的优点”——参数远多于样本的机器学习模型反而提升收益预测（Journal of Finance）","id":"10.1111/jofi.13298","url":"https://doi.org/10.1111/jofi.13298","venue":"The Journal of Finance","date":"2023-12-21","why":{"zh":"为“大模型式”高复杂度预测在金融中的有效性给出理论与实证依据，改变了对过拟合的传统看法","en":"Theory and evidence that highly over-parameterised models improve return prediction"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Deep Learning in Asset Pricing","titleZh":"Chen、Pelger、Zhu：用深度学习估计满足无套利条件的资产定价模型（Management Science）","id":"10.1287/mnsc.2023.4695","url":"https://doi.org/10.1287/mnsc.2023.4695","venue":"Management Science","date":"2024-02","why":{"zh":"把经济约束（无套利）嵌入深度网络，是机器学习资产定价的代表作","en":"Embeds no-arbitrage in deep nets; representative ML asset-pricing work"},"fieldImpact":"major","aiImpact":"incremental"}],"latest_papers":[{"id":"arXiv:2610.29663","title":"Lineage-Aware Search Control for LLM-Based Evolutionary Factor Mining","titleZh":"面向大模型进化式因子挖掘的谱系感知搜索控制","summaryZh":"在用大模型做变异算子的进化式 alpha 因子挖掘中，引入谱系信息引导搜索，避免集中在少数相似因子。","summaryEn":"Lineage-aware search control for LLM-driven evolutionary alpha-factor mining avoids collapsing onto similar factors.","authors":["Xintong Liu","Chenkai Zhang","Yanzhao Ma","Peng Yang"],"date":"2026-10-21","url":"https://arxiv.org/abs/2610.29663","site":"arXiv","venue":null,"category":"q-fin.CP","fetchedAt":"2026-10-22T10:09:40"},{"id":"arXiv:2610.26031","title":"AuDQ: Auditable Data Quality through Agentic Rule Discovery","titleZh":"AuDQ：通过智能体规则发现实现可审计的数据质量","summaryZh":"提出与治理文本关联、可版本化和可回放的数据质量规则层，并实现其智能体规则发现核心。","summaryEn":"An auditable, governance-linked data-quality layer with agentic rule discovery for regulated tabular pipelines.","authors":["Jacopo Biggiogera","Nicolas Colombo","Krati Saxena","Bhaskarjit Sarmah"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.26031","site":"arXiv","venue":null,"category":"cs.AI","fetchedAt":"2026-10-22T10:09:40"},{"id":"arXiv:2610.25468","title":"More Mentions, Less Information? Diagnosing Identity and Alignment in Reddit Embeddings for Volatility Forecasting","titleZh":"提及越多信息越少？诊断 Reddit 嵌入在波动率预测中的身份与对齐","summaryZh":"仅用华尔街赌场版帖子的大模型嵌入预测对数已实现方差的样本外 R² 为 0.13，研究进一步诊断信息来源。","summaryEn":"LLM embeddings of WallStreetBets posts reach out-of-sample R² 0.13 for log realised variance; the study diagnoses its source.","authors":["Zhiyu Zheng"],"date":"2026-10-18","url":"https://arxiv.org/abs/2610.25468","site":"arXiv","venue":null,"category":"q-fin.ST","fetchedAt":"2026-10-22T10:09:40"},{"id":"arXiv:2610.18435","title":"How Well Do LLMs Perform in Peer Review? Evidence from Finance and Accounting","titleZh":"大模型做同行评审表现如何？来自金融与会计的证据","summaryZh":"用 GPT-4o 评审 2024 年 AFA 和 NBER 会议论文，得分高一个标准差的论文两年内发表在三大金融期刊的可能性高 44%。","summaryEn":"GPT-4o scores for 2024 AFA/NBER papers: +1 SD predicts 44% higher odds of top-three finance publication.","authors":["Hai Lu","Yichen Luo","Luping Yu"],"date":"2026-10-13","url":"https://arxiv.org/abs/2610.18435","site":"arXiv","venue":null,"category":"q-fin.GN","fetchedAt":"2026-10-22T10:09:40"},{"id":"arXiv:2610.14959","title":"AlphaRLM: Aligning Language Models for Literature-Grounded Symbolic Factor Discovery","titleZh":"AlphaRLM：让大模型基于文献发现符号化因子","summaryZh":"用强化学习训练大模型从金融文献直接构造可执行的符号化 alpha 因子。","summaryEn":"RL-trained LLMs construct executable symbolic alpha factors grounded in financial literature.","authors":["Jie Huang","Yuecheng Hong","Yao Chen","Hao Zhong"],"date":"2026-10-10","url":"https://arxiv.org/abs/2610.14959","site":"arXiv","venue":null,"category":"cs.AI","fetchedAt":"2026-10-22T10:09:40"},{"id":"arXiv:2610.12718","title":"A Distributional Framework for Behavioral Characterization and Risk Management of Large Language Models","titleZh":"大模型行为刻画与风险管理的分布框架","summaryZh":"用蒙特卡洛模拟估计大模型回答的经验分布，并通过语义干预研究分布变化，示例包括借款人分类。","summaryEn":"A distributional, Monte Carlo framework for characterising and managing LLM behavioural risk (e.g. borrower classification).","authors":["David García-Lorite","Raul Merino"],"date":"2026-10-08","url":"https://arxiv.org/abs/2610.12718","site":"arXiv","venue":null,"category":"q-fin.RM","fetchedAt":"2026-10-22T10:09:40"},{"id":"arXiv:2610.10727","title":"Deep Learning vs. Statistical Models for Multi-Horizon Price Forecasting of Second-Hand Electronics: A Systematic Benchmark","titleZh":"二手电子产品多期价格预测：深度学习与统计模型的系统基准","summaryZh":"用波兰在线市场 2022–2025 年日度挂牌数据，首次系统比较统计与深度学习模型预测二手电子产品价格。","summaryEn":"First multi-horizon benchmark of statistical vs deep-learning models for used-electronics price forecasting.","authors":["Mateusz Buczyński","Michał Woźniak","Konrad Kaczyński","Anna Wróblewska"],"date":"2026-10-07","url":"https://arxiv.org/abs/2610.10727","site":"arXiv","venue":null,"category":"q-fin.CP","fetchedAt":"2026-10-22T10:09:40"},{"id":"arXiv:2610.06947","title":"FactorBench: A Portfolio-Aware Benchmark for Automated Factor Mining","titleZh":"FactorBench：面向投资组合的自动因子挖掘基准","summaryZh":"统一比较遗传规划、强化学习、生成模型和大模型智能体等自动因子挖掘方法的泛化性、独特性和经济价值。","summaryEn":"A portfolio-aware benchmark comparing automated factor-mining paradigms on generalisation, distinctness and economic value.","authors":["Zhuohan Wang","Carmine Ventre"],"date":"2026-10-03","url":"https://arxiv.org/abs/2610.06947","site":"arXiv","venue":null,"category":"q-fin.PM","fetchedAt":"2026-10-22T10:09:40"}]},{"id":"psychology","slug":"psychology","category":"social","categoryName":{"zh":"社会科学","en":"Social sciences"},"name":{"zh":"心理学","en":"Psychology"},"classification":{"oecdFos":"5.1","moe":"0402 心理学"},"level":2,"levelKey":"breakthrough","levelLabel":{"zh":"局部突破","en":"Local breakthroughs"},"trend":"flat","reason":{"zh":"Centaur 用 1000 多万条人类选择数据微调大模型，预测行为优于传统认知模型（Nature 2025），但多篇评估指出其生成行为常偏离真人；与 AI 对话可持久降低阴谋论信念（Science 2024）。","en":"Centaur, fine-tuned on 10M+ human choices, predicts behaviour better than classic cognitive models (Nature 2025), though evaluations show divergence from humans; AI dialogues durably reduced conspiracy beliefs (Science 2024)."},"reasonSources":[{"url":"https://www.nature.com/articles/s41586-025-09215-4","linkCheck":"ok"},{"url":"https://arxiv.org/abs/2508.07887","linkCheck":"ok"},{"url":"https://www.science.org/doi/10.1126/science.adq1814","linkCheck":"ok-crossref"}],"directions":{"zh":["认知基础模型","AI 心理治疗评估","大模型被试","AI 劝说干预"],"en":["Cognitive foundation models","AI therapy evaluation","LLM participants","AI persuasion interventions"]},"stats":{"milestones":9,"milestonesSince2025":7,"milestones2026":4,"milestonesLast6Months":2,"news":5,"latestProgress":"2026-10-19","latestMilestone":"2026-10-16","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":0,"leaps":{"ai":0,"field":0},"latestPapers":8,"keyPapers":9},"milestones":[{"date":"2023-02-02","title":{"zh":"用认知心理学方法研究 GPT-3（PNAS，Binz & Schulz）","en":"Using cognitive psychology to understand GPT-3 (PNAS)"},"url":"https://doi.org/10.1073/pnas.2218523120","source":"PNAS","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"开启把大模型当作“被试”研究的方向","en":"Opened the “LLM as participant” line"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"开启把大模型当作“被试”研究的方向","en":"Opened the “LLM as participant” line"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"开启把大模型当作“被试”研究的方向","en":"Opened the “LLM as participant” line"}},"impactNotes":[],"verified":{"id":"10.1073/pnas.2218523120","title":"Using cognitive psychology to understand GPT-3","venue":"Proceedings of the National Academy of Sciences","date":"2023-02-02"},"addedIn":"v2"},{"date":"2024-09-13","title":{"zh":"与 AI 对话持久降低阴谋论信念（Science）","en":"Dialogues with AI durably reduce conspiracy beliefs (Science)"},"url":"https://www.science.org/doi/10.1126/science.adq1814","source":"Science","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"与 GPT-4 对话使阴谋论信念降低约 20% 且持续两个月","en":"~20% belief reduction lasting two months"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"与 GPT-4 对话使阴谋论信念降低约 20% 且持续两个月","en":"~20% belief reduction lasting two months"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"与 GPT-4 对话使阴谋论信念降低约 20% 且持续两个月","en":"~20% belief reduction lasting two months"}},"impactNotes":[],"verified":{"id":"10.1126/science.adq1814","title":"Durably reducing conspiracy beliefs through dialogues with AI","venue":"Science","date":"2024-09-13"},"addedIn":"v1"},{"date":"2025-03-27","title":{"zh":"Therabot 随机试验：生成式 AI 治疗抑郁、焦虑、进食障碍症状（NEJM AI）","en":"Therabot RCT for depression, anxiety and eating-disorder symptoms (NEJM AI)"},"url":"https://doi.org/10.1056/aioa2400802","source":"NEJM AI","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"clinical","typeLabel":{"zh":"临床","en":"Clinical"},"why":{"zh":"首个全生成式心理治疗聊天机器人的随机对照试验","en":"First RCT of a fully generative therapy chatbot"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首个全生成式心理治疗聊天机器人的随机对照试验","en":"First RCT of a fully generative therapy chatbot"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首个全生成式心理治疗聊天机器人的随机对照试验","en":"First RCT of a fully generative therapy chatbot"}},"impactNotes":[],"verified":{"id":"10.1056/aioa2400802","title":"Randomized Trial of a Generative AI Chatbot for Mental Health Treatment","venue":"NEJM AI","date":"2025-03-27"},"addedIn":"v2"},{"date":"2025-03-27","title":{"zh":"Therabot 随机对照试验：生成式 AI 聊天机器人显著缓解抑郁、焦虑症状（NEJM AI）","en":"Therabot RCT: generative AI chatbot for mental health treatment (NEJM AI)"},"url":"https://doi.org/10.1056/AIoa2400802","source":"NEJM AI","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"首个生成式 AI 心理治疗的随机对照试验","en":"First RCT of a generative-AI therapy chatbot"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"首个生成式 AI 心理治疗的随机对照试验","en":"First RCT of a generative-AI therapy chatbot"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首个生成式 AI 心理治疗的随机对照试验","en":"First RCT of a generative-AI therapy chatbot"}},"impactNotes":[],"verified":{"id":"10.1056/AIoa2400802","title":"Randomized Trial of a Generative AI Chatbot for Mental Health Treatment","venue":"NEJM AI","date":"2025-03-27"},"addedIn":"v2"},{"date":"2025-07-02","title":{"zh":"Centaur：预测并刻画人类认知的基础模型（Nature）","en":"Centaur: a foundation model of human cognition (Nature)"},"url":"https://www.nature.com/articles/s41586-025-09215-4","source":"Nature","flags":["contested"],"flagLabels":[{"zh":"有争议","en":"Contested"}],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"用 1000 多万条选择数据训练的认知基础模型；生成行为偏离真人，有争议","en":"Cognitive foundation model; disputed"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"用 1000 多万条选择数据训练的认知基础模型；生成行为偏离真人，有争议","en":"Cognitive foundation model; disputed"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"用 1000 多万条选择数据训练的认知基础模型；生成行为偏离真人，有争议","en":"Cognitive foundation model; disputed"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-025-09215-4","title":"A foundation model to predict and capture human cognition","venue":"Nature","date":"2025-07-02"},"addedIn":"v1"},{"date":"2026-03-12","title":{"zh":"大模型心理治疗“认知层”架构：227 人双盲评估、近 2 万份真实对话（Nature Medicine）","en":"Cognitive-layer architecture for LLM psychotherapy: 227-participant double-blind evaluation (Nature Medicine)"},"url":"https://doi.org/10.1038/s41591-026-04278-w","source":"Nature Medicine","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"clinical","typeLabel":{"zh":"临床","en":"Clinical"},"why":{"zh":"把心理治疗 AI 的评估推进到双盲和真实世界规模","en":"Double-blind, real-world-scale evaluation"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"把心理治疗 AI 的评估推进到双盲和真实世界规模","en":"Double-blind, real-world-scale evaluation"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"把心理治疗 AI 的评估推进到双盲和真实世界规模","en":"Double-blind, real-world-scale evaluation"}},"impactNotes":[],"verified":{"id":"10.1038/s41591-026-04278-w","title":"A cognitive layer architecture to support large-language model performance in psychotherapy interactions","venue":"Nature Medicine","date":"2026-03-12"},"addedIn":"v2"},{"date":"2026-04-14","title":{"zh":"JAMA Network Open 随机试验（995 名大学生）：对话式 AI 心理干预在焦虑、抑郁和幸福感上优于团体治疗与候补对照","en":"JAMA Netw Open RCT (n=995): conversational AI agent improves psychiatric symptoms vs group therapy and waitlist"},"url":"https://doi.org/10.1001/jamanetworkopen.2026.6713","source":"JAMA Network Open","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"clinical","typeLabel":{"zh":"临床","en":"Clinical"},"why":{"zh":"大样本 RCT 显示 AI 干预与传统团体治疗相当或更优，但需长期随访","en":"Large RCT; long-term follow-up needed"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"大样本 RCT 显示 AI 干预与传统团体治疗相当或更优，但需长期随访","en":"Large RCT; long-term follow-up needed"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"大样本 RCT 显示 AI 干预与传统团体治疗相当或更优，但需长期随访","en":"Large RCT; long-term follow-up needed"}},"impactNotes":[],"verified":{"id":"10.1001/jamanetworkopen.2026.6713","title":"Efficacy of a Conversational AI Agent for Psychiatric Symptoms and Digital Therapeutic Alliance","venue":"JAMA Network Open","date":"2026-04-14"},"addedIn":"v2"},{"date":"2026-06-09","title":{"zh":"Nature Human Behaviour 发布行为科学中使用大语言模型的报告清单（共识声明）","en":"A reporting checklist for large language models in behavioural science (Nat. Hum. Behav.)"},"url":"https://www.nature.com/articles/s41562-026-02492-7","source":"Nature Human Behaviour","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"policy","typeLabel":{"zh":"政策","en":"Policy"},"why":{"zh":"AI：规范模型版本、提示词与参数等报告要素；学科：心理与行为科学首个面向 LLM 研究的共识报告标准","en":"AI: standardises reporting of model versions, prompts and settings; Field: first consensus reporting standard for LLM research in behavioural science"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：规范模型版本、提示词与参数等报告要素；学科：心理与行为科学首个面向 LLM 研究的共识报告标准","en":"AI: standardises reporting of model versions, prompts and settings; Field: first consensus reporting standard for LLM research in behavioural science"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：规范模型版本、提示词与参数等报告要素；学科：心理与行为科学首个面向 LLM 研究的共识报告标准","en":"AI: standardises reporting of model versions, prompts and settings; Field: first consensus reporting standard for LLM research in behavioural science"}},"impactNotes":[],"verified":{"id":"10.1038/s41562-026-02492-7","title":"A reporting checklist for large language models in behavioural science","venue":"Nature Human Behaviour","date":"2026-06-09"},"addedIn":"v2"},{"date":"2026-10-16","title":{"zh":"六臂随机对照试验（263 名社交焦虑障碍患者）：网络 CBT 中 AI 生成反馈与心理师反馈的症状改善无显著差异（Internet Interventions）","en":"Six-arm RCT: AI-generated vs therapist feedback in internet CBT for social anxiety (Internet Interventions)"},"url":"https://doi.org/10.1016/j.invent.2026.101012","source":"Internet Interventions","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"clinical","typeLabel":{"zh":"临床","en":"Clinical"},"why":{"zh":"AI：经心理师审核的 AI 反馈在疗效上未见劣势；学科：但未能证明非劣效，仍需更大样本","en":"AI: psychologist-reviewed AI feedback showed no efficacy disadvantage; Field: non-inferiority not established"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：经心理师审核的 AI 反馈在疗效上未见劣势；学科：但未能证明非劣效，仍需更大样本","en":"AI: psychologist-reviewed AI feedback showed no efficacy disadvantage; Field: non-inferiority not established"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：经心理师审核的 AI 反馈在疗效上未见劣势；学科：但未能证明非劣效，仍需更大样本","en":"AI: psychologist-reviewed AI feedback showed no efficacy disadvantage; Field: non-inferiority not established"}},"impactNotes":[],"verified":{"id":"10.1016/j.invent.2026.101012","title":"AI-generated vs. human therapist feedback in internet-based CBT for social anxiety disorder: A six-arm randomized controlled trial with deception design","venue":"Internet Interventions","date":"2026-10-16"},"addedIn":"v2"}],"papersRepos":[{"title":"Using cognitive psychology to understand GPT-3","url":"https://doi.org/10.1073/pnas.2218523120","date":"2023-02-02","kind":"paper","linkCheck":"ok-crossref"},{"title":"Durably reducing conspiracy beliefs through dialogues with AI","url":"https://doi.org/10.1126/science.adq1814","date":"2024-09-13","kind":"paper","linkCheck":"ok-crossref"},{"title":"Randomized Trial of a Generative AI Chatbot for Mental Health Treatment","url":"https://doi.org/10.1056/aioa2400802","date":"2025-03-27","kind":"paper","linkCheck":"ok-crossref"},{"title":"Randomized Trial of a Generative AI Chatbot for Mental Health Treatment","url":"https://doi.org/10.1056/AIoa2400802","date":"2025-03-27","kind":"paper","linkCheck":"ok-crossref"},{"title":"A foundation model to predict and capture human cognition","url":"https://doi.org/10.1038/s41586-025-09215-4","date":"2025-07-02","kind":"paper","linkCheck":"ok-crossref"},{"title":"A cognitive layer architecture to support large-language model performance in psychotherapy interactions","url":"https://doi.org/10.1038/s41591-026-04278-w","date":"2026-03-12","kind":"paper","linkCheck":"ok-crossref"},{"title":"Efficacy of a Conversational AI Agent for Psychiatric Symptoms and Digital Therapeutic Alliance","url":"https://doi.org/10.1001/jamanetworkopen.2026.6713","date":"2026-04-14","kind":"paper","linkCheck":"ok-crossref"},{"title":"A reporting checklist for large language models in behavioural science","url":"https://doi.org/10.1038/s41562-026-02492-7","date":"2026-06-09","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"Helmholtz Munich","url":"https://www.helmholtz-munich.de"},{"name":"APA","url":"https://www.apa.org"}],"news":[{"id":"c036ce2758bd","date":"2026-10-20","title":"Pluribus News: Advocates seek state guardrails as seniors turn to AI companions","url":"https://epic.org/pluribus-news-advocates-seek-state-guardrails-as-seniors-turn-to-ai-companions/","source":"EPIC","lang":"en","kind":"news","summaryZh":"美国电子隐私信息中心（EPIC）转载报道：越来越多老年人使用 AI 陪伴产品，倡导团体呼吁各州立法设立保护措施。","via":"pool-v2-20261022","linkCheck":"ok"},{"id":"beee2458b080","date":"2026-10-19","title":"AI chatbots could help mental health but also reinforce harmful thinking","url":"https://techxplore.com/news/2026-10-ai-chatbots-mental-health.html","source":"Phys.org 机器学习与AI","lang":"en","kind":"news","summaryZh":"Phys.org 机器学习与AI于2026-10-19发布的文章，涉及AI 科研与调研、AI 教育，原题《AI chatbots could help mental health but also reinforce harmful thinking》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"6da560af8407","date":"2026-10-09","title":"Role of the Digital Mental Health Technology (DMHT) Working Group","url":"https://www.gov.uk/government/publications/role-of-the-digital-mental-health-technology-dmht-working-group","source":"GOV.UK","lang":"en","kind":"news","summaryZh":"英国政府公布“数字心理健康技术工作组”的职责说明，涉及 AI 等数字心理健康产品的监管。","via":"pool-v2-20261021","linkCheck":"from-feed"},{"id":"56503e6a7c8e","date":"2026-10-08","title":"Partnership on AI Launches New Area of Work Focused on AI and Psychological Well-Being","url":"https://partnershiponai.org/partnership-on-ai-launches-new-area-of-work-focused-on-ai-and-psychological-well-being/","source":"Partnership on AI","lang":"en","kind":"news","summaryZh":"Partnership on AI 新设“AI 与心理健康”工作方向。","via":"pool-v2-20261021","linkCheck":"from-feed"},{"id":"7b6613eda154","date":"2026-10-08","title":"Partnership on AI Welcomes Eight New Partners Advancing Mental Health and Well-being","url":"https://partnershiponai.org/partnership-on-ai-welcomes-eight-new-partners-advancing-mental-health-and-well-being/","source":"Partnership on AI","lang":"en","kind":"news","summaryZh":"Partnership on AI 新增 8 家心理健康领域合作伙伴。","via":"pool-v2-20261021","linkCheck":"from-feed"}],"review":{"lastReviewed":"2026-10-21","cycleDays":90,"nextReview":"2027-01-19"},"urls":{"zh":"/zh-hans/disciplines/psychology/","en":"/en/disciplines/psychology/"},"dimensions":{"method":{"score":2,"label":{"zh":"研究方法","en":"Research method"},"evidence":"大模型被用作认知模型、实验被试和干预工具","source":"https://doi.org/10.1073/pnas.2218523120","weight":0.3},"discovery":{"score":2,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"AI 对话持久降低阴谋论信念（Science 2024）；Centaur 预测行为（Nature 2025，有争议）","source":"https://www.science.org/doi/10.1126/science.adq1814","weight":0.35},"adoption":{"score":1,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"临床心理 AI 仍在试验阶段，FDA 尚未批准生成式 AI 治疗产品","source":"https://www.fda.gov/media/189391/download","weight":0.25},"education":{"score":1,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到课程证据","source":null,"weight":0.1}},"compositeScore":1.65,"levelCapped":false,"levelChange":{"from":2,"to":2,"direction":"same","reason":null},"biggestBreakthrough":{"date":"2025-07-02","title":{"zh":"Centaur：预测并刻画人类认知的基础模型（Nature）","en":"Centaur: a foundation model of human cognition (Nature)"},"url":"https://www.nature.com/articles/s41586-025-09215-4","aiImpact":"major","fieldImpact":"major","why":{"zh":"用 1000 多万条选择数据训练的认知基础模型；生成行为偏离真人，有争议","en":"Cognitive foundation model; disputed"}},"keyPapers":[{"title":"Using cognitive psychology to understand GPT-3","titleZh":"用认知心理学方法研究 GPT-3（PNAS，Binz & Schulz）","id":"10.1073/pnas.2218523120","url":"https://doi.org/10.1073/pnas.2218523120","venue":"Proceedings of the National Academy of Sciences","date":"2023-02-02","why":{"zh":"开启把大模型当作“被试”研究的方向","en":"Opened the “LLM as participant” line"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Durably reducing conspiracy beliefs through dialogues with AI","titleZh":"与 AI 对话持久降低阴谋论信念（Science）","id":"10.1126/science.adq1814","url":"https://doi.org/10.1126/science.adq1814","venue":"Science","date":"2024-09-13","why":{"zh":"与 GPT-4 对话使阴谋论信念降低约 20% 且持续两个月","en":"~20% belief reduction lasting two months"},"fieldImpact":"major","aiImpact":"major"},{"title":"Randomized Trial of a Generative AI Chatbot for Mental Health Treatment","titleZh":"Therabot 随机试验：生成式 AI 治疗抑郁、焦虑、进食障碍症状（NEJM AI）","id":"10.1056/aioa2400802","url":"https://doi.org/10.1056/aioa2400802","venue":"NEJM AI","date":"2025-03-27","why":{"zh":"首个全生成式心理治疗聊天机器人的随机对照试验","en":"First RCT of a fully generative therapy chatbot"},"fieldImpact":"major","aiImpact":"major"},{"title":"Randomized Trial of a Generative AI Chatbot for Mental Health Treatment","titleZh":"Therabot 随机对照试验：生成式 AI 聊天机器人显著缓解抑郁、焦虑症状（NEJM AI）","id":"10.1056/AIoa2400802","url":"https://doi.org/10.1056/AIoa2400802","venue":"NEJM AI","date":"2025-03-27","why":{"zh":"首个生成式 AI 心理治疗的随机对照试验","en":"First RCT of a generative-AI therapy chatbot"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"A foundation model to predict and capture human cognition","titleZh":"Centaur：预测并刻画人类认知的基础模型（Nature）","id":"10.1038/s41586-025-09215-4","url":"https://doi.org/10.1038/s41586-025-09215-4","venue":"Nature","date":"2025-07-02","why":{"zh":"用 1000 多万条选择数据训练的认知基础模型；生成行为偏离真人，有争议","en":"Cognitive foundation model; disputed"},"fieldImpact":"major","aiImpact":"major"},{"title":"A cognitive layer architecture to support large-language model performance in psychotherapy interactions","titleZh":"大模型心理治疗“认知层”架构：227 人双盲评估、近 2 万份真实对话（Nature Medicine）","id":"10.1038/s41591-026-04278-w","url":"https://doi.org/10.1038/s41591-026-04278-w","venue":"Nature Medicine","date":"2026-03-12","why":{"zh":"把心理治疗 AI 的评估推进到双盲和真实世界规模","en":"Double-blind, real-world-scale evaluation"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Efficacy of a Conversational AI Agent for Psychiatric Symptoms and Digital Therapeutic Alliance","titleZh":"JAMA Network Open 随机试验（995 名大学生）：对话式 AI 心理干预在焦虑、抑郁和幸福感上优于团体治疗与候补对照","id":"10.1001/jamanetworkopen.2026.6713","url":"https://doi.org/10.1001/jamanetworkopen.2026.6713","venue":"JAMA Network Open","date":"2026-04-14","why":{"zh":"大样本 RCT 显示 AI 干预与传统团体治疗相当或更优，但需长期随访","en":"Large RCT; long-term follow-up needed"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"A reporting checklist for large language models in behavioural science","titleZh":"Nature Human Behaviour 发布行为科学中使用大语言模型的报告清单（共识声明）","id":"10.1038/s41562-026-02492-7","url":"https://doi.org/10.1038/s41562-026-02492-7","venue":"Nature Human Behaviour","date":"2026-06-09","why":{"zh":"AI：规范模型版本、提示词与参数等报告要素；学科：心理与行为科学首个面向 LLM 研究的共识报告标准","en":"AI: standardises reporting of model versions, prompts and settings; Field: first consensus reporting standard for LLM research in behavioural science"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"AI-generated vs. human therapist feedback in internet-based CBT for social anxiety disorder: A six-arm randomized controlled trial with deception design","titleZh":"六臂随机对照试验（263 名社交焦虑障碍患者）：网络 CBT 中 AI 生成反馈与心理师反馈的症状改善无显著差异（Internet Interventions）","id":"10.1016/j.invent.2026.101012","url":"https://doi.org/10.1016/j.invent.2026.101012","venue":"Internet Interventions","date":"2026-10-16","why":{"zh":"AI：经心理师审核的 AI 反馈在疗效上未见劣势；学科：但未能证明非劣效，仍需更大样本","en":"AI: psychologist-reviewed AI feedback showed no efficacy disadvantage; Field: non-inferiority not established"},"fieldImpact":"incremental","aiImpact":"incremental"}],"latest_papers":[{"id":"arXiv:2610.29294","title":"Same Probabilities, Different Advice: Gain-Loss Framing Effects in High-Stakes LLM Decision Support","titleZh":"概率相同、建议不同：高风险大模型决策支持中的得失框架效应","summaryZh":"构建 FrameBench-HPS 基准，检验同一概率以“得”或“失”表述时，大模型给出的高风险决策建议是否改变。","summaryEn":"FrameBench-HPS tests whether LLM high-stakes advice shifts when the same probability is framed as gain or loss.","authors":["Yuyao Zhang","Yakun Cui","Han Zhu","Jinsong Lu"],"date":"2026-10-21","url":"https://arxiv.org/abs/2610.29294","site":"arXiv","venue":null,"category":"cs.AI","fetchedAt":"2026-10-22T10:09:44"},{"id":"arXiv:2610.28772","title":"Scaling Therapy Supervision: Assessing Psychotherapy Fidelity Using Large Language Models","titleZh":"扩展心理治疗督导：用大模型评估治疗忠实度","summaryZh":"用大模型自动评估治疗师对治疗方案的遵循程度并生成可操作反馈，缓解依赖专家人工评审的瓶颈。","summaryEn":"Uses LLMs to automate psychotherapy fidelity assessment and generate actionable feedback for clinicians.","authors":["Arvind Pillai","Alexis Roach","Jason Holden","Varsha D Badal"],"date":"2026-10-20","url":"https://arxiv.org/abs/2610.28772","site":"arXiv","venue":null,"category":"cs.CL","fetchedAt":"2026-10-22T10:09:44"},{"id":"arXiv:2610.28105","title":"Reading Too Much Into It? Interpretive Preferences and Commitment in LLM Mentalisation under Ambiguity","titleZh":"想得太多？模糊情境下大模型“心智化”的解读偏好与坚持程度","summaryZh":"构建 604 个模糊社交情境并配以抑郁、社交焦虑表述，测试 8 个大模型：用户背景会改变模型的解读倾向，重复追问会让模型在证据不变时更加确定。","summaryEn":"MentalLoom shows user distress disclosures and repeated questioning shift LLM interpretations and certainty without new evidence.","authors":["Stephanie Fong","Yiwen Jiang","Qingyang Xu","Zimu Wang"],"date":"2026-10-20","url":"https://arxiv.org/abs/2610.28105","site":"arXiv","venue":null,"category":"cs.CL","fetchedAt":"2026-10-22T10:09:44"},{"id":"arXiv:2610.28092","title":"BAIT: Can LLMs Find What Biases a Reader? A Benchmark for Cognitive Bias Triggers in News","titleZh":"BAIT：大模型能找出新闻中触发读者认知偏差的语言特征吗？","summaryZh":"提出 9 类认知偏差触发器的标注规范和 50 篇人工标注新闻，测试 6 个前沿模型：能接近人类判断文章含哪些触发器，但很少完整定位。","summaryEn":"A benchmark of cognitive-bias triggers in news; LLMs approach human agreement on presence but miss spans.","authors":["Yousra Fettach","Tijl De Bie"],"date":"2026-10-20","url":"https://arxiv.org/abs/2610.28092","site":"arXiv","venue":null,"category":"cs.CL","fetchedAt":"2026-10-22T10:09:44"},{"id":"arXiv:2610.27928","title":"Why Do LLMs Accept Generics? Explaining Acceptability using Psycholinguistic and Model-Derived Factors","titleZh":"大模型为何接受泛指句？用心理语言学因素与模型内部因素解释","summaryZh":"12 个模型像人一样权衡属性的普遍性与区分性，但忽视“醒目”属性，最看重陈述是否属实；沿各因素方向调整隐状态可改变判断。","summaryEn":"LLMs weigh prevalence and distinctiveness like humans but rely most on factual truth when judging generics.","authors":["Mervat Abassy","Zhaoxin Feng","Tatsuki Kuribayashi","Kentaro Inui"],"date":"2026-10-20","url":"https://arxiv.org/abs/2610.27928","site":"arXiv","venue":null,"category":"cs.CL","fetchedAt":"2026-10-22T10:09:44"},{"id":"arXiv:2610.25596","title":"Cognitive Map in Large Language Models","titleZh":"大模型中的认知地图","summaryZh":"通过关系推理任务和操纵内部激活的“马赛克实验”发现，百亿参数以上的模型能推断未明说的关系，形成类似认知地图的抽象表示。","summaryEn":"Models above ~10B parameters infer unstated relations and form abstract, cognitive-map-like relational representations.","authors":["Albert Qin","Jonathan Cohen","Gautam Reddy"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.25596","site":"arXiv","venue":null,"category":"cs.CL","fetchedAt":"2026-10-22T10:09:44"},{"id":"doi:10.3389/fpsyg.2026.1877044","title":"Switching intention toward conversational AI for health information seeking: a push–pull–mooring analysis","titleZh":"用户从社交媒体转向对话式 AI 获取健康信息的意向：推-拉-锚定分析","summaryZh":"基于 2876 份问卷，用推-拉-锚定框架解释用户为何从社交媒体转向对话式 AI 获取健康信息。","summaryEn":"Survey of 2,876 users explains switching from social media to conversational AI for health information.","authors":["Zhaoqi Zhou","Yan Lin","Quanzhi Chen","Yanting Wu"],"date":"2026-10-19","url":"https://doi.org/10.3389/fpsyg.2026.1877044","site":"OpenAlex","venue":"Frontiers in Psychology","category":null,"fetchedAt":"2026-10-22T10:09:44"},{"id":"doi:10.1080/15614263.2026.2741740","title":"Reporting crime to police via human and AI live chats: the salience of procedural justice and outcome favourability","titleZh":"通过人工与 AI 在线聊天向警方报案：程序正义与结果有利性的作用","summaryZh":"在线实验（N=648）比较人们如何评价由人工或 AI 处理的报案聊天，考察程序正义与结果对公平感的影响。","summaryEn":"An online experiment (N=648) compares perceived fairness of crime reporting via human vs AI live chat.","authors":["Ben Bradford","Arabella Kyprianides","Helen Wells","Elizabeth Aston"],"date":"2026-10-18","url":"https://doi.org/10.1080/15614263.2026.2741740","site":"OpenAlex","venue":"Police Practice and Research","category":null,"fetchedAt":"2026-10-22T10:09:44"}]},{"id":"sociology-politics","slug":"sociology-politics","category":"social","categoryName":{"zh":"社会科学","en":"Social sciences"},"name":{"zh":"社会学与政治学","en":"Sociology & political science"},"classification":{"oecdFos":"5.4","moe":"0303 社会学 / 0302 政治学"},"level":1,"levelKey":"assistive","levelLabel":{"zh":"辅助起步","en":"Assistive"},"trend":"flat","reason":{"zh":"“硅样本”（用大模型模拟调查人群，Political Analysis 2023）引发方法论热议，但可靠性存疑（Nat. Hum. Behav. 2026）；主要用途仍是文本编码、调查辅助。","en":"“Silicon samples” (LLM-simulated respondents, Political Analysis 2023) spark debate but reliability is doubtful (Nat. Hum. Behav. 2026); main uses are text coding and survey support."},"reasonSources":[{"url":"https://doi.org/10.1017/pan.2023.2","linkCheck":"ok"},{"url":"https://www.nature.com/articles/s41562-026-02611-4","linkCheck":"ok"}],"directions":{"zh":["硅样本与社会模拟","大规模文本编码","舆论研究"],"en":["Silicon samples & simulation","Large-scale text coding","Public opinion"]},"stats":{"milestones":7,"milestonesSince2025":3,"milestones2026":2,"milestonesLast6Months":2,"news":5,"latestProgress":"2026-10-20","latestMilestone":"2026-10-12","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":0,"leaps":{"ai":0,"field":0},"latestPapers":8,"keyPapers":7},"milestones":[{"date":"2020-03-30","title":{"zh":"Salganik 等 160 支团队的“脆弱家庭挑战”：机器学习对人生结果的预测力很低（PNAS）","en":"Salganik et al.: Measuring the predictability of life outcomes with a scientific mass collaboration (PNAS)"},"url":"https://doi.org/10.1073/pnas.1915006117","source":"PNAS","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"重要反证：即使用最好的机器学习，社会结果仍难以预测","en":"Key negative result: ML barely predicts life outcomes"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"重要反证：即使用最好的机器学习，社会结果仍难以预测","en":"Key negative result: ML barely predicts life outcomes"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"重要反证：即使用最好的机器学习，社会结果仍难以预测","en":"Key negative result: ML barely predicts life outcomes"}},"impactNotes":[],"verified":{"id":"10.1073/pnas.1915006117","title":"Measuring the predictability of life outcomes with a scientific mass collaboration","venue":"Proceedings of the National Academy of Sciences","date":"2020-03-30"},"addedIn":"v2"},{"date":"2023-02-21","title":{"zh":"“Out of One, Many”：用语言模型模拟人类样本（Political Analysis）","en":"“Out of One, Many”: LMs simulate human samples (Political Analysis)"},"url":"https://doi.org/10.1017/pan.2023.2","source":"Political Analysis","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"提出用大模型模拟人群调查的“硅样本”","en":"Introduced LLM “silicon samples”"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"提出用大模型模拟人群调查的“硅样本”","en":"Introduced LLM “silicon samples”"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"提出用大模型模拟人群调查的“硅样本”","en":"Introduced LLM “silicon samples”"}},"impactNotes":[],"verified":{"id":"10.1017/pan.2023.2","title":"Out of One, Many: Using Language Models to Simulate Human Samples","venue":"Political Analysis","date":"2023-02-21"},"addedIn":"v1"},{"date":"2024-03-01","title":{"zh":"大模型能否改造计算社会科学？（Computational Linguistics，Ziems 等）","en":"Can LLMs transform computational social science? (Computational Linguistics)"},"url":"https://doi.org/10.1162/coli_a_00502","source":"Computational Linguistics","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"系统评估大模型在 25 项社会科学编码任务上的表现","en":"Benchmarks LLMs on 25 social-science coding tasks"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"系统评估大模型在 25 项社会科学编码任务上的表现","en":"Benchmarks LLMs on 25 social-science coding tasks"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"系统评估大模型在 25 项社会科学编码任务上的表现","en":"Benchmarks LLMs on 25 social-science coding tasks"}},"impactNotes":[],"verified":{"id":"10.1162/coli_a_00502","title":"Can Large Language Models Transform Computational Social\n                    Science?","venue":"Computational Linguistics","date":"2024-03-01"},"addedIn":"v2"},{"date":"2024-05-09","title":{"zh":"Bail：生成式 AI 能否改进社会科学？（PNAS）","en":"Bail: Can generative AI improve social science? (PNAS)"},"url":"https://doi.org/10.1073/pnas.2314021121","source":"PNAS","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"社会学家系统讨论机遇与风险","en":"Sociologist’s systematic assessment"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"社会学家系统讨论机遇与风险","en":"Sociologist’s systematic assessment"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"社会学家系统讨论机遇与风险","en":"Sociologist’s systematic assessment"}},"impactNotes":[],"verified":{"id":"10.1073/pnas.2314021121","title":"Can Generative AI improve social science?","venue":"Proceedings of the National Academy of Sciences","date":"2024-05-09"},"addedIn":"v2"},{"date":"2025-12-04","title":{"zh":"Hackenburg 等：对话式 AI 政治说服的杠杆——19 个模型、近 7.7 万次回应，后训练使说服力提升最多 51%，但准确性下降（Science）","en":"Hackenburg et al.: The levers of political persuasion with conversational AI (Science)"},"url":"https://doi.org/10.1126/science.aea3884","source":"Science","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"迄今最大规模 AI 说服实验，直接关系选举与舆论研究","en":"Largest AI persuasion experiment to date"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"迄今最大规模 AI 说服实验，直接关系选举与舆论研究","en":"Largest AI persuasion experiment to date"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"迄今最大规模 AI 说服实验，直接关系选举与舆论研究","en":"Largest AI persuasion experiment to date"}},"impactNotes":[],"verified":{"id":"10.1126/science.aea3884","title":"The levers of political persuasion with conversational artificial intelligence","venue":"Science","date":"2025-12-04"},"addedIn":"v2"},{"date":"2026-08-05","title":{"zh":"预注册面板实验（4293 名美国选民）：AI 生成的“预先辟谣”文章降低对五种选举谣言的相信程度，效果维持一周（Royal Society Open Science）","en":"Pre-registered US panel experiment: AI-generated pre-bunking reduces belief in five election rumours, effects persist a week (R. Soc. Open Sci.)"},"url":"https://doi.org/10.1098/rsos.252226","source":"Royal Society Open Science","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"AI：由专家设计提示词后，大模型可批量生成辟谣文章，无需逐篇人工审阅；学科：为以 AI 规模化对抗选举谣言提供预注册实验证据","en":"AI: once experts design the prompt, LLMs mass-produce pre-bunks without per-article review. Field: pre-registered evidence for scalable AI counter-misinformation"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：由专家设计提示词后，大模型可批量生成辟谣文章，无需逐篇人工审阅；学科：为以 AI 规模化对抗选举谣言提供预注册实验证据","en":"AI: once experts design the prompt, LLMs mass-produce pre-bunks without per-article review. Field: pre-registered evidence for scalable AI counter-misinformation"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：由专家设计提示词后，大模型可批量生成辟谣文章，无需逐篇人工审阅；学科：为以 AI 规模化对抗选举谣言提供预注册实验证据","en":"AI: once experts design the prompt, LLMs mass-produce pre-bunks without per-article review. Field: pre-registered evidence for scalable AI counter-misinformation"}},"impactNotes":[],"verified":{"id":"10.1098/rsos.252226","title":"Towards scalable AI-assisted pre-bunking of election misinformation: evidence from a pre-registered US panel experiment","venue":"Royal Society Open Science","date":"2026-08-05"},"addedIn":"v2-20261022pm"},{"date":"2026-10-12","title":{"zh":"跨 74 国研究：大模型模拟回答只是人类偏好的弱代理（Nat. Hum. Behav.）","en":"74-country study: LLM simulated responses are weak proxies for humans (Nat. Hum. Behav.)"},"url":"https://www.nature.com/articles/s41562-026-02611-4","source":"Nature Human Behaviour","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"跨 74 国研究表明硅样本只是人类偏好的弱代理","en":"74-country study: silicon samples are weak proxies"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"跨 74 国研究表明硅样本只是人类偏好的弱代理","en":"74-country study: silicon samples are weak proxies"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"跨 74 国研究表明硅样本只是人类偏好的弱代理","en":"74-country study: silicon samples are weak proxies"}},"impactNotes":[],"verified":{"id":"10.1038/s41562-026-02611-4","title":"Simulated responses from large language models are weak proxies for human economic preferences across 74 countries","venue":"Nature Human Behaviour","date":"2026-10-12"},"addedIn":"v1"}],"papersRepos":[{"title":"Measuring the predictability of life outcomes with a scientific mass collaboration","url":"https://doi.org/10.1073/pnas.1915006117","date":"2020-03-30","kind":"paper","linkCheck":"ok-crossref"},{"title":"Out of One, Many: Using Language Models to Simulate Human Samples","url":"https://doi.org/10.1017/pan.2023.2","date":"2023-02-21","kind":"paper","linkCheck":"ok-crossref"},{"title":"Can Large Language Models Transform Computational Social\n                    Science?","url":"https://doi.org/10.1162/coli_a_00502","date":"2024-03-01","kind":"paper","linkCheck":"ok-crossref"},{"title":"Can Generative AI improve social science?","url":"https://doi.org/10.1073/pnas.2314021121","date":"2024-05-09","kind":"paper","linkCheck":"ok-crossref"},{"title":"The levers of political persuasion with conversational artificial intelligence","url":"https://doi.org/10.1126/science.aea3884","date":"2025-12-04","kind":"paper","linkCheck":"ok-crossref"},{"title":"Simulated responses from large language models are weak proxies for human economic preferences across 74 countries","url":"https://doi.org/10.1038/s41562-026-02611-4","date":"2026-10-12","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"Stanford HAI","url":"https://hai.stanford.edu"},{"name":"Oxford Internet Institute","url":"https://www.oii.ox.ac.uk"}],"news":[{"id":"5f396f1bac18","date":"2026-10-20","title":"How governments worldwide are shaping what AI can say about politics","url":"https://theconversation.com/how-governments-worldwide-are-shaping-what-ai-can-say-about-politics-294097","source":"The Conversation","lang":"en","kind":"news","summaryZh":"The Conversation：研究 12 个国家和地区后发现，政府常借国家安全或反宣传等既有法律而非新 AI 法来控制 AI 对政治话题的回答。","via":"pool-v2-20261022","linkCheck":"ok"},{"id":"c8c1817da325","date":"2026-10-20","title":"As Midterms Approach, Top American Fears Include AI Interference in the Political Process","url":"https://www.cnet.com/tech/services-and-software/midterms-top-american-fears-ai-interference-political-process/","source":"CNET","lang":"en","kind":"news","summaryZh":"CNET：年度“美国人的恐惧”调查首次加入 AI 题目，“AI 损害政治进程”排名第三，“AI 损害经济”进入前十。","via":"pool-v2-20261022","linkCheck":"ok"},{"id":"9279cf6bb0c3","date":"2026-10-19","title":"Democrats think cracking down on AI could give them an edge","url":"https://www.theverge.com/policy/1013113/democrats-ai-midterms","source":"The Verge AI","lang":"en","kind":"news","summaryZh":"The Verge AI于2026-10-19发布的文章，涉及AI 综合、AI 法律政策，原题《Democrats think cracking down on AI could give them an edge》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"2ee15e5e5538","date":"2026-10-09","title":"Democrats prepare to battle over AI regulation","url":"https://www.transformernews.ai/p/democrats-prepare-to-battle-over-ai-regulation","source":"Transformer (AI 政策)","lang":"en","kind":"news","summaryZh":"Transformer (AI 政策)于2026-10-09发布的文章，涉及AI 法律政策、AI 安全，原题《Democrats prepare to battle over AI regulation》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"8d1dc43a37f7","date":"2026-10-01","title":"Democrats want to lead on AI – but can’t agree how","url":"https://www.transformernews.ai/p/democrats-cant-agree-on-ai-midterms","source":"Transformer (AI 政策)","lang":"en","kind":"news","summaryZh":"Transformer (AI 政策)于2026-10-01发布的文章，涉及AI 法律政策，原题《Democrats want to lead on AI – but can’t agree how》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"}],"review":{"lastReviewed":"2026-10-22","cycleDays":90,"nextReview":"2027-01-19"},"urls":{"zh":"/zh-hans/disciplines/sociology-politics/","en":"/en/disciplines/sociology-politics/"},"dimensions":{"method":{"score":2,"label":{"zh":"研究方法","en":"Research method"},"evidence":"大模型文本编码、“硅样本”模拟调查成为方法热点","source":"https://doi.org/10.1073/pnas.2314021121","weight":0.3},"discovery":{"score":1,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"没有以 AI 为关键的公认新发现；硅样本可靠性被跨国研究否定","source":"https://www.nature.com/articles/s41562-026-02611-4","weight":0.35},"adoption":{"score":1,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"缺少使用率调查","source":null,"weight":0.25},"education":{"score":1,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到课程证据","source":null,"weight":0.1}},"compositeScore":1.3,"levelCapped":false,"levelChange":{"from":1,"to":1,"direction":"same","reason":null},"biggestBreakthrough":{"date":"2023-02-21","title":{"zh":"“Out of One, Many”：用语言模型模拟人类样本（Political Analysis）","en":"“Out of One, Many”: LMs simulate human samples (Political Analysis)"},"url":"https://doi.org/10.1017/pan.2023.2","aiImpact":"major","fieldImpact":"major","why":{"zh":"提出用大模型模拟人群调查的“硅样本”","en":"Introduced LLM “silicon samples”"}},"keyPapers":[{"title":"Measuring the predictability of life outcomes with a scientific mass collaboration","titleZh":"Salganik 等 160 支团队的“脆弱家庭挑战”：机器学习对人生结果的预测力很低（PNAS）","id":"10.1073/pnas.1915006117","url":"https://doi.org/10.1073/pnas.1915006117","venue":"Proceedings of the National Academy of Sciences","date":"2020-03-30","why":{"zh":"重要反证：即使用最好的机器学习，社会结果仍难以预测","en":"Key negative result: ML barely predicts life outcomes"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Out of One, Many: Using Language Models to Simulate Human Samples","titleZh":"“Out of One, Many”：用语言模型模拟人类样本（Political Analysis）","id":"10.1017/pan.2023.2","url":"https://doi.org/10.1017/pan.2023.2","venue":"Political Analysis","date":"2023-02-21","why":{"zh":"提出用大模型模拟人群调查的“硅样本”","en":"Introduced LLM “silicon samples”"},"fieldImpact":"major","aiImpact":"major"},{"title":"Can Large Language Models Transform Computational Social\n                    Science?","titleZh":"大模型能否改造计算社会科学？（Computational Linguistics，Ziems 等）","id":"10.1162/coli_a_00502","url":"https://doi.org/10.1162/coli_a_00502","venue":"Computational Linguistics","date":"2024-03-01","why":{"zh":"系统评估大模型在 25 项社会科学编码任务上的表现","en":"Benchmarks LLMs on 25 social-science coding tasks"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Can Generative AI improve social science?","titleZh":"Bail：生成式 AI 能否改进社会科学？（PNAS）","id":"10.1073/pnas.2314021121","url":"https://doi.org/10.1073/pnas.2314021121","venue":"Proceedings of the National Academy of Sciences","date":"2024-05-09","why":{"zh":"社会学家系统讨论机遇与风险","en":"Sociologist’s systematic assessment"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"The levers of political persuasion with conversational artificial intelligence","titleZh":"Hackenburg 等：对话式 AI 政治说服的杠杆——19 个模型、近 7.7 万次回应，后训练使说服力提升最多 51%，但准确性下降（Science）","id":"10.1126/science.aea3884","url":"https://doi.org/10.1126/science.aea3884","venue":"Science","date":"2025-12-04","why":{"zh":"迄今最大规模 AI 说服实验，直接关系选举与舆论研究","en":"Largest AI persuasion experiment to date"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Towards scalable AI-assisted pre-bunking of election misinformation: evidence from a pre-registered US panel experiment","titleZh":"预注册面板实验（4293 名美国选民）：AI 生成的“预先辟谣”文章降低对五种选举谣言的相信程度，效果维持一周（Royal Society Open Science）","id":"10.1098/rsos.252226","url":"https://doi.org/10.1098/rsos.252226","venue":"Royal Society Open Science","date":"2026-08-05","why":{"zh":"AI：由专家设计提示词后，大模型可批量生成辟谣文章，无需逐篇人工审阅；学科：为以 AI 规模化对抗选举谣言提供预注册实验证据","en":"AI: once experts design the prompt, LLMs mass-produce pre-bunks without per-article review. Field: pre-registered evidence for scalable AI counter-misinformation"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Simulated responses from large language models are weak proxies for human economic preferences across 74 countries","titleZh":"跨 74 国研究：大模型模拟回答只是人类偏好的弱代理（Nat. Hum. Behav.）","id":"10.1038/s41562-026-02611-4","url":"https://doi.org/10.1038/s41562-026-02611-4","venue":"Nature Human Behaviour","date":"2026-10-12","why":{"zh":"跨 74 国研究表明硅样本只是人类偏好的弱代理","en":"74-country study: silicon samples are weak proxies"},"fieldImpact":"major","aiImpact":"incremental"}],"latest_papers":[{"id":"arXiv:2610.29785","title":"Where Do LLMs Learn to Recommend AI?","titleZh":"大模型从哪里学会推荐 AI？","summaryZh":"发现大模型在升学、职业和创业建议中常偏向推荐 AI 相关选项，且指令微调后这种倾向明显增强。","summaryEn":"Finds LLMs often recommend AI-related options and that instruction tuning sharply increases this tendency.","authors":["Benaya Trabelsi","Jonathan Shaki","Sarit Kraus"],"date":"2026-10-21","url":"https://arxiv.org/abs/2610.29785","site":"arXiv","venue":null,"category":"cs.CL","fetchedAt":"2026-10-22T10:09:48"},{"id":"arXiv:2610.29304","title":"Situating Vulnerability: An Ethical Governance Framework for Anthropomorphic AI Interaction","titleZh":"情境化脆弱性：拟人化 AI 交互的伦理治理框架","summaryZh":"指出拟人化、带情感的大模型回应对不同用户可能是安慰、诱陷或强化妄想，提出按情境评估脆弱性的治理框架。","summaryEn":"An ethical governance framework for anthropomorphic AI that situates user vulnerability rather than applying uniform rules.","authors":["Hanhui Xu","Jiacheng Ji","Yuhan Wang","Haoan Jin"],"date":"2026-10-21","url":"https://arxiv.org/abs/2610.29304","site":"arXiv","venue":null,"category":"cs.CY","fetchedAt":"2026-10-22T10:09:48"},{"id":"arXiv:2610.27573","title":"When LLMs Manipulate: Measuring Behavioral, Social, and Temporal Effects","titleZh":"当大模型操纵用户：行为、社会与时间效应的测量","summaryZh":"用可控的操纵性回复（权威、紧迫、从众）开展 120 人实验，操纵性措辞显著改变信任校准与顺从，并在社交情境中传播。","summaryEn":"A 120-participant study finds manipulative LLM framing shifts trust and compliance, with effects propagating socially.","authors":["Sadia Asif","Israel Antonio Rosales Laguan","Shumaila Asif","Haris Khan"],"date":"2026-10-20","url":"https://arxiv.org/abs/2610.27573","site":"arXiv","venue":null,"category":"cs.AI","fetchedAt":"2026-10-22T10:09:48"},{"id":"arXiv:2610.26730","title":"EpiLens: Lightweight RAG for Epidemiological Knowledge Acquisition","titleZh":"EpiLens：用于获取流行病学知识的轻量检索增强工具","summaryZh":"开源 RAG 工具，对每篇论文按固定模板抽取信息并链接原文；在 2297 篇论文语料上与 214 篇人工标注论文对比评估。","summaryEn":"Open-source RAG package for mapping epidemiological literature, evaluated on 2,297 papers against 214 labels.","authors":["Vincenzo Perri","Ciro Cattuto","Daniela Paolotti"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.26730","site":"arXiv","venue":null,"category":"cs.CL","fetchedAt":"2026-10-22T10:09:48"},{"id":"arXiv:2610.26412","title":"Where Does Rhetorical Distortion Live? A Multi-Platform, Multi-Language Measurement of Influencer Content","titleZh":"修辞扭曲在哪里？对网红内容的多平台、多语言测量","summaryZh":"把“修辞扭曲”分为夸大重要性、制造焦虑、新颖性宣称、情绪化用语和时间扭曲五个维度，用分层流程在多平台测量。","summaryEn":"Measures five dimensions of rhetorical distortion in influencer content across platforms and languages.","authors":["Xucheng Yu","Jiangping Chen","Haohan Wang"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.26412","site":"arXiv","venue":null,"category":"cs.SI","fetchedAt":"2026-10-22T10:09:48"},{"id":"arXiv:2610.25298","title":"Aligning Language Models with Indigenous Community Cosmovision through Constitutional AI","titleZh":"通过宪法式 AI 让语言模型对齐原住民社区宇宙观","summaryZh":"访谈 5 位拉美原住民领袖，构建含 12 条原则的文化“宪法”，用于对齐大模型。","summaryEn":"Builds a 12-principle cultural constitution from interviews with Indigenous leaders to align LLMs.","authors":["Silvia Fernández-Sabido","Jorge Reyes-Magaña","Jesús Adrián Chay Tun","Jean Carlos Buenfil"],"date":"2026-10-18","url":"https://arxiv.org/abs/2610.25298","site":"arXiv","venue":null,"category":"cs.AI","fetchedAt":"2026-10-22T10:09:48"},{"id":"arXiv:2610.26961","title":"Do LLMs Really Lack Difference Awareness? Uncovering Heuristic Substitution in Group-Differentiating Fairness","titleZh":"大模型真的缺乏“差异意识”吗？群体区分公平中的启发式替代","summaryZh":"提出大模型并非不能识别群体差异，而是用“平等对待”启发式替代复杂判断；两阶段“先区分、再决定”框架在 4 个基准、10 个模型上改善表现。","summaryEn":"Argues LLMs substitute an equal-treatment heuristic for difference awareness and proposes a two-stage fix.","authors":["Jingshen Zhang","Bo Wang","Yanlin Fu","Yuewen Dang"],"date":"2026-10-18","url":"https://arxiv.org/abs/2610.26961","site":"arXiv","venue":null,"category":"cs.SI","fetchedAt":"2026-10-22T10:09:48"},{"id":"arXiv:2610.25040","title":"Scam Town: Can Scam Education Spread Like a Vaccine?","titleZh":"骗局小镇：反诈教育能像疫苗一样传播吗？","summaryZh":"构建由大模型扮演每位居民的模拟社区，研究老年人之间口口相传的反诈提醒能否像疫苗一样保护社区。","summaryEn":"An LLM-simulated community tests whether word-of-mouth scam warnings spread protection like a vaccine.","authors":["Simiao Ren","Tommy Duong","Xingyu Shen","Jiaqi Gan"],"date":"2026-10-18","url":"https://arxiv.org/abs/2610.25040","site":"arXiv","venue":null,"category":"cs.SI","fetchedAt":"2026-10-22T10:09:48"}]},{"id":"law","slug":"law","category":"social","categoryName":{"zh":"社会科学","en":"Social sciences"},"name":{"zh":"法学","en":"Law"},"classification":{"oecdFos":"5.5","moe":"0301 法学"},"level":1,"levelKey":"assistive","levelLabel":{"zh":"辅助起步","en":"Assistive"},"trend":"up","reason":{"zh":"法律实务广泛试用 AI，但问题突出：Charlotin 数据库收录 2202 起涉及 AI 幻觉的法庭裁决，路透社统计美国 1395 起；调查显示 60% 以上联邦法官用过 AI、仅 22.4% 每周使用。法学研究本身改造有限。","en":"Legal practice uses AI widely, with problems: Charlotin's database lists 2,202 decisions on AI hallucinations, Reuters counts 1,395 US cases; >60% of surveyed federal judges have used AI but only 22.4% weekly."},"reasonSources":[{"url":"https://www.damiencharlotin.com/hallucinations/","linkCheck":"ok"},{"url":"https://www.reuters.com/legal/government/ai-error-ridden-court-filings-surge-despite-three-years-court-sanctions-2026-09-17/","linkCheck":"site-blocks-bots"},{"url":"https://www.nycbar.org/reports/artificial-intelligence-in-federal-courts-a-random-sample-survey-of-judges/","linkCheck":"ok"}],"directions":{"zh":["法律检索与起草","合同审查","司法辅助","AI 治理法（研究对象）"],"en":["Legal research & drafting","Contract review","Judicial assistance","AI law (as subject)"]},"stats":{"milestones":7,"milestonesSince2025":4,"milestones2026":3,"milestonesLast6Months":2,"news":12,"latestProgress":"2026-10-20","latestMilestone":"2026-10-18","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":1,"leaps":{"ai":0,"field":0},"latestPapers":8,"keyPapers":3},"milestones":[{"date":"2023-06-22","title":{"zh":"Mata v. Avianca：纽约联邦法院因律师提交 ChatGPT 编造的 6 个判例处以制裁","en":"Mata v. Avianca: SDNY sanctions lawyers for six ChatGPT-fabricated cases"},"url":"https://www.reuters.com/legal/new-york-lawyers-sanctioned-using-fake-chatgpt-cases-legal-brief-2023-06-22/","source":"Reuters","flags":[],"flagLabels":[],"linkCheck":"site-blocks-bots","datePrecision":"day","type":"regulatory","typeLabel":{"zh":"监管","en":"Regulatory"},"why":{"zh":"首个广为人知的 AI 幻觉判例制裁，推动各地法院出台 AI 使用规则","en":"Landmark sanction that triggered court AI rules"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"首个广为人知的 AI 幻觉判例制裁，推动各地法院出台 AI 使用规则","en":"Landmark sanction that triggered court AI rules"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首个广为人知的 AI 幻觉判例制裁，推动各地法院出台 AI 使用规则","en":"Landmark sanction that triggered court AI rules"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2024-02-26","title":{"zh":"GPT-4 通过美国律师资格考试（Phil. Trans. R. Soc. A）","en":"GPT-4 passes the bar exam (Phil. Trans. R. Soc. A)"},"url":"https://doi.org/10.1098/rsta.2023.0254","source":"Phil. Trans. R. Soc. A","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"competition","typeLabel":{"zh":"竞赛","en":"Competition"},"why":{"zh":"GPT-4 在美国律师资格统一考试中达到及格以上","en":"GPT-4 passes the Uniform Bar Exam"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"GPT-4 在美国律师资格统一考试中达到及格以上","en":"GPT-4 passes the Uniform Bar Exam"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"GPT-4 在美国律师资格统一考试中达到及格以上","en":"GPT-4 passes the Uniform Bar Exam"}},"impactNotes":[],"verified":{"id":"10.1098/rsta.2023.0254","title":"GPT-4 passes the bar exam","venue":"Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences","date":"2024-02-26"},"addedIn":"v1"},{"date":"2024-06-26","title":{"zh":"“大型法律虚构”：大模型法律幻觉的系统测量（Journal of Legal Analysis）","en":"Large Legal Fictions: profiling legal hallucinations in LLMs (J. Legal Analysis)"},"url":"https://doi.org/10.1093/jla/laae003","source":"Journal of Legal Analysis","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"量化大模型在判例问题上的幻觉率","en":"Quantifies LLM hallucination on case-law questions"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"量化大模型在判例问题上的幻觉率","en":"Quantifies LLM hallucination on case-law questions"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"量化大模型在判例问题上的幻觉率","en":"Quantifies LLM hallucination on case-law questions"}},"impactNotes":[],"verified":{"id":"10.1093/jla/laae003","title":"Large Legal Fictions: Profiling Legal Hallucinations in Large Language Models","venue":"Journal of Legal Analysis","date":"2024-06-26"},"addedIn":"v2"},{"date":"2025-04-23","title":{"zh":"法律研究 AI 工具的幻觉：主流商业工具仍有 17%–33% 幻觉率（Journal of Empirical Legal Studies）","en":"Hallucination-free? Assessing AI legal research tools (JELS)"},"url":"https://doi.org/10.1111/jels.12413","source":"Journal of Empirical Legal Studies","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"首个对商业法律 AI 工具的独立实证评估，反驳“无幻觉”宣传","en":"First independent test of commercial legal AI tools"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"首个对商业法律 AI 工具的独立实证评估，反驳“无幻觉”宣传","en":"First independent test of commercial legal AI tools"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首个对商业法律 AI 工具的独立实证评估，反驳“无幻觉”宣传","en":"First independent test of commercial legal AI tools"}},"impactNotes":[],"verified":{"id":"10.1111/jels.12413","title":"Hallucination‐Free? Assessing the Reliability of Leading\n                    AI\n                    Legal Research Tools","venue":"Journal of Empirical Legal Studies","date":"2025-04-23"},"addedIn":"v2"},{"date":"2026-03-30","title":{"zh":"纽约市律师协会：联邦法官随机抽样调查，60% 以上用过 AI、约 22% 每日或每周使用","en":"NYC Bar random-sample survey: 60%+ of federal judges have used AI, ~22% daily or weekly"},"url":"https://www.nycbar.org/reports/artificial-intelligence-in-federal-courts-a-random-sample-survey-of-judges/","source":"New York City Bar","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"industry","typeLabel":{"zh":"产业/采用","en":"Industry"},"why":{"zh":"首个随机抽样的法官 AI 使用调查","en":"First random-sample survey of judges’ AI use"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"首个随机抽样的法官 AI 使用调查","en":"First random-sample survey of judges’ AI use"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首个随机抽样的法官 AI 使用调查","en":"First random-sample survey of judges’ AI use"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-09-17","title":{"zh":"路透社：AI 错误法庭文书激增，三年制裁未能遏制","en":"Reuters: AI error-ridden court filings surge despite sanctions"},"url":"https://www.reuters.com/legal/government/ai-error-ridden-court-filings-surge-despite-three-years-court-sanctions-2026-09-17/","source":"Reuters","flags":[],"flagLabels":[],"linkCheck":"site-blocks-bots","datePrecision":"day","type":"industry","typeLabel":{"zh":"产业/采用","en":"Industry"},"why":{"zh":"AI 错误文书在三年制裁后仍激增，反映实践风险","en":"AI-error filings keep surging"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI 错误文书在三年制裁后仍激增，反映实践风险","en":"AI-error filings keep surging"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI 错误文书在三年制裁后仍激增，反映实践风险","en":"AI-error filings keep surging"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-10-18","title":{"zh":"美国联邦上诉法院维持 Thomson Reuters 诉 Ross Intelligence 案判决：用 Westlaw 内容训练法律 AI 不构成合理使用（判决书仍密封）","en":"Appeals court upholds Thomson Reuters’ win over Ross Intelligence on AI-training fair use (opinion sealed)"},"url":"https://www.theverge.com/ai-artificial-intelligence/1013500/appeals-court-upholds-thomson-reuters-win-in-ai-training-lawsuit","source":"The Verge","flags":["tbv"],"flagLabels":[{"zh":"日期或细节待核实","en":"Date/details to be verified"}],"linkCheck":"ok","datePrecision":"day","type":"legal","typeLabel":{"zh":"诉讼/判决","en":"Legal"},"why":{"zh":"AI：训练数据合理使用抗辩在上诉审被驳回；学科：AI 训练版权纠纷获得上诉层级判决，对法律科技和其他版权诉讼有参照意义（判决细节待公开）","en":"AI: fair-use defence for training data fails on appeal; Field: an appellate-level ruling on AI training copyright (details pending)"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：训练数据合理使用抗辩在上诉审被驳回；学科：AI 训练版权纠纷获得上诉层级判决，对法律科技和其他版权诉讼有参照意义（判决细节待公开）","en":"AI: fair-use defence for training data fails on appeal; Field: an appellate-level ruling on AI training copyright (details pending)"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：训练数据合理使用抗辩在上诉审被驳回；学科：AI 训练版权纠纷获得上诉层级判决，对法律科技和其他版权诉讼有参照意义（判决细节待公开）","en":"AI: fair-use defence for training data fails on appeal; Field: an appellate-level ruling on AI training copyright (details pending)"}},"impactNotes":[],"verified":null,"addedIn":"v2"}],"papersRepos":[{"title":"GPT-4 passes the bar exam","url":"https://doi.org/10.1098/rsta.2023.0254","date":"2024-02-26","kind":"paper","linkCheck":"ok-crossref"},{"title":"Large Legal Fictions: Profiling Legal Hallucinations in Large Language Models","url":"https://doi.org/10.1093/jla/laae003","date":"2024-06-26","kind":"paper","linkCheck":"ok-crossref"},{"title":"Hallucination‐Free? Assessing the Reliability of Leading\n                    AI\n                    Legal Research Tools","url":"https://doi.org/10.1111/jels.12413","date":"2025-04-23","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"Harvey","url":"https://www.harvey.ai"},{"name":"Stanford RegLab","url":"https://reglab.stanford.edu"}],"news":[{"id":"72921a439360","date":"2026-10-20","title":"在人人都能做漫剧之后，阅文打开了自己的版权库","url":"https://36kr.com/p/4032661941883525?f=rss","source":"36氪","lang":"zh","kind":"news","summaryZh":"36氪于2026-10-20发布的文章，涉及AI 法律政策、AI 商业融资，原题《在人人都能做漫剧之后，阅文打开了自己的版权库》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"0e9d1fa2350e","date":"2026-10-20","title":"What Happens When Lawyers Can Afford to Look Further?","url":"https://www.artificiallawyer.com/2026/10/20/what-happens-when-lawyers-can-afford-to-look-further/","source":"Artificial Lawyer","lang":"en","kind":"news","summaryZh":"Artificial Lawyer于2026-10-20发布的文章，涉及AI 法律政策，原题《What Happens When Lawyers Can Afford to Look Further?》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"5a05475451ef","date":"2026-10-20","title":"More Legal Research Moves as Moonlit Launches Free Commentaries","url":"https://www.artificiallawyer.com/2026/10/20/more-legal-research-moves-as-moonlit-launches-free-commentaries/","source":"Artificial Lawyer","lang":"en","kind":"news","summaryZh":"Artificial Lawyer于2026-10-20发布的文章，涉及AI 法律政策、AI 创意媒体，原题《More Legal Research Moves as Moonlit Launches Free Commentaries》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"49643288fe24","date":"2026-10-20","title":"ICMP says AI firms are trying to ‘strong-arm’ Australia on copyright – and that Kylie Minogue’s songs have already been used to train AI (report)","url":"https://www.musicbusinessworldwide.com/icmp-says-ai-firms-are-trying-to-strong-arm-australia-on-copyright-and-that-kylie-minogues-songs-have-already-been-used-to-train-ai-report/","source":"Music Business Worldwide","lang":"en","kind":"news","summaryZh":"Music Business Worldwide于2026-10-20发布的文章，涉及AI 创意媒体、AI 法律政策，原题《ICMP says AI firms are trying to ‘strong-arm’ Australia on copyright – and that Kylie Mino》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"ffbaa317e67b","date":"2026-10-20","title":"OpenAI 与 Meta 齐发声：深度伪造将严重威胁法律诉讼中的证据可靠性","url":"https://www.ithome.com/1/015/421.htm","source":"IT之家","lang":"zh","kind":"news","summaryZh":"IT之家于2026-10-20发布的文章，涉及AI 法律政策、AI 安全，原题《OpenAI 与 Meta 齐发声：深度伪造将严重威胁法律诉讼中的证据可靠性》，详见原文链接。","via":"collector-run-20261020","linkCheck":"from-feed"},{"id":"c93fa8396a7c","date":"2026-10-19","title":"Big Law’s Greatest Gifts","url":"https://www.artificiallawyer.com/2026/10/19/big-laws-greatest-gifts/","source":"Artificial Lawyer","lang":"en","kind":"news","summaryZh":"Artificial Lawyer于2026-10-19发布的文章，涉及AI 法律政策，原题《Big Law’s Greatest Gifts》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"8fbce512feec","date":"2026-10-19","title":"Disney Asked Character.AI to Go Bankrupt for a Get-Out-of-Lawsuits Card","url":"https://www.wired.com/story/disney-asked-characterai-to-go-bankrupt-for-a-get-out-of-lawsuits-card/","source":"WIRED · AI","lang":"en","kind":"news","summaryZh":"WIRED · AI于2026-10-19发布的文章，涉及AI 法律政策、AI 商业融资、AI 安全，原题《Disney Asked Character.AI to Go Bankrupt for a Get-Out-of-Lawsuits Card》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"79910d3ca669","date":"2026-10-19","title":"Harvey Webinar – Agents in Legal: A Practical Guide","url":"https://www.artificiallawyer.com/2026/10/19/harvey-webinar-agents-in-legal-a-practical-guide/","source":"Artificial Lawyer","lang":"en","kind":"news","summaryZh":"Artificial Lawyer于2026-10-19发布的文章，涉及AI 法律政策，原题《Harvey Webinar – Agents in Legal: A Practical Guide》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"b3d41182461b","date":"2026-10-19","title":"Disney found a way to buy Character.AI without buying its lawsuits","url":"https://arstechnica.com/tech-policy/2026/10/disney-found-a-way-to-buy-character-ai-without-buying-its-lawsuits/","source":"Ars Technica AI","lang":"en","kind":"news","summaryZh":"Ars Technica AI于2026-10-19发布的文章，涉及AI 法律政策，原题《Disney found a way to buy Character.AI without buying its lawsuits》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"3f4528f1e176","date":"2026-10-19","title":"Two German courts now hold Google liable for what its AI overviews say","url":"https://the-decoder.com/two-german-courts-now-hold-google-liable-for-what-its-ai-overviews-say/","source":"The Decoder","lang":"en","kind":"news","summaryZh":"The Decoder于2026-10-19发布的文章，涉及AI 法律政策、AI 商业融资，原题《Two German courts now hold Google liable for what its AI overviews say》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"d3fd113d24a7","date":"2026-09-17","title":"AI error-ridden court filings surge despite three years of sanctions","url":"https://www.reuters.com/legal/government/ai-error-ridden-court-filings-surge-despite-three-years-court-sanctions-2026-09-17/","source":"Reuters","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"site-blocks-bots"},{"id":"976cb7a26755","date":"2026-03-30","title":"Artificial Intelligence in Federal Courts: A Random Sample Survey of Judges","url":"https://www.nycbar.org/reports/artificial-intelligence-in-federal-courts-a-random-sample-survey-of-judges/","source":"New York City Bar","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"ok"}],"review":{"lastReviewed":"2026-10-21","cycleDays":90,"nextReview":"2027-01-19"},"urls":{"zh":"/zh-hans/disciplines/law/","en":"/en/disciplines/law/"},"dimensions":{"method":{"score":1,"label":{"zh":"研究方法","en":"Research method"},"evidence":"法学研究中 AI 主要用于检索与文本分析","source":null,"weight":0.3},"discovery":{"score":0,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"没有以 AI 为关键的法学研究成果","source":null,"weight":0.35},"adoption":{"score":2,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"超过 60% 美国联邦法官用过 AI，但仅约 22% 每日或每周使用；AI 幻觉文书大量出现","source":"https://www.nycbar.org/reports/artificial-intelligence-in-federal-courts-a-random-sample-survey-of-judges/","weight":0.25},"education":{"score":2,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"ABA 发布 AI 职业伦理意见，法学院陆续开设 AI 课程","source":null,"weight":0.1}},"compositeScore":1,"levelCapped":false,"levelChange":{"from":1,"to":1,"direction":"same","reason":null},"biggestBreakthrough":{"date":"2026-10-18","title":{"zh":"美国联邦上诉法院维持 Thomson Reuters 诉 Ross Intelligence 案判决：用 Westlaw 内容训练法律 AI 不构成合理使用（判决书仍密封）","en":"Appeals court upholds Thomson Reuters’ win over Ross Intelligence on AI-training fair use (opinion sealed)"},"url":"https://www.theverge.com/ai-artificial-intelligence/1013500/appeals-court-upholds-thomson-reuters-win-in-ai-training-lawsuit","aiImpact":"incremental","fieldImpact":"major","why":{"zh":"AI：训练数据合理使用抗辩在上诉审被驳回；学科：AI 训练版权纠纷获得上诉层级判决，对法律科技和其他版权诉讼有参照意义（判决细节待公开）","en":"AI: fair-use defence for training data fails on appeal; Field: an appellate-level ruling on AI training copyright (details pending)"}},"keyPapers":[{"title":"GPT-4 passes the bar exam","titleZh":"GPT-4 通过美国律师资格考试（Phil. Trans. R. Soc. A）","id":"10.1098/rsta.2023.0254","url":"https://doi.org/10.1098/rsta.2023.0254","venue":"Philosophical Transactions of the Royal Society A: Mathematical, Physical and Engineering Sciences","date":"2024-02-26","why":{"zh":"GPT-4 在美国律师资格统一考试中达到及格以上","en":"GPT-4 passes the Uniform Bar Exam"},"fieldImpact":"incremental","aiImpact":"major"},{"title":"Large Legal Fictions: Profiling Legal Hallucinations in Large Language Models","titleZh":"“大型法律虚构”：大模型法律幻觉的系统测量（Journal of Legal Analysis）","id":"10.1093/jla/laae003","url":"https://doi.org/10.1093/jla/laae003","venue":"Journal of Legal Analysis","date":"2024-06-26","why":{"zh":"量化大模型在判例问题上的幻觉率","en":"Quantifies LLM hallucination on case-law questions"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Hallucination‐Free? Assessing the Reliability of Leading\n                    AI\n                    Legal Research Tools","titleZh":"法律研究 AI 工具的幻觉：主流商业工具仍有 17%–33% 幻觉率（Journal of Empirical Legal Studies）","id":"10.1111/jels.12413","url":"https://doi.org/10.1111/jels.12413","venue":"Journal of Empirical Legal Studies","date":"2025-04-23","why":{"zh":"首个对商业法律 AI 工具的独立实证评估，反驳“无幻觉”宣传","en":"First independent test of commercial legal AI tools"},"fieldImpact":"major","aiImpact":"incremental"}],"latest_papers":[{"id":"arXiv:2610.29961","title":"Commit or Abstain? System One Decision Models, LLMs, and Missing Legal Authority","titleZh":"该作答还是弃权？“系统一”决策模型、大模型与缺失的法律依据","summaryZh":"测试输出各选项校准概率的决策模型在正确法律依据缺失时能否选择弃权，用于可扩展的法律依据核验。","summaryEn":"Tests whether calibrated “System One” decision models abstain when the correct legal authority is missing.","authors":["M. Mikail Demir"],"date":"2026-10-21","url":"https://arxiv.org/abs/2610.29961","site":"arXiv","venue":null,"category":"cs.AI","fetchedAt":"2026-10-22T10:09:52"},{"id":"arXiv:2610.27347","title":"An LLM-Based Agent for Legal Compliance Assessment: Architecture, Memory, and Evaluation in a Canadian and Quebec Context","titleZh":"面向法律合规评估的大模型智能体：架构、记忆与在加拿大和魁北克的评估","summaryZh":"ComplianceAgent 依据参考法律语料按固定量表给文本打分，低分时自动改写、标注、生成解释并升级给人工审核；在 13 部联邦、省级与市级法律文件上评估。","summaryEn":"An LLM compliance agent grounded in a legal corpus, evaluated on 13 Canadian and Quebec instruments.","authors":["Soukaina Alaoui Ismaili","Jean-Sebastien Dessureault"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.27347","site":"arXiv","venue":null,"category":"cs.AI","fetchedAt":"2026-10-22T10:09:52"},{"id":"arXiv:2610.26313","title":"StrucChunk: Structure-Aware Retrieval for Legal Document RAG","titleZh":"StrucChunk：面向法律文档 RAG 的结构感知检索","summaryZh":"把法律文档视为条款的有向无环图，保留条款边界、先解析交叉引用，再结合 BM25 与 BGE-M3 检索。","summaryEn":"Structure-aware chunking and hybrid retrieval for hierarchical legal documents.","authors":["Surya Deep Singh","Ashutosh Tripathi","Gaurav Deep Singh"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.26313","site":"arXiv","venue":null,"category":"cs.AI","fetchedAt":"2026-10-22T10:09:52"},{"id":"arXiv:2610.26211","title":"LLMs Prompted for Legal Context Refuse More: Overrefusal from Small On-Premises LLMs in Criminal Legal Context","titleZh":"提示法律语境的大模型拒答更多：小型本地大模型在刑事法律语境中的过度拒答","summaryZh":"律师在提示中声明身份等习惯会让小型开源模型选择性拒绝合法请求，研究评估了 4 个模型的过度拒答。","summaryEn":"Stating a lawyer role makes small on-prem LLMs over-refuse benign criminal-law requests.","authors":["Anastasiia Kucherenko","Francois Brouchoud","Dimitri Percia David","Andrei Kucharavy"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.26211","site":"arXiv","venue":null,"category":"cs.AI","fetchedAt":"2026-10-22T10:09:52"},{"id":"doi:10.5281/zenodo.23481759","title":"Bilingual Legal Citation Fabrication Benchmark: Saudi PDPL and GDPR (Arabic–English)","titleZh":"双语法律引用编造基准：沙特 PDPL 与 GDPR（阿英双语）","summaryZh":"120 道双语问题，检验大模型能否正确引用数据保护法条款、在无适用条款时拒答，还是编造引用。","summaryEn":"A 120-question Arabic–English benchmark testing whether LLMs cite correct GDPR/PDPL articles or fabricate.","authors":["Noura Alrajeh"],"date":"2026-10-19","url":"https://doi.org/10.5281/zenodo.23481759","site":"OpenAlex","venue":"arXiv (Cornell University)","category":null,"fetchedAt":"2026-10-22T10:09:52"},{"id":"arXiv:2610.26946","title":"Evidence-Grounded LLM-Assisted Android Mobile Forensic Reporting","titleZh":"基于证据的 LLM 辅助安卓取证报告","summaryZh":"把 Gemini、Llama、Qwen 生成的取证报告拆成 612 条原子陈述逐条核验：公开案例中三者都找全了 27 个事实，但引用有效性差异巨大，Llama 未产生有效引用。","summaryEn":"Claim-level evaluation of LLM-drafted Android forensic reports reveals large differences in citation validity.","authors":["Madhab Chandra Das","Sajid Ahmed","Tasnia Tabassum","Qingzhong Liu"],"date":"2026-10-18","url":"https://arxiv.org/abs/2610.26946","site":"arXiv","venue":null,"category":"cs.CR","fetchedAt":"2026-10-22T10:09:52"},{"id":"doi:10.1017/bhj.2026.10070","title":"Corporate Accountability by Algorithm: AI and the Interpretation of Human Rights Due Diligence","titleZh":"算法问责：AI 与人权尽职调查的解释","summaryZh":"企业和政府越来越多用 AI 工具做供应链人权尽职调查，文章分析其方法和科技公司角色，指出可能边缘化权利人、削弱企业问责。","summaryEn":"Examines AI tools for human-rights due diligence and their risks to corporate accountability.","authors":["Galit A. Sarfaty"],"date":"2026-10-19","url":"https://doi.org/10.1017/bhj.2026.10070","site":"OpenAlex","venue":"Business and Human Rights Journal","category":null,"fetchedAt":"2026-10-22T10:09:52"},{"id":"doi:10.25172/slrf.79.1.6","title":"A Response to “How AI Destroys Institutions”","titleZh":"回应《AI 如何摧毁制度》","summaryZh":"针对“AI 系统天然削弱专业性、决策和人际联系”的论点提出法学回应。","summaryEn":"A law-review response to the argument that AI systems inherently destroy institutions.","authors":["Andrew M. Perlman"],"date":"2026-10-19","url":"https://doi.org/10.25172/slrf.79.1.6","site":"OpenAlex","venue":"SMU Law Review Forum","category":null,"fetchedAt":"2026-10-22T10:09:52"}]},{"id":"education","slug":"education","category":"social","categoryName":{"zh":"社会科学","en":"Social sciences"},"name":{"zh":"教育学","en":"Education"},"classification":{"oecdFos":"5.3","moe":"0401 教育学"},"level":2,"levelKey":"breakthrough","levelLabel":{"zh":"局部突破","en":"Local breakthroughs"},"trend":"up","reason":{"zh":"随机对照试验开始给出证据：哈佛物理课 AI 导师的学习增益高于主动学习课堂（Sci. Rep. 2025）；2026 年 NBER 两年期 Khanmigo 试验显示数学成绩每年提高约 0.06–0.08 个标准差，但使用率低。","en":"RCT evidence is arriving: a Harvard physics AI tutor beat active-learning classes (Sci. Rep. 2025); a 2026 NBER two-year Khanmigo trial found ~0.06–0.08 SD/year maths gains, with low engagement."},"reasonSources":[{"url":"https://www.nature.com/articles/s41598-025-97652-6","linkCheck":"ok"},{"url":"https://www.nber.org/papers/w35620","linkCheck":"ok"}],"directions":{"zh":["AI 导师","自适应练习","自动评分与反馈"],"en":["AI tutors","Adaptive practice","Automated grading & feedback"]},"stats":{"milestones":6,"milestonesSince2025":6,"milestones2026":4,"milestonesLast6Months":3,"news":12,"latestProgress":"2026-10-20","latestMilestone":"2026-10-17","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":2,"leaps":{"ai":0,"field":0},"latestPapers":8,"keyPapers":3},"milestones":[{"date":"2025-06-03","title":{"zh":"哈佛物理课随机试验：AI 导师优于主动学习课堂（Scientific Reports）","en":"Harvard physics RCT: AI tutoring beats active learning (Scientific Reports)"},"url":"https://www.nature.com/articles/s41598-025-97652-6","source":"Scientific Reports","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"随机试验中 AI 导师组学习增益约为主动学习课堂的两倍","en":"RCT: AI-tutor learning gains about double active learning"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"随机试验中 AI 导师组学习增益约为主动学习课堂的两倍","en":"RCT: AI-tutor learning gains about double active learning"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"随机试验中 AI 导师组学习增益约为主动学习课堂的两倍","en":"RCT: AI-tutor learning gains about double active learning"}},"impactNotes":[],"verified":{"id":"10.1038/s41598-025-97652-6","title":"AI tutoring outperforms in-class active learning: an RCT introducing a novel research-based design in an authentic educational setting","venue":"Scientific Reports","date":"2025-06-03"},"addedIn":"v1"},{"date":"2025-06-25","title":{"zh":"不加防护的生成式 AI 可能损害学习：土耳其高中近千人实验（PNAS）","en":"Generative AI without guardrails can harm learning (PNAS)"},"url":"https://doi.org/10.1073/pnas.2422633122","source":"PNAS","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"使用 GPT-4 练习时成绩提升，撤掉后考试成绩反而下降 17%","en":"Practice gains vanish and exam scores drop 17% after removal"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"使用 GPT-4 练习时成绩提升，撤掉后考试成绩反而下降 17%","en":"Practice gains vanish and exam scores drop 17% after removal"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"使用 GPT-4 练习时成绩提升，撤掉后考试成绩反而下降 17%","en":"Practice gains vanish and exam scores drop 17% after removal"}},"impactNotes":[],"verified":{"id":"10.1073/pnas.2422633122","title":"Generative AI without guardrails can harm learning: Evidence from high school mathematics","venue":"Proceedings of the National Academy of Sciences","date":"2025-06-25"},"addedIn":"v2"},{"date":"2026-04-15","title":{"zh":"中国推进构建 AI 素养教育体系，中小学 AI 教育 2030 年基本普及","en":"China moves to build an AI-literacy system; AI education in schools to be near-universal by 2030"},"url":"https://english.www.gov.cn/news/202604/15/content_WS69df29e6c6d00ca5f9a0a6b1.html","source":"gov.cn","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"policy","typeLabel":{"zh":"政策","en":"Policy"},"why":{"zh":"全国性课程政策，支撑“教育课程”维度","en":"National curriculum policy"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"全国性课程政策，支撑“教育课程”维度","en":"National curriculum policy"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"全国性课程政策，支撑“教育课程”维度","en":"National curriculum policy"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-08","title":{"zh":"NBER：Khanmigo 两年期学校实验","en":"NBER: two-year Khanmigo school experiment"},"url":"https://www.nber.org/papers/w35620","source":"NBER","flags":["tbv"],"flagLabels":[{"zh":"日期或细节待核实","en":"Date/details to be verified"}],"linkCheck":"ok","datePrecision":"month","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"两年期学校实验，数学成绩每年提高约 0.06–0.08 个标准差","en":"Two-year school trial, ~0.06–0.08 SD/yr"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"两年期学校实验，数学成绩每年提高约 0.06–0.08 个标准差","en":"Two-year school trial, ~0.06–0.08 SD/yr"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"两年期学校实验，数学成绩每年提高约 0.06–0.08 个标准差","en":"Two-year school trial, ~0.06–0.08 SD/yr"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-08","title":{"zh":"NBER：掌握式数学练习中的 AI 辅导实验（6000+ 学生）","en":"NBER: AI tutoring in mastery-based maths practice (6,000+ students)"},"url":"https://www.nber.org/papers/w35621","source":"NBER","flags":["tbv"],"flagLabels":[{"zh":"日期或细节待核实","en":"Date/details to be verified"}],"linkCheck":"ok","datePrecision":"month","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"6000 多名学生的掌握式练习 AI 辅导实验","en":"6,000+ student tutoring experiment"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"6000 多名学生的掌握式练习 AI 辅导实验","en":"6,000+ student tutoring experiment"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"6000 多名学生的掌握式练习 AI 辅导实验","en":"6,000+ student tutoring experiment"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-10-17","title":{"zh":"中国农村 8647 名小学生大规模随机试验：AI 个性化学习未显著提高数学成绩（Economics of Education Review）","en":"Large RCT of 8,647 rural Chinese pupils: AI-personalised learning no significant maths gain"},"url":"https://doi.org/10.1016/j.econedurev.2026.102850","source":"Economics of Education Review","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"大样本无效结果，对“AI 导师普遍有效”的说法构成反证","en":"Large null result"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"大样本无效结果，对“AI 导师普遍有效”的说法构成反证","en":"Large null result"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"大样本无效结果，对“AI 导师普遍有效”的说法构成反证","en":"Large null result"}},"impactNotes":[],"verified":{"id":"10.1016/j.econedurev.2026.102850","title":"Putting AI to the test: Evidence from a large-scale RCT in rural China","venue":"Economics of Education Review","date":"2026-10-17"},"addedIn":"v2"}],"papersRepos":[{"title":"AI tutoring outperforms in-class active learning: an RCT introducing a novel research-based design in an authentic educational setting","url":"https://doi.org/10.1038/s41598-025-97652-6","date":"2025-06-03","kind":"paper","linkCheck":"ok-crossref"},{"title":"Generative AI without guardrails can harm learning: Evidence from high school mathematics","url":"https://doi.org/10.1073/pnas.2422633122","date":"2025-06-25","kind":"paper","linkCheck":"ok-crossref"},{"title":"Putting AI to the test: Evidence from a large-scale RCT in rural China","url":"https://doi.org/10.1016/j.econedurev.2026.102850","date":"2026-10-17","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"Khan Academy","url":"https://www.khanacademy.org"},{"name":"NBER","url":"https://www.nber.org"},{"name":"UNESCO","url":"https://www.unesco.org"}],"news":[{"id":"967a44fdf4a0","date":"2026-10-20","title":"Show HN: Penmentor – Claude marks up your handwritten math like a teacher","url":"https://github.com/AbdullahAlhariri/penmentor","source":"HN · AI 数学/定理证明","lang":"en","kind":"news","summaryZh":"HN · AI 数学/定理证明于2026-10-20发布的文章，涉及AI 数学、AI 教育、AI 软件开发，原题《Show HN: Penmentor – Claude marks up your handwritten math like a teacher》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"d2b0bbf7b60b","date":"2026-10-20","title":"儿童手表品牌成小学生社交准入门槛，“小天才封闭社交倒逼家长购买”冲上热搜","url":"https://www.ithome.com/1/015/567.htm","source":"IT之家","lang":"zh","kind":"news","summaryZh":"IT之家于2026-10-20发布的文章，涉及AI 教育、AI 安全、AI 商业融资，原题《儿童手表品牌成小学生社交准入门槛，“小天才封闭社交倒逼家长购买”冲上热搜》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"9d0f6ffd1f47","date":"2026-10-20","title":"SLA-LENS: From Classroom Transcripts to CSLA-Informed and Evidence-Linked L2 Interaction Analytics","url":"https://arxiv.org/abs/2610.24259","source":"arXiv cs.CL","lang":"en","kind":"paper","summaryZh":"arXiv cs.CL于2026-10-20发布的论文，涉及AI 教育、AI 商业融资、AI 软件开发，原题《SLA-LENS: From Classroom Transcripts to CSLA-Informed and Evidence-Linked L2 Interaction A》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"0f2042a540fe","date":"2026-10-20","title":"Voice-Based AI Feedback in Higher Education: A Field Study on Quality and Actionability","url":"https://arxiv.org/abs/2610.25560","source":"arXiv cs.CY 计算机与社会","lang":"en","kind":"paper","summaryZh":"arXiv cs.CY 计算机与社会于2026-10-20发布的论文，涉及AI 教育，原题《Voice-Based AI Feedback in Higher Education: A Field Study on Quality and Actionability》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"5681a2a9fffe","date":"2026-10-20","title":"Exploring Visual Access and Proactive AI Tutoring for Handwritten Mathematics","url":"https://arxiv.org/abs/2610.26411","source":"arXiv cs.CY 计算机与社会","lang":"en","kind":"paper","summaryZh":"arXiv cs.CY 计算机与社会于2026-10-20发布的论文，涉及AI 教育、AI 数学、AI 科研与调研，原题《Exploring Visual Access and Proactive AI Tutoring for Handwritten Mathematics》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"053eb508cbac","date":"2026-10-19","title":"Extending synthetic medical education: generating OSCE cases with LLMs","url":"https://www.nature.com/articles/s41746-026-03421-8","source":"npj Digital Medicine","lang":"en","kind":"news","summaryZh":"npj Digital Medicine于2026-10-19发布的文章，涉及AI 医疗、AI 教育，原题《Extending synthetic medical education: generating OSCE cases with LLMs》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"07df16b2bfff","date":"2026-10-19","title":"EF英孚、网易有道、伴鱼、伯索科技齐聚「iRTE2026教育专场」，共探AI时代教育行业新变化","url":"https://www.jiemodui.com/N/139608.html","source":"芥末堆","lang":"zh","kind":"news","summaryZh":"芥末堆于2026-10-19发布的文章，涉及AI 教育，原题《EF英孚、网易有道、伴鱼、伯索科技齐聚「iRTE2026教育专场」，共探AI时代教育行业新变化》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"ca08df55e72c","date":"2026-10-18","title":"Why the ugly history of IQ testing matters for AI and education","url":"https://theconversation.com/why-the-ugly-history-of-iq-testing-matters-for-ai-and-education-293661","source":"The Conversation · AI","lang":"en","kind":"news","summaryZh":"The Conversation · AI于2026-10-18发布的文章，涉及AI 科研与调研、AI 生物/医药、AI 教育，原题《Why the ugly history of IQ testing matters for AI and education》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"98ae48ba0bd7","date":"2026-10-16","title":"PhD students who write dissertations with AI less likely to stay in academia","url":"https://www.nature.com/articles/d41586-026-03253-2","source":"Nature 学科·机器学习","lang":"en","kind":"news","summaryZh":"Nature 学科·机器学习于2026-10-16发布的文章，涉及AI 科研与调研、AI 教育，原题《PhD students who write dissertations with AI less likely to stay in academia》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"0f185be104ce","date":"2026-10-16","title":"Educators who use AI say it frees up time to interact with students","url":"https://www.k12dive.com/news/educators-who-use-ai-say-it-frees-up-time-to-interact-with-students/833264/","source":"K-12 Dive","lang":"en","kind":"news","summaryZh":"K-12 Dive于2026-10-17发布的文章，涉及AI 教育，原题《Educators who use AI say it frees up time to interact with students》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"37c1ac19aa2c","date":"2026-10-15","title":"Do you think teachers are similar to coaches, pastors or something else? It depends on where you live","url":"https://theconversation.com/do-you-think-teachers-are-similar-to-coaches-pastors-or-something-else-it-depends-on-where-you-live-294029","source":"The Conversation · AI","lang":"en","kind":"news","summaryZh":"The Conversation · AI于2026-10-15发布的文章，涉及AI 科研与调研、AI 教育，原题《Do you think teachers are similar to coaches, pastors or something else? It depends on whe》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"d59a0a79113c","date":"2026-04-15","title":"China issues guideline on AI literacy education","url":"https://english.www.gov.cn/news/202604/15/content_WS69df29e6c6d00ca5f9a0a6b1.html","source":"gov.cn","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"ok"}],"review":{"lastReviewed":"2026-10-21","cycleDays":90,"nextReview":"2027-01-19"},"urls":{"zh":"/zh-hans/disciplines/education/","en":"/en/disciplines/education/"},"dimensions":{"method":{"score":2,"label":{"zh":"研究方法","en":"Research method"},"evidence":"随机对照试验开始系统评估 AI 导师","source":"https://www.nature.com/articles/s41598-025-97652-6","weight":0.3},"discovery":{"score":2,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"哈佛 AI 导师试验、NBER Khanmigo 试验给出正向效果，但也有无效或有害的证据","source":"https://doi.org/10.1073/pnas.2422633122","weight":0.35},"adoption":{"score":2,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"学生使用普遍，学校层面采用不一","source":null,"weight":0.25},"education":{"score":3,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"中国教育部推动中小学 AI 教育，目标 2030 年基本普及","source":"https://english.www.gov.cn/news/202604/15/content_WS69df29e6c6d00ca5f9a0a6b1.html","weight":0.1}},"compositeScore":2.1,"levelCapped":false,"levelChange":{"from":2,"to":2,"direction":"same","reason":null},"biggestBreakthrough":{"date":"2025-06-03","title":{"zh":"哈佛物理课随机试验：AI 导师优于主动学习课堂（Scientific Reports）","en":"Harvard physics RCT: AI tutoring beats active learning (Scientific Reports)"},"url":"https://www.nature.com/articles/s41598-025-97652-6","aiImpact":"major","fieldImpact":"major","why":{"zh":"随机试验中 AI 导师组学习增益约为主动学习课堂的两倍","en":"RCT: AI-tutor learning gains about double active learning"}},"keyPapers":[{"title":"AI tutoring outperforms in-class active learning: an RCT introducing a novel research-based design in an authentic educational setting","titleZh":"哈佛物理课随机试验：AI 导师优于主动学习课堂（Scientific Reports）","id":"10.1038/s41598-025-97652-6","url":"https://doi.org/10.1038/s41598-025-97652-6","venue":"Scientific Reports","date":"2025-06-03","why":{"zh":"随机试验中 AI 导师组学习增益约为主动学习课堂的两倍","en":"RCT: AI-tutor learning gains about double active learning"},"fieldImpact":"major","aiImpact":"major"},{"title":"Generative AI without guardrails can harm learning: Evidence from high school mathematics","titleZh":"不加防护的生成式 AI 可能损害学习：土耳其高中近千人实验（PNAS）","id":"10.1073/pnas.2422633122","url":"https://doi.org/10.1073/pnas.2422633122","venue":"Proceedings of the National Academy of Sciences","date":"2025-06-25","why":{"zh":"使用 GPT-4 练习时成绩提升，撤掉后考试成绩反而下降 17%","en":"Practice gains vanish and exam scores drop 17% after removal"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Putting AI to the test: Evidence from a large-scale RCT in rural China","titleZh":"中国农村 8647 名小学生大规模随机试验：AI 个性化学习未显著提高数学成绩（Economics of Education Review）","id":"10.1016/j.econedurev.2026.102850","url":"https://doi.org/10.1016/j.econedurev.2026.102850","venue":"Economics of Education Review","date":"2026-10-17","why":{"zh":"大样本无效结果，对“AI 导师普遍有效”的说法构成反证","en":"Large null result"},"fieldImpact":"major","aiImpact":"incremental"}],"latest_papers":[{"id":"arXiv:2610.29312","title":"A Stratified Multi-Rater Evaluation of LLM-Based Virtual Standardized Patients with a Deployed Data-Generation Platform","titleZh":"基于大模型的虚拟标准化病人：分层多评分者评估与已部署的数据生成平台","summaryZh":"用多种骨干模型和较大规模评分者小组，系统评估医学教育 OSCE 中大模型虚拟标准化病人的表现。","summaryEn":"A stratified multi-rater evaluation of LLM-based virtual standardised patients for OSCE medical education.","authors":["Yongjin Yi","Jinyong Park","Sejoong Kim"],"date":"2026-10-21","url":"https://arxiv.org/abs/2610.29312","site":"arXiv","venue":null,"category":"cs.HC","fetchedAt":"2026-10-22T10:09:58"},{"id":"arXiv:2610.27448","title":"Bias and Reliability in Learning-Style Inference from Student--AI Dialogues: An Evaluation of GPT-4o and Gemini 2.0 Flash","titleZh":"从学生–AI 对话推断学习风格的偏差与可靠性：GPT-4o 与 Gemini 2.0 Flash 评估","summaryZh":"用 153 名本科生的量表标签检验两模型从简短解题对话推断学习风格的能力，表面准确率中等，但多靠预测多数类，严重漏判直觉型和整体型学习者。","summaryEn":"LLMs infer learning styles from dialogues mostly by predicting majority labels, under-identifying intuitive and global learners.","authors":["Samaneh Zamanifard","Sajad Goudarzi","Andrew Robb"],"date":"2026-10-20","url":"https://arxiv.org/abs/2610.27448","site":"arXiv","venue":null,"category":"cs.HC","fetchedAt":"2026-10-22T10:09:58"},{"id":"arXiv:2610.27248","title":"ValiAssist: Scaling LLM-generated Programming Feedback Validation by Demonstration","titleZh":"ValiAssist：通过示范规模化验证 LLM 生成的编程反馈","summaryZh":"从教师修改 AI 反馈的操作中推断可复用的验证逻辑并应用到其他学生作业；16 名教师的对比研究显示能帮助制定和完善验证标准。","summaryEn":"Infers reusable validation logic from instructors’ edits to AI programming feedback; tested with 16 instructors.","authors":["Xiaohang Tang","Rehema Abulikemu","Peng Gao","Yan Chen"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.27248","site":"arXiv","venue":null,"category":"cs.HC","fetchedAt":"2026-10-22T10:09:58"},{"id":"arXiv:2610.25934","title":"MITRA-lexicon: Lexica of attested translations from Sanskrit into English, Tibetan and Chinese","titleZh":"MITRA 词典：梵语到英语、藏语和汉语的已证译文词典","summaryZh":"基于梵语为枢纽的译文对齐标注构建 6 部方向性词典，共 218 万词条、1040 万条译例，涵盖佛教文献的梵藏、梵汉对照。","summaryEn":"Corpus-grounded Sanskrit–English/Tibetan/Chinese lexica with 2.18M headwords and 10.4M attestations.","authors":["Sebastian Nehrdich","Kengo Harimoto","Sven Sellmer","Oliver Hellwig"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.25934","site":"arXiv","venue":null,"category":"cs.CL","fetchedAt":"2026-10-22T10:09:58"},{"id":"arXiv:2610.25244","title":"It's Complicated: Factors Influencing the Effectiveness of LLM-Generated Feedback in a Live Programming Course","titleZh":"情况很复杂：影响大模型生成反馈在真实编程课中效果的因素","summaryZh":"在网页开发课程中部署三轮大模型反馈系统，发现反馈在部分比较中与更好的即时成绩相关，但较长反馈与下一次提交成绩较低相关。","summaryEn":"Three iterations of LLM feedback in a live course: mixed associations, with longer feedback linked to lower next-submission scores.","authors":["Tsvetomila Mihaylova","Arto Hellas","Bita Akram","Narges Norouzi"],"date":"2026-10-18","url":"https://arxiv.org/abs/2610.25244","site":"arXiv","venue":null,"category":"cs.HC","fetchedAt":"2026-10-22T10:09:58"},{"id":"arXiv:2610.25192","title":"Acharya 2.0: An AI Tutor for Self-Learning Subjects in Hindi","titleZh":"Acharya 2.0：面向印地语自学科目的 AI 导师","summaryZh":"面向低资源语言印地语开发 AI 辅导系统，支持学生自学。","summaryEn":"An AI tutor for self-learning school subjects in Hindi.","authors":["Shantipriya Parida","Debasish Dhal","Sonal Khosla","Sambit Sekhar"],"date":"2026-10-18","url":"https://arxiv.org/abs/2610.25192","site":"arXiv","venue":null,"category":"cs.CL","fetchedAt":"2026-10-22T10:09:58"},{"id":"arXiv:2610.23530","title":"EduWild: How Teacher-Configured Chatbots Fail in Real-World Use and How to Fix Them","titleZh":"EduWild：教师配置的聊天机器人在真实使用中如何失败及如何修复","summaryZh":"收集 217 个教师配置的教育聊天机器人的 1726 段对话，最常见失败是不遵循教师意图，如跳过必要流程（53.8% 的终端用户对话）。","summaryEn":"1,726 conversations with 217 teacher-configured chatbots; most common failure is skipping required workflow steps (53.8%).","authors":["Ziqi Liu","Ruiwei Xiao","Yuxuan Li","John Stamper"],"date":"2026-10-16","url":"https://arxiv.org/abs/2610.23530","site":"arXiv","venue":null,"category":"cs.CL","fetchedAt":"2026-10-22T10:09:58"},{"id":"arXiv:2610.23149","title":"Can AI interview? Comparing AI-led to Human-led Chat Interviews","titleZh":"AI 能做访谈吗？AI 主导与人类主导的聊天访谈比较","summaryZh":"首次直接比较大模型聊天机器人与人类访谈者收集质性访谈数据的质量。","summaryEn":"First direct comparison of data quality between AI-led and human-led chat interviews.","authors":["Semra Guven","Tobias Gårdhus","Andreas Bjerre-Nielsen","Anna Rogers"],"date":"2026-10-16","url":"https://arxiv.org/abs/2610.23149","site":"arXiv","venue":null,"category":"cs.CY","fetchedAt":"2026-10-22T10:09:58"}]},{"id":"linguistics","slug":"linguistics","category":"social","categoryName":{"zh":"社会科学","en":"Social sciences"},"name":{"zh":"语言学","en":"Linguistics"},"classification":{"oecdFos":"6.2","moe":"050102 语言学及应用语言学"},"level":2,"levelKey":"breakthrough","levelLabel":{"zh":"局部突破","en":"Local breakthroughs"},"trend":"flat","reason":{"zh":"NLLB 把神经机器翻译扩展到 200 种语言（Nature 2024），低资源语言的记录与翻译明显受益；大模型也成为语言习得与句法理论争论的新证据来源。","en":"NLLB extended neural MT to 200 languages (Nature 2024), helping low-resource documentation; LLMs are now evidence in debates on acquisition and syntax."},"reasonSources":[{"url":"https://www.nature.com/articles/s41586-024-07335-x","linkCheck":"ok"}],"directions":{"zh":["低资源语言翻译","语言记录","大模型与语言理论"],"en":["Low-resource MT","Language documentation","LLMs & linguistic theory"]},"stats":{"milestones":9,"milestonesSince2025":4,"milestones2026":4,"milestonesLast6Months":3,"news":5,"latestProgress":"2026-10-20","latestMilestone":"2026-10-14","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":1,"leaps":{"ai":0,"field":0},"latestPapers":8,"keyPapers":8},"milestones":[{"date":"2022-12-06","title":{"zh":"OpenAI Whisper：68 万小时弱监督数据训练的多语种语音识别（arXiv 2212.04356）","en":"Whisper: robust multilingual speech recognition via large-scale weak supervision (arXiv 2212.04356)"},"url":"https://arxiv.org/abs/2212.04356","source":"arXiv","flags":["preprint","vendor"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"},{"zh":"厂商自述","en":"Vendor claim"}],"linkCheck":"ok-crossref","datePrecision":"day","type":"model","typeLabel":{"zh":"模型发布","en":"Model"},"why":{"zh":"开源多语种转写让田野录音、语料库语言学的转写成本大降","en":"Open multilingual ASR cuts transcription cost for fieldwork and corpora"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"开源多语种转写让田野录音、语料库语言学的转写成本大降","en":"Open multilingual ASR cuts transcription cost for fieldwork and corpora"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"开源多语种转写让田野录音、语料库语言学的转写成本大降","en":"Open multilingual ASR cuts transcription cost for fieldwork and corpora"}},"impactNotes":[],"verified":{"id":"arXiv:2212.04356","title":"Robust Speech Recognition via Large-Scale Weak Supervision","venue":"arXiv","date":"2022-12-06"},"addedIn":"v2"},{"date":"2023-02-25","title":{"zh":"大模型展示统计学习在语言习得中的潜力（Cognitive Science）","en":"LLMs demonstrate the potential of statistical learning in language (Cognitive Science)"},"url":"https://doi.org/10.1111/cogs.13256","source":"Cognitive Science","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"大模型成为“刺激贫乏论”争论的新证据","en":"LLMs enter the poverty-of-the-stimulus debate"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"大模型成为“刺激贫乏论”争论的新证据","en":"LLMs enter the poverty-of-the-stimulus debate"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"大模型成为“刺激贫乏论”争论的新证据","en":"LLMs enter the poverty-of-the-stimulus debate"}},"impactNotes":[],"verified":{"id":"10.1111/cogs.13256","title":"Large Language Models Demonstrate the Potential of Statistical Learning in Language","venue":"Cognitive Science","date":"2023-02-25"},"addedIn":"v2"},{"date":"2023-09-28","title":{"zh":"MTOB 基准：大模型仅凭一本语法书学习翻译卡拉曼语（Kalamang，约 200 名使用者）","en":"MTOB: learning to translate Kalamang from a single grammar book (arXiv 2309.16575)"},"url":"https://arxiv.org/abs/2309.16575","source":"arXiv","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok-crossref","datePrecision":"day","type":"benchmark","typeLabel":{"zh":"基准","en":"Benchmark"},"why":{"zh":"把“从描写语法习得语言”变成可测任务，直接关联濒危语言研究","en":"Turns learning from a descriptive grammar into a measurable task"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"把“从描写语法习得语言”变成可测任务，直接关联濒危语言研究","en":"Turns learning from a descriptive grammar into a measurable task"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"把“从描写语法习得语言”变成可测任务，直接关联濒危语言研究","en":"Turns learning from a descriptive grammar into a measurable task"}},"impactNotes":[],"verified":{"id":"arXiv:2309.16575","title":"A Benchmark for Learning to Translate a New Language from One Grammar Book","venue":"arXiv","date":"2023-09-28"},"addedIn":"v2"},{"date":"2024-06","title":{"zh":"区分大模型中的语言能力与思维（Trends in Cognitive Sciences）","en":"Dissociating language and thought in LLMs (Trends Cogn. Sci.)"},"url":"https://doi.org/10.1016/j.tics.2024.01.011","source":"Trends in Cognitive Sciences","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"month","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"提出形式语言能力与功能语言能力的区分，影响广泛","en":"Formal vs functional competence distinction"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"提出形式语言能力与功能语言能力的区分，影响广泛","en":"Formal vs functional competence distinction"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"提出形式语言能力与功能语言能力的区分，影响广泛","en":"Formal vs functional competence distinction"}},"impactNotes":[],"verified":{"id":"10.1016/j.tics.2024.01.011","title":"Dissociating language and thought in large language models","venue":"Trends in Cognitive Sciences","date":"2024-06"},"addedIn":"v2"},{"date":"2024-06-05","title":{"zh":"NLLB：将神经机器翻译扩展到 200 种语言（Nature）","en":"NLLB: scaling neural MT to 200 languages (Nature)"},"url":"https://www.nature.com/articles/s41586-024-07335-x","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"单一模型覆盖 200 种语言，低资源语言平均提升 44%","en":"200 languages in one model, +44% for low-resource"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"单一模型覆盖 200 种语言，低资源语言平均提升 44%","en":"200 languages in one model, +44% for low-resource"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"单一模型覆盖 200 种语言，低资源语言平均提升 44%","en":"200 languages in one model, +44% for low-resource"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-024-07335-x","title":"Scaling neural machine translation to 200 languages","venue":"Nature","date":"2024-06-05"},"addedIn":"v1"},{"date":"2026","title":{"zh":"评估前沿大模型的拉科塔语翻译能力：仍不可靠（AmericasNLP 2026）","en":"Evaluating frontier LLM translation for Lakota (AmericasNLP 2026)"},"url":"https://aclanthology.org/2026.americasnlp-6.2/","source":"ACL Anthology","flags":["tbv"],"flagLabels":[{"zh":"日期或细节待核实","en":"Date/details to be verified"}],"linkCheck":"ok","datePrecision":"year","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"濒危语言上大模型仍明显不足","en":"Frontier LLMs still unreliable on endangered languages"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"濒危语言上大模型仍明显不足","en":"Frontier LLMs still unreliable on endangered languages"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"濒危语言上大模型仍明显不足","en":"Frontier LLMs still unreliable on endangered languages"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-08-24","title":{"zh":"Nature Human Behaviour：大语言模型时代语言多样性收缩","en":"The shrinking landscape of linguistic diversity in the age of large language models (Nat. Hum. Behav.)"},"url":"https://www.nature.com/articles/s41562-026-02550-0","source":"Nature Human Behaviour","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"AI：大模型参与写作带来可测量的同质化；学科：语言变异与多样性研究出现新的外部驱动因素","en":"AI: measurable homogenisation from LLM-assisted writing; Field: a new driver of change in linguistic variation"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：大模型参与写作带来可测量的同质化；学科：语言变异与多样性研究出现新的外部驱动因素","en":"AI: measurable homogenisation from LLM-assisted writing; Field: a new driver of change in linguistic variation"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：大模型参与写作带来可测量的同质化；学科：语言变异与多样性研究出现新的外部驱动因素","en":"AI: measurable homogenisation from LLM-assisted writing; Field: a new driver of change in linguistic variation"}},"impactNotes":[],"verified":{"id":"10.1038/s41562-026-02550-0","title":"The shrinking landscape of linguistic diversity in the age of large language models","venue":"Nature Human Behaviour","date":"2026-08-24"},"addedIn":"v2"},{"date":"2026-10-12","title":{"zh":"PNAS：CHAT 框架显示，经多种校准后 9 个开源大模型生成的社交媒体语言仍可与人类语言系统区分","en":"AI-generated language remains systematically distinguishable from human communication (PNAS)"},"url":"https://doi.org/10.1073/pnas.2531748123","source":"PNAS","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"AI：指令微调模型反而不如基础模型像人，更大模型也无明显改善；学科：情感语气等特征仍是人机语言的稳定差异，给用大模型模拟语言行为的研究划定边界","en":"AI: instruction-tuned models are less human-like than base models; Field: affective tone remains a stable human–machine difference, limiting LLM language simulation"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：指令微调模型反而不如基础模型像人，更大模型也无明显改善；学科：情感语气等特征仍是人机语言的稳定差异，给用大模型模拟语言行为的研究划定边界","en":"AI: instruction-tuned models are less human-like than base models; Field: affective tone remains a stable human–machine difference, limiting LLM language simulation"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：指令微调模型反而不如基础模型像人，更大模型也无明显改善；学科：情感语气等特征仍是人机语言的稳定差异，给用大模型模拟语言行为的研究划定边界","en":"AI: instruction-tuned models are less human-like than base models; Field: affective tone remains a stable human–machine difference, limiting LLM language simulation"}},"impactNotes":[],"verified":{"id":"10.1073/pnas.2531748123","title":"AI-generated language remains systematically distinguishable from human communication","venue":"Proceedings of the National Academy of Sciences","date":"2026-10-12"},"addedIn":"v2"},{"date":"2026-10-14","title":{"zh":"ELARSpeech：从濒危语言档案整理 675 种少数语言、约 3 万小时语音（其中 443 种语言 2791 小时带标注），并发布识别与合成模型","en":"ELARSpeech: a 30K-hour speech dataset across 675 underrepresented languages from the Endangered Languages Archive"},"url":"https://arxiv.org/abs/2610.19388","source":"arXiv","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok","datePrecision":"day","type":"dataset","typeLabel":{"zh":"数据集","en":"Dataset"},"why":{"zh":"AI：为长尾语种语音识别与合成提供迄今覆盖语种最多的公开语料之一；学科：把语言记录档案转为可计算资源，使用须遵守社区同意与非商业条款（预印本）","en":"AI: one of the widest-coverage open speech corpora for long-tail languages. Field: turns documentation archives into computable resources under community-consent terms (preprint)"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：为长尾语种语音识别与合成提供迄今覆盖语种最多的公开语料之一；学科：把语言记录档案转为可计算资源，使用须遵守社区同意与非商业条款（预印本）","en":"AI: one of the widest-coverage open speech corpora for long-tail languages. Field: turns documentation archives into computable resources under community-consent terms (preprint)"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：为长尾语种语音识别与合成提供迄今覆盖语种最多的公开语料之一；学科：把语言记录档案转为可计算资源，使用须遵守社区同意与非商业条款（预印本）","en":"AI: one of the widest-coverage open speech corpora for long-tail languages. Field: turns documentation archives into computable resources under community-consent terms (preprint)"}},"impactNotes":[],"verified":{"id":"arXiv:2610.19388","title":"ELARSpeech: A 30K-Hour Dataset Across 675 Underrepresented Languages for Long-Tail Multilingual Speech Technologies","venue":"arXiv","date":"2026-10-14"},"addedIn":"v2-20261022pm"}],"papersRepos":[{"title":"facebookresearch/fairseq (NLLB branch)","url":"https://github.com/facebookresearch/fairseq/tree/nllb","date":null,"kind":"repo","linkCheck":"ok"},{"title":"Robust Speech Recognition via Large-Scale Weak Supervision","url":"https://arxiv.org/abs/2212.04356","date":"2022-12-06","kind":"paper","linkCheck":"ok-crossref"},{"title":"Large Language Models Demonstrate the Potential of Statistical Learning in Language","url":"https://doi.org/10.1111/cogs.13256","date":"2023-02-25","kind":"paper","linkCheck":"ok-crossref"},{"title":"A Benchmark for Learning to Translate a New Language from One Grammar Book","url":"https://arxiv.org/abs/2309.16575","date":"2023-09-28","kind":"paper","linkCheck":"ok-crossref"},{"title":"Dissociating language and thought in large language models","url":"https://doi.org/10.1016/j.tics.2024.01.011","date":"2024-06","kind":"paper","linkCheck":"ok-crossref"},{"title":"Scaling neural machine translation to 200 languages","url":"https://doi.org/10.1038/s41586-024-07335-x","date":"2024-06-05","kind":"paper","linkCheck":"ok-crossref"},{"title":"The shrinking landscape of linguistic diversity in the age of large language models","url":"https://doi.org/10.1038/s41562-026-02550-0","date":"2026-08-24","kind":"paper","linkCheck":"ok-crossref"},{"title":"AI-generated language remains systematically distinguishable from human communication","url":"https://doi.org/10.1073/pnas.2531748123","date":"2026-10-12","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"Meta FAIR","url":"https://ai.meta.com"},{"name":"SIL","url":"https://www.sil.org"}],"news":[{"id":"68a1c2784fc5","date":"2026-10-20","title":"Translating in style: An open-weight alternative to GPT-4 for culturally aligned machine translation","url":"https://arxiv.org/abs/2610.23868","source":"arXiv cs.CL","lang":"en","kind":"paper","summaryZh":"arXiv cs.CL于2026-10-20发布的论文，涉及AI 综合、AI 法律政策、AI 安全，原题《Translating in style: An open-weight alternative to GPT-4 for culturally aligned machine t》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"abc6c25e33b3","date":"2026-10-19","title":"ELARSpeech: A 30K-Hour Dataset Across 675 Underrepresented Languages for Long-Tail Multilingual Speech Technologies","url":"https://arxiv.org/abs/2610.19388","source":"arXiv","lang":"en","kind":"paper","summaryZh":"ELARSpeech 从濒危语言档案整理出覆盖 675 种少数语言、约 3 万小时的语音数据集，并发布语音识别/合成模型与处理流程。","via":"manual-websearch-20261022","linkCheck":"ok"},{"id":"78d6d66a2b4f","date":"2026-10-12","title":"Large Language Models for Machine Translation Quality Annotation: Humans and Models Are Both Challenged","url":"https://machinelearning.apple.com/research/llm-translation-quality-annotation","source":"Apple Machine Learning Research","lang":"en","kind":"news","summaryZh":"Apple Machine Learning Research于2026-10-12发布的文章，涉及AI 综合、AI 科研与调研、AI 商业融资，原题《Large Language Models for Machine Translation Quality Annotation: Humans and Models Are Bo》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"d660d5118443","date":"2026-10-10","title":"斩获融资的这8家语言学习平台，都在用AI做什么？","url":"https://www.jiemodui.com/N/139594.html","source":"芥末堆","lang":"zh","kind":"news","summaryZh":"芥末堆于2026-10-10发布的文章，涉及AI 教育、AI 商业融资，原题《斩获融资的这8家语言学习平台，都在用AI做什么？》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"c31d7b915419","date":"2026-09-24","title":"将Gemini当作语音模型，又一家AI语言学习公司拿到融资","url":"https://www.jiemodui.com/N/139577.html","source":"芥末堆","lang":"zh","kind":"news","summaryZh":"芥末堆于2026-09-24发布的文章，涉及AI 教育、AI 商业融资，原题《将Gemini当作语音模型，又一家AI语言学习公司拿到融资》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"}],"review":{"lastReviewed":"2026-10-22","cycleDays":90,"nextReview":"2027-01-19"},"urls":{"zh":"/zh-hans/disciplines/linguistics/","en":"/en/disciplines/linguistics/"},"dimensions":{"method":{"score":2,"label":{"zh":"研究方法","en":"Research method"},"evidence":"计算语言学被大模型重塑，理论语言学采用有限","source":"https://doi.org/10.1016/j.tics.2024.01.011","weight":0.3},"discovery":{"score":2,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"NLLB 覆盖 200 种语言，低资源语言翻译平均提升 44%","source":"https://www.nature.com/articles/s41586-024-07335-x","weight":0.35},"adoption":{"score":2,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"翻译与语言记录工作广泛使用","source":null,"weight":0.25},"education":{"score":1,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到课程证据","source":null,"weight":0.1}},"compositeScore":1.9,"levelCapped":false,"levelChange":{"from":2,"to":2,"direction":"same","reason":null},"biggestBreakthrough":{"date":"2024-06-05","title":{"zh":"NLLB：将神经机器翻译扩展到 200 种语言（Nature）","en":"NLLB: scaling neural MT to 200 languages (Nature)"},"url":"https://www.nature.com/articles/s41586-024-07335-x","aiImpact":"major","fieldImpact":"major","why":{"zh":"单一模型覆盖 200 种语言，低资源语言平均提升 44%","en":"200 languages in one model, +44% for low-resource"}},"keyPapers":[{"title":"Robust Speech Recognition via Large-Scale Weak Supervision","titleZh":"OpenAI Whisper：68 万小时弱监督数据训练的多语种语音识别（arXiv 2212.04356）","id":"arXiv:2212.04356","url":"https://arxiv.org/abs/2212.04356","venue":"arXiv","date":"2022-12-06","why":{"zh":"开源多语种转写让田野录音、语料库语言学的转写成本大降","en":"Open multilingual ASR cuts transcription cost for fieldwork and corpora"},"fieldImpact":"incremental","aiImpact":"major"},{"title":"Large Language Models Demonstrate the Potential of Statistical Learning in Language","titleZh":"大模型展示统计学习在语言习得中的潜力（Cognitive Science）","id":"10.1111/cogs.13256","url":"https://doi.org/10.1111/cogs.13256","venue":"Cognitive Science","date":"2023-02-25","why":{"zh":"大模型成为“刺激贫乏论”争论的新证据","en":"LLMs enter the poverty-of-the-stimulus debate"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"A Benchmark for Learning to Translate a New Language from One Grammar Book","titleZh":"MTOB 基准：大模型仅凭一本语法书学习翻译卡拉曼语（Kalamang，约 200 名使用者）","id":"arXiv:2309.16575","url":"https://arxiv.org/abs/2309.16575","venue":"arXiv","date":"2023-09-28","why":{"zh":"把“从描写语法习得语言”变成可测任务，直接关联濒危语言研究","en":"Turns learning from a descriptive grammar into a measurable task"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Dissociating language and thought in large language models","titleZh":"区分大模型中的语言能力与思维（Trends in Cognitive Sciences）","id":"10.1016/j.tics.2024.01.011","url":"https://doi.org/10.1016/j.tics.2024.01.011","venue":"Trends in Cognitive Sciences","date":"2024-06","why":{"zh":"提出形式语言能力与功能语言能力的区分，影响广泛","en":"Formal vs functional competence distinction"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Scaling neural machine translation to 200 languages","titleZh":"NLLB：将神经机器翻译扩展到 200 种语言（Nature）","id":"10.1038/s41586-024-07335-x","url":"https://doi.org/10.1038/s41586-024-07335-x","venue":"Nature","date":"2024-06-05","why":{"zh":"单一模型覆盖 200 种语言，低资源语言平均提升 44%","en":"200 languages in one model, +44% for low-resource"},"fieldImpact":"major","aiImpact":"major"},{"title":"The shrinking landscape of linguistic diversity in the age of large language models","titleZh":"Nature Human Behaviour：大语言模型时代语言多样性收缩","id":"10.1038/s41562-026-02550-0","url":"https://doi.org/10.1038/s41562-026-02550-0","venue":"Nature Human Behaviour","date":"2026-08-24","why":{"zh":"AI：大模型参与写作带来可测量的同质化；学科：语言变异与多样性研究出现新的外部驱动因素","en":"AI: measurable homogenisation from LLM-assisted writing; Field: a new driver of change in linguistic variation"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"AI-generated language remains systematically distinguishable from human communication","titleZh":"PNAS：CHAT 框架显示，经多种校准后 9 个开源大模型生成的社交媒体语言仍可与人类语言系统区分","id":"10.1073/pnas.2531748123","url":"https://doi.org/10.1073/pnas.2531748123","venue":"Proceedings of the National Academy of Sciences","date":"2026-10-12","why":{"zh":"AI：指令微调模型反而不如基础模型像人，更大模型也无明显改善；学科：情感语气等特征仍是人机语言的稳定差异，给用大模型模拟语言行为的研究划定边界","en":"AI: instruction-tuned models are less human-like than base models; Field: affective tone remains a stable human–machine difference, limiting LLM language simulation"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"ELARSpeech: A 30K-Hour Dataset Across 675 Underrepresented Languages for Long-Tail Multilingual Speech Technologies","titleZh":"ELARSpeech：从濒危语言档案整理 675 种少数语言、约 3 万小时语音（其中 443 种语言 2791 小时带标注），并发布识别与合成模型","id":"arXiv:2610.19388","url":"https://arxiv.org/abs/2610.19388","venue":"arXiv","date":"2026-10-14","why":{"zh":"AI：为长尾语种语音识别与合成提供迄今覆盖语种最多的公开语料之一；学科：把语言记录档案转为可计算资源，使用须遵守社区同意与非商业条款（预印本）","en":"AI: one of the widest-coverage open speech corpora for long-tail languages. Field: turns documentation archives into computable resources under community-consent terms (preprint)"},"fieldImpact":"major","aiImpact":"incremental"}],"latest_papers":[{"id":"arXiv:2610.29681","title":"Scaling Multilingual Pretraining of Pixel Language Models","titleZh":"扩展像素语言模型的多语言预训练","summaryZh":"把文本渲染成图像处理的像素语言模型在更多语言、文字和更大容量下预训练，并在覆盖 83 种语言的任务上评估跨语言迁移。","summaryEn":"Scales multilingual pretraining of pixel language models across languages, scripts and capacity, evaluated on 83 languages.","authors":["Ilker Kesen","Jenny Kunz","Kasra Malihi","Desmond Elliott"],"date":"2026-10-21","url":"https://arxiv.org/abs/2610.29681","site":"arXiv","venue":null,"category":"cs.CL","fetchedAt":"2026-10-22T10:10:01"},{"id":"arXiv:2610.28607","title":"Benchmarking Arabic Dialects for Math Problems","titleZh":"阿拉伯语方言数学题基准","summaryZh":"发布首个人工整理的方言阿拉伯语数学推理基准 DAR-MGSM，覆盖 5 种方言和标准阿拉伯语；11 个模型在文化本地化版本上普遍表现下降。","summaryEn":"DAR-MGSM, the first dialectal Arabic math benchmark; models drop on culturally localised versions.","authors":["Noor Abo Mokh","Teresa Lynn","Nizar Habash","Ahmed Elshabrawy"],"date":"2026-10-20","url":"https://arxiv.org/abs/2610.28607","site":"arXiv","venue":null,"category":"cs.CL","fetchedAt":"2026-10-22T10:10:01"},{"id":"arXiv:2610.27304","title":"Global Security Blind Spots: Translation as a Safety Bypass in Low-Resource Languages","titleZh":"全球安全盲区：低资源语言翻译成为安全绕过手段","summaryZh":"评测 10 个大模型在 78 种低资源语言上的翻译绕过：英语中安全拒绝的请求翻译后仍有 21.9%–44.4% 产生不安全回复，风险集中在部分语系。","summaryEn":"Across 78 low-resource languages, translation bypasses English safety alignment 21.9–44.4% of the time.","authors":["Jalil Alva","David Harper","Abigail E. Thornton","Rafael S. Villalobos Alva"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.27304","site":"arXiv","venue":null,"category":"cs.CL","fetchedAt":"2026-10-22T10:10:01"},{"id":"arXiv:2610.26155","title":"DeepSEA: A Multilingual Evaluation Benchmark for Southeast Asia","titleZh":"DeepSEA：东南亚多语言评测基准","summaryZh":"由专业译者把成熟数据集翻译成 12 种东南亚语言，评估语言模型在这些语言上的能力。","summaryEn":"A benchmark professionally translated into 12 Southeast Asian languages.","authors":["Jiuzhou Han","Abhishek Bijay Mishra","Trevor Cohn","Shubham Mittal"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.26155","site":"arXiv","venue":null,"category":"cs.CL","fetchedAt":"2026-10-22T10:10:01"},{"id":"arXiv:2610.24990","title":"The World as Viewed by 136 Countries: An Open, Curated, Annotated, and Multilingual Dataset of 118 Million State Media Articles","titleZh":"136 国眼中的世界：1.18 亿篇国家媒体文章的开放多语言数据集","summaryZh":"从预训练语料中提取 136 个国家、114 种语言的 1.18 亿篇国家媒体文章，研究其对大模型的潜在影响。","summaryEn":"WorldStateMedia: 118M state-media articles from 136 countries in 114 languages, extracted from pretraining corpora.","authors":["Bryan Li","Samar Haider","Grace Kim","Fiona Luo"],"date":"2026-10-18","url":"https://arxiv.org/abs/2610.24990","site":"arXiv","venue":null,"category":"cs.CL","fetchedAt":"2026-10-22T10:10:01"},{"id":"arXiv:2610.23205","title":"MAS-Bai: Linguistically Informed Adaptive Multi-Agent Translation for Bowen-Script Bai","titleZh":"MAS-Bai：面向白文（僰文）白语的语言学引导多智能体翻译","summaryZh":"白文常借汉字表音，导致直译幻觉；MAS-Bai 用检索翻译和带词源、句法智能体的语文学路径改进白语-汉语翻译。","summaryEn":"A linguistically informed multi-agent system for Bowen-script Bai–Chinese translation.","authors":["Jingting Zheng","Hong Ren","Yuehan Cui","Shaolin Zhu"],"date":"2026-10-16","url":"https://arxiv.org/abs/2610.23205","site":"arXiv","venue":null,"category":"cs.CL","fetchedAt":"2026-10-22T10:10:01"},{"id":"arXiv:2610.20370","title":"From Triage to Impact: A Multi-Hop Reasoning Benchmark for Explainable Persian Hate Speech","titleZh":"从分诊到影响：可解释波斯语仇恨言论的多跳推理基准","summaryZh":"构建细粒度波斯语仇恨言论基准，涵盖目标识别、策略、道德基础和严重度评估，并测试大模型。","summaryEn":"A multi-hop benchmark for explainable Persian hate-speech analysis.","authors":["Ghazal Kalhor","Sima Kamandloo","Mehrsa Shamsian","Behnam Bahrak"],"date":"2026-10-14","url":"https://arxiv.org/abs/2610.20370","site":"arXiv","venue":null,"category":"cs.CL","fetchedAt":"2026-10-22T10:10:01"},{"id":"arXiv:2610.19231","title":"Do Syntax Heads Influence Model Behavior?","titleZh":"句法注意力头会影响模型行为吗？","summaryZh":"检验专门关注句法关系的注意力头是否对模型句法预测起因果作用，证据有限。","summaryEn":"Tests whether specialised syntax attention heads causally drive syntactic predictions; evidence is limited.","authors":["William Timkey","Tal Linzen"],"date":"2026-10-13","url":"https://arxiv.org/abs/2610.19231","site":"arXiv","venue":null,"category":"cs.CL","fetchedAt":"2026-10-22T10:10:01"}]},{"id":"journalism","slug":"journalism","category":"social","categoryName":{"zh":"社会科学","en":"Social sciences"},"name":{"zh":"新闻传播学","en":"Journalism & communication"},"classification":{"oecdFos":"5.8","moe":"0503 新闻传播学"},"level":1,"levelKey":"assistive","levelLabel":{"zh":"辅助起步","en":"Assistive"},"trend":"up","reason":{"zh":"路透新闻研究所《2026 数字新闻报告》显示用 AI 聊天机器人获取新闻的比例从 7% 升至 10%；英国新闻编辑室广泛用 AI 转写、翻译、核查，但深度整合有限。","en":"Reuters Institute DNR 2026: use of AI chatbots for news rose from 7% to 10%; UK newsrooms use AI for transcription, translation and verification, but integration is limited."},"reasonSources":[{"url":"https://reutersinstitute.politics.ox.ac.uk/digital-news-report/2026/emerging-uses-ai-chatbots-news-and-what-it-means-journalism","linkCheck":"ok"},{"url":"https://reutersinstitute.politics.ox.ac.uk/ai-adoption-uk-journalists-and-their-newsrooms-surveying-applications-approaches-and-attitudes","linkCheck":"ok"}],"directions":{"zh":["新闻生产辅助","事实核查","AI 搜索与分发"],"en":["Newsroom assistance","Fact-checking","AI search & distribution"]},"stats":{"milestones":8,"milestonesSince2025":4,"milestones2026":4,"milestonesLast6Months":4,"news":6,"latestProgress":"2026-10-21","latestMilestone":"2026-10-15","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":0,"leaps":{"ai":0,"field":0},"latestPapers":8,"keyPapers":0},"milestones":[{"date":"2014-07-01","title":{"zh":"美联社用 Automated Insights 自动生成季度财报新闻，每季从约 300 篇扩到最多 4400 篇","en":"AP automates quarterly earnings stories, from ~300 to up to 4,400 per quarter"},"url":"https://www.nytimes.com/2014/07/01/business/media/the-ap-plans-for-computers-to-write-corporate-earnings-news.html","source":"The New York Times","flags":[],"flagLabels":[],"linkCheck":"site-blocks-bots","datePrecision":"day","type":"deployment","typeLabel":{"zh":"落地部署","en":"Deployment"},"why":{"zh":"主流通讯社首次规模化使用自动写作","en":"First large-scale automated writing at a major wire service"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"主流通讯社首次规模化使用自动写作","en":"First large-scale automated writing at a major wire service"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"主流通讯社首次规模化使用自动写作","en":"First large-scale automated writing at a major wire service"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2023-07-13","title":{"zh":"美联社与 OpenAI 达成协议：授权 1985 年以来部分新闻档案用于训练，换取技术合作","en":"AP and OpenAI agree to share select news content and technology"},"url":"https://www.ap.org/media-center/press-releases/2023/ap-open-ai-agree-to-share-select-news-content-and-technology-in-new-collaboration/","source":"The Associated Press","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"industry","typeLabel":{"zh":"产业/采用","en":"Industry"},"why":{"zh":"首个大型通讯社与 AI 公司的内容授权协议，开启“授权 vs 诉讼”两条路线","en":"First major wire-service licensing deal with an AI firm"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"首个大型通讯社与 AI 公司的内容授权协议，开启“授权 vs 诉讼”两条路线","en":"First major wire-service licensing deal with an AI firm"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首个大型通讯社与 AI 公司的内容授权协议，开启“授权 vs 诉讼”两条路线","en":"First major wire-service licensing deal with an AI firm"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2023-12-27","title":{"zh":"《纽约时报》起诉 OpenAI 和微软，指其用数百万篇报道训练模型侵犯版权","en":"The New York Times sues OpenAI and Microsoft over use of its articles for training"},"url":"https://www.reuters.com/legal/transactional/ny-times-sues-openai-microsoft-infringing-copyrighted-work-2023-12-27/","source":"Reuters","flags":["contested"],"flagLabels":[{"zh":"有争议","en":"Contested"}],"linkCheck":"site-blocks-bots","datePrecision":"day","type":"legal","typeLabel":{"zh":"诉讼/判决","en":"Legal"},"why":{"zh":"新闻业与 AI 公司关于训练数据的标志性诉讼，案件仍在审理","en":"Landmark training-data lawsuit; ongoing"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"新闻业与 AI 公司关于训练数据的标志性诉讼，案件仍在审理","en":"Landmark training-data lawsuit; ongoing"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"新闻业与 AI 公司关于训练数据的标志性诉讼，案件仍在审理","en":"Landmark training-data lawsuit; ongoing"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2024-03-11","title":{"zh":"普利策奖首次要求披露 AI 使用：45 个新闻类入围作品中 5 个在调查或呈现中使用了 AI","en":"Pulitzer requires AI disclosure for the first time: 5 of 45 journalism finalists used AI"},"url":"https://www.niemanlab.org/2024/03/five-of-this-years-pulitzer-finalists-are-ai-powered/","source":"Nieman Lab","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"prize","typeLabel":{"zh":"奖项","en":"Prize"},"why":{"zh":"AI 辅助调查报道进入新闻业最高奖项","en":"AI-assisted investigative work reaches the top prize"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI 辅助调查报道进入新闻业最高奖项","en":"AI-assisted investigative work reaches the top prize"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI 辅助调查报道进入新闻业最高奖项","en":"AI-assisted investigative work reaches the top prize"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-06-16","title":{"zh":"路透新闻研究所《2026 数字新闻报告》：10% 受访者每周用 AI 聊天机器人获取新闻（2025 年 7%）","en":"Reuters Institute DNR 2026: 10% use AI chatbots for news weekly (7% in 2025)"},"url":"https://reutersinstitute.politics.ox.ac.uk/digital-news-report/2026/dnr-executive-summary","source":"Reuters Institute","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"industry","typeLabel":{"zh":"产业/采用","en":"Industry"},"why":{"zh":"AI 开始改变新闻分发入口；只有 1% 视 AI 为主要来源（原“待核实”已确认 2026-06-16）","en":"AI starts to reshape news access; only 1% main source"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI 开始改变新闻分发入口；只有 1% 视 AI 为主要来源（原“待核实”已确认 2026-06-16）","en":"AI starts to reshape news access; only 1% main source"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI 开始改变新闻分发入口；只有 1% 视 AI 为主要来源（原“待核实”已确认 2026-06-16）","en":"AI starts to reshape news access; only 1% main source"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-10-06","title":{"zh":"加拿大 Postmedia、Torstar、《环球邮报》和加拿大通讯社在安大略起诉 Anthropic，指其抓取新闻训练 Claude 并绕过防复制措施","en":"Postmedia, Torstar, The Globe and Mail and The Canadian Press sue Anthropic in Ontario over use of their journalism to train Claude"},"url":"https://www.cbc.ca/news/business/anthropic-postmedia-torstar-globe-canadian-press-copyright-lawsuit-9.7387205","source":"CBC News","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"legal","typeLabel":{"zh":"诉讼/判决","en":"Legal"},"why":{"zh":"AI：诉状称 Anthropic 违反网站条款并规避技术保护措施；学科：继 2024 年起诉 OpenAI 后，加拿大主要新闻机构再次联合起诉 AI 公司（10-06 立案，10-19 公开报道）","en":"AI: alleges breach of site terms and circumvention of anti-copy measures. Field: Canada’s major news groups again jointly sue an AI firm (filed 10-06, reported 10-19)"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：诉状称 Anthropic 违反网站条款并规避技术保护措施；学科：继 2024 年起诉 OpenAI 后，加拿大主要新闻机构再次联合起诉 AI 公司（10-06 立案，10-19 公开报道）","en":"AI: alleges breach of site terms and circumvention of anti-copy measures. Field: Canada’s major news groups again jointly sue an AI firm (filed 10-06, reported 10-19)"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：诉状称 Anthropic 违反网站条款并规避技术保护措施；学科：继 2024 年起诉 OpenAI 后，加拿大主要新闻机构再次联合起诉 AI 公司（10-06 立案，10-19 公开报道）","en":"AI: alleges breach of site terms and circumvention of anti-copy measures. Field: Canada’s major news groups again jointly sue an AI firm (filed 10-06, reported 10-19)"}},"impactNotes":[],"verified":null,"addedIn":"v2-20261022pm"},{"date":"2026-10-08","title":{"zh":"USA Today 及 13 家关联报纸起诉 OpenAI 侵犯版权，索赔逾 2.5 亿美元","en":"USA Today Co. and 13 affiliated papers sue OpenAI for copyright infringement, seeking $250M+"},"url":"https://www.theverge.com/ai-artificial-intelligence/1008198/usa-today-openai-copyright-lawsuit","source":"The Verge","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"legal","typeLabel":{"zh":"诉讼/判决","en":"Legal"},"why":{"zh":"AI：诉求包括销毁涉案模型与训练数据；学科：美国新闻业版权诉讼继续扩大，聚焦新闻授权市场被替代","en":"AI: seeks destruction of implicated models and data; Field: US newspaper litigation widens, focusing on market substitution"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：诉求包括销毁涉案模型与训练数据；学科：美国新闻业版权诉讼继续扩大，聚焦新闻授权市场被替代","en":"AI: seeks destruction of implicated models and data; Field: US newspaper litigation widens, focusing on market substitution"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：诉求包括销毁涉案模型与训练数据；学科：美国新闻业版权诉讼继续扩大，聚焦新闻授权市场被替代","en":"AI: seeks destruction of implicated models and data; Field: US newspaper litigation widens, focusing on market substitution"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-10-15","title":{"zh":"谷歌 News AI 授权试点扩展到 400 多家媒体，纳入 CNN、FT、Axios 及美国地方新闻机构","en":"Google’s News AI pilot expands to 400+ publications incl. CNN, FT, Axios and US local outlets"},"url":"https://www.niemanlab.org/2026/10/googles-news-ai-pilot-expands-to-over-400-publications-including-u-s-local-news-outlets/","source":"Nieman Lab","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"industry","typeLabel":{"zh":"产业/采用","en":"Industry"},"why":{"zh":"AI：授权内容用于 AI 摘要“故事预览”和定制音频简报；学科：新闻内容授权从个别大厂交易扩展到地方媒体，但按次付费模式未采用","en":"AI: licensed content powers AI story previews and audio briefings; Field: licensing extends to local newsrooms"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：授权内容用于 AI 摘要“故事预览”和定制音频简报；学科：新闻内容授权从个别大厂交易扩展到地方媒体，但按次付费模式未采用","en":"AI: licensed content powers AI story previews and audio briefings; Field: licensing extends to local newsrooms"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：授权内容用于 AI 摘要“故事预览”和定制音频简报；学科：新闻内容授权从个别大厂交易扩展到地方媒体，但按次付费模式未采用","en":"AI: licensed content powers AI story previews and audio briefings; Field: licensing extends to local newsrooms"}},"impactNotes":[],"verified":null,"addedIn":"v2"}],"papersRepos":[],"orgsTools":[{"name":"Reuters Institute","url":"https://reutersinstitute.politics.ox.ac.uk"}],"news":[{"id":"8540e6f2083f","date":"2026-10-21","title":"Who’s suing AI and who’s signing latest: Canadian publishers sue Anthropic, Reach and CNN are latest Google AI news partners","url":"https://pressgazette.co.uk/platforms/news-publisher-ai-deals-lawsuits-openai-google/","source":"Press Gazette","lang":"en","kind":"news","summaryZh":"Press Gazette 汇总新闻出版商与 AI 公司的诉讼和授权：加拿大出版商起诉 Anthropic，Reach 与 CNN 成为谷歌 AI 新闻的新合作方。","via":"pool-v2-20261022","linkCheck":"site-blocks-bots"},{"id":"ad660a3bc964","date":"2026-10-19","title":"How publishers can compete with AI overviews by reviving their archives","url":"https://www.journalism.co.uk/how-publishers-can-compete-with-ai-overviews-by-reviving-their-archives/","source":"Journalism.co.uk","lang":"en","kind":"news","summaryZh":"Journalism.co.uk 社区文章：面对 AI 摘要分流流量，出版商与其限制抓取，不如主动盘活自家旧稿档案。","via":"pool-v2-20261022","linkCheck":"ok"},{"id":"a08ca06c8554","date":"2026-10-08","title":"USA Today becomes the latest publisher to sue OpenAI","url":"https://www.theverge.com/ai-artificial-intelligence/1008198/usa-today-openai-copyright-lawsuit","source":"The Verge AI","lang":"en","kind":"news","summaryZh":"USA Today 及旗下地方报纸起诉 OpenAI，指控其复制数十万篇文章用于训练。","via":"research-samples-20261010","linkCheck":"from-feed"},{"id":"3f8fa372e5f2","date":"2026-10-07","title":"“Uber agents,” undercover AI personas, and other ways newsrooms are using AI in investigations","url":"https://www.niemanlab.org/2026/10/uber-agents-undercover-ai-personas-and-other-ways-newsrooms-are-using-ai-in-investigations/","source":"Nieman Lab","lang":"en","kind":"news","summaryZh":"Nieman Lab于2026-10-08发布的文章，涉及AI 创意媒体、AI 科研与调研，原题《“Uber agents,” undercover AI personas, and other ways newsrooms are using AI in investigat》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"9c32e6a8654a","date":"2026-09-22","title":"How scientific publishers are fighting back against fraudulent papermills","url":"https://physicsworld.com/a/how-scientific-publishers-are-fighting-back-against-fraudulent-papermills/","source":"Physics World（IOP）","lang":"en","kind":"news","summaryZh":"Physics World（IOP）于2026-09-22发布的文章，涉及AI 科研与调研、AI 创意媒体，原题《How scientific publishers are fighting back against fraudulent papermills》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"51669a1551d4","date":"2026-06-16","title":"Digital News Report 2026: executive summary","url":"https://reutersinstitute.politics.ox.ac.uk/digital-news-report/2026/dnr-executive-summary","source":"Reuters Institute","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"ok"}],"review":{"lastReviewed":"2026-10-22","cycleDays":90,"nextReview":"2027-01-19"},"urls":{"zh":"/zh-hans/disciplines/journalism/","en":"/en/disciplines/journalism/"},"dimensions":{"method":{"score":1,"label":{"zh":"研究方法","en":"Research method"},"evidence":"新闻学研究中 AI 主要作为研究对象","source":null,"weight":0.3},"discovery":{"score":0,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"无以 AI 为关键的新闻学研究成果","source":null,"weight":0.35},"adoption":{"score":2,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"AP 自 2014 年自动生成财报稿；2026 年 10% 受众每周用 AI 聊天机器人获取新闻","source":"https://reutersinstitute.politics.ox.ac.uk/digital-news-report/2026/dnr-executive-summary","weight":0.25},"education":{"score":1,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到课程证据","source":null,"weight":0.1}},"compositeScore":0.9,"levelCapped":false,"levelChange":{"from":1,"to":1,"direction":"same","reason":null},"biggestBreakthrough":{"date":"2014-07-01","title":{"zh":"美联社用 Automated Insights 自动生成季度财报新闻，每季从约 300 篇扩到最多 4400 篇","en":"AP automates quarterly earnings stories, from ~300 to up to 4,400 per quarter"},"url":"https://www.nytimes.com/2014/07/01/business/media/the-ap-plans-for-computers-to-write-corporate-earnings-news.html","aiImpact":"major","fieldImpact":"major","why":{"zh":"主流通讯社首次规模化使用自动写作","en":"First large-scale automated writing at a major wire service"}},"keyPapers":[],"latest_papers":[{"id":"doi:10.70382/hijaerd.v13i6.073","title":"AI-ASSISTED FINANCIAL JOURNALISM AND THE CREDIBILITY OF CORPORATE FINANCIAL INFORMATION IN NIGERIA","titleZh":"尼日利亚 AI 辅助财经新闻与企业财务信息可信度","summaryZh":"讨论生成式 AI 等工具在尼日利亚财经新闻生产中的机遇与风险，以及对企业财务信息可信度的影响。","summaryEn":"Examines AI-assisted financial journalism in Nigeria and its effect on the credibility of corporate financial information.","authors":["ATUKUM MARYAM AWUNGHAY","OLADOKUN PAUL OLAYIWOLA"],"date":"2026-10-20","url":"https://doi.org/10.70382/hijaerd.v13i6.073","site":"OpenAlex","venue":"International Journal of Anthropology and Ethnology Research Development","category":null,"fetchedAt":"2026-10-22T10:10:05"},{"id":"arXiv:2610.25777","title":"Standing on the Shoulders of Giants: Enhancing LLM Agents for Automated Fact-Checking with Human Expert Skills","titleZh":"站在巨人肩上：用人类专家技能增强自动事实核查智能体","summaryZh":"把法新社事实核查公开指南提炼为可复用的专业核查技能，构建 SkillCheck 系统。","summaryEn":"Distils AFP Fact Check guidelines into a reusable skill for LLM fact-checking agents (SkillCheck).","authors":["Xinwen Chen","Haorui He","Yupeng Li"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.25777","site":"arXiv","venue":null,"category":"cs.AI","fetchedAt":"2026-10-22T10:10:05"},{"id":"doi:10.3126/irjmmc.v7i4.101334","title":"Rethinking Lasswell's 5W Model of Communication in the Era of Generative AI","titleZh":"生成式 AI 时代重新思考拉斯韦尔 5W 传播模式","summaryZh":"论证生成式 AI 动摇了“人类信源、静态信息、渠道只是载体”等假设，主张扩展而非抛弃 5W 模式。","summaryEn":"Argues generative AI unsettles the assumptions of Lasswell’s 5W model and proposes extending rather than abandoning it.","authors":["Anjali Sapkota"],"date":"2026-10-19","url":"https://doi.org/10.3126/irjmmc.v7i4.101334","site":"OpenAlex","venue":"International Research Journal of MMC","category":null,"fetchedAt":"2026-10-22T10:10:05"},{"id":"doi:10.36948/ijfmr.2026.v08i05.90001","title":"Artificial Intelligence Adoption in Media Organizations: Emerging Practices, Opportunities and Implications for Social and Cultural Development","titleZh":"媒体机构中的 AI 应用：新兴实践、机遇及对社会文化发展的影响","summaryZh":"综述 AI 在内容写作、编辑、转写、翻译、事实核查、受众分析和个性化推荐等媒体生产环节中的应用。","summaryEn":"Reviews AI adoption across media production stages from writing to personalisation.","authors":["Shikeb Majeed"],"date":"2026-10-17","url":"https://doi.org/10.36948/ijfmr.2026.v08i05.90001","site":"OpenAlex","venue":"International Journal For Multidisciplinary Research","category":null,"fetchedAt":"2026-10-22T10:10:05"},{"id":"arXiv:2610.22724","title":"A misinformation playbook sharpens judgement without breeding cynicism","titleZh":"错误信息“套路手册”提升判断力而不滋生犬儒","summaryZh":"从 51 种语言的 214646 条已核查说法中归纳说服套路，学习这些套路能提升人与机器对说法的判断。","summaryEn":"Learning persuasion strategies mapped from 214,646 fact-checked claims in 51 languages sharpens human and machine judgement.","authors":["Tong Guo","Teng Ye","Yutong Xie","Qiaozhu Mei"],"date":"2026-10-15","url":"https://arxiv.org/abs/2610.22724","site":"arXiv","venue":null,"category":"cs.CY","fetchedAt":"2026-10-22T10:10:05"},{"id":"doi:10.1007/s44382-026-00042-3","title":"Cues for you: a comparative test of agenda cueing and attitudes toward algorithmic news in the US and the UK","titleZh":"为你提示：美国与英国议程提示及对算法新闻态度的比较检验","summaryZh":"在线实验比较美英两国：新闻编辑的议程提示能提高议题重要性认知，算法提示的效果则不同。","summaryEn":"A US–UK experiment compares journalistic vs algorithmic agenda cues.","authors":["Elizabeth Stoycheff","Robert B Swetlic"],"date":"2026-10-15","url":"https://doi.org/10.1007/s44382-026-00042-3","site":"OpenAlex","venue":"Communication and Change","category":null,"fetchedAt":"2026-10-22T10:10:05"},{"id":"doi:10.17645/mac.12492","title":"Do Malaysian Newsrooms Jump on the AI Bandwagon? Empirical Evidence From Journalists’ Perspectives","titleZh":"马来西亚新闻编辑室是否追赶 AI 潮流？来自记者视角的证据","summaryZh":"研究 AI 在马来西亚新闻编辑室中如何被理解、组织和使用，补充非西方媒体环境的证据。","summaryEn":"How AI is interpreted, organised and used in Malaysian newsrooms.","authors":["Jiankun Gong","Yu Zuo","Izzati Aziz","Mumtaz Aini Alivi"],"date":"2026-10-15","url":"https://doi.org/10.17645/mac.12492","site":"OpenAlex","venue":"Media and Communication","category":null,"fetchedAt":"2026-10-22T10:10:05"},{"id":"doi:10.71016/prtfs/pps9t528","title":"AI and Journalism: Exploring Ethics, Transparency and Self-Regulation among Pakistani TV Journalists","titleZh":"AI 与新闻：巴基斯坦电视记者的伦理、透明与自律","summaryZh":"研究巴基斯坦电视记者如何看待新闻工作中使用 AI 的伦理、透明度与自律问题。","summaryEn":"Pakistani TV journalists’ views on ethics, transparency and self-regulation in AI use.","authors":["Adiba Akhtar","Naeem Ullah Tariq","Aqeel Ahmed","Nur Nadia Abd Mubin"],"date":"2026-10-15","url":"https://doi.org/10.71016/prtfs/pps9t528","site":"OpenAlex","venue":"Print Radio TV and Film Studies","category":null,"fetchedAt":"2026-10-22T10:10:05"}]},{"id":"sports-science","slug":"sports-science","category":"social","categoryName":{"zh":"社会科学","en":"Social sciences"},"name":{"zh":"体育学","en":"Sports science"},"classification":{"oecdFos":"3.3","moe":"0403 体育学"},"level":1,"levelKey":"assistive","levelLabel":{"zh":"辅助起步","en":"Assistive"},"trend":"flat","reason":{"zh":"下调为 L1：AI 在体育中主要是追踪、视频分析和判罚辅助（如 2022 世界杯半自动越位）；以 AI 为关键的研究成果只有 TacticAI（Nat. Commun. 2024）等少数，且属战术建议工具，尚不足以称“局部突破”。","en":"Lowered to L1: AI in sport is mainly tracking, video analysis and officiating aids (e.g. semi-automated offside at the 2022 World Cup); AI-key research is limited to a few tools such as TacticAI (Nat. Commun. 2024)."},"reasonSources":[{"url":"https://www.nature.com/articles/s41467-024-45965-x","linkCheck":"ok"},{"url":"https://inside.fifa.com/innovation/media-releases/semi-automated-offside-technology-to-be-used-at-fifa-world-cup-2022-tm","linkCheck":"ok"}],"directions":{"zh":["战术分析","动作捕捉与生物力学","伤病预测"],"en":["Tactical analysis","Motion capture & biomechanics","Injury prediction"]},"stats":{"milestones":10,"milestonesSince2025":6,"milestones2026":6,"milestonesLast6Months":6,"news":5,"latestProgress":"2026-10-19","latestMilestone":"2026-10-08","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":0,"leaps":{"ai":0,"field":0},"latestPapers":8,"keyPapers":2},"milestones":[{"date":"2022-07-01","title":{"zh":"FIFA 宣布 2022 世界杯使用半自动越位技术：12 台摄像机每秒 50 次追踪 29 个身体点位，AI 提示视频裁判","en":"FIFA announces semi-automated offside for the 2022 World Cup (12 cameras, 29 body points, 50 Hz)"},"url":"https://inside.fifa.com/innovation/media-releases/semi-automated-offside-technology-to-be-used-at-fifa-world-cup-2022-tm","source":"FIFA","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"deployment","typeLabel":{"zh":"落地部署","en":"Deployment"},"why":{"zh":"AI 进入最高级别赛事的判罚流程","en":"AI enters officiating at the top tournament"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI 进入最高级别赛事的判罚流程","en":"AI enters officiating at the top tournament"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI 进入最高级别赛事的判罚流程","en":"AI enters officiating at the top tournament"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2024-03-19","title":{"zh":"TacticAI：足球战术 AI 助手（Nat. Commun.）","en":"TacticAI: AI assistant for football tactics (Nat. Commun.)"},"url":"https://www.nature.com/articles/s41467-024-45965-x","source":"Nature Communications","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"与利物浦合作，角球战术建议在盲评中 90% 被专家偏好","en":"Liverpool collaboration; experts preferred suggestions 90% of the time"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"与利物浦合作，角球战术建议在盲评中 90% 被专家偏好","en":"Liverpool collaboration; experts preferred suggestions 90% of the time"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"与利物浦合作，角球战术建议在盲评中 90% 被专家偏好","en":"Liverpool collaboration; experts preferred suggestions 90% of the time"}},"impactNotes":[],"verified":{"id":"10.1038/s41467-024-45965-x","title":"TacticAI: an AI assistant for football tactics","venue":"Nature Communications","date":"2024-03-19"},"addedIn":"v1"},{"date":"2024-04-19","title":{"zh":"国际奥委会发布《奥林匹克 AI 议程》，覆盖选材、个性化训练、裁判与运动员保护","en":"IOC launches the Olympic AI Agenda (talent ID, training, judging, athlete protection)"},"url":"https://www.olympics.com/ioc/news/ioc-takes-the-lead-for-the-olympic-movement-and-launches-olympic-ai-agenda","source":"IOC","flags":[],"flagLabels":[],"linkCheck":"site-blocks-bots","datePrecision":"day","type":"policy","typeLabel":{"zh":"政策","en":"Policy"},"why":{"zh":"全球体育最高治理机构把 AI 纳入战略","en":"Top sports governing body adopts an AI strategy"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"全球体育最高治理机构把 AI 纳入战略","en":"Top sports governing body adopts an AI strategy"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"全球体育最高治理机构把 AI 纳入战略","en":"Top sports governing body adopts an AI strategy"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2024-10-09","title":{"zh":"温布尔登宣布 2025 年起以 Hawk-Eye 实时电子司线全面取代人工司线员（147 年来首次）","en":"Wimbledon replaces line judges with Hawk-Eye Live electronic line calling from 2025"},"url":"https://www.reuters.com/sports/tennis/out-wimbledon-calls-time-line-judges-favour-electronic-system-2024-10-09/","source":"Reuters","flags":[],"flagLabels":[],"linkCheck":"site-blocks-bots","datePrecision":"day","type":"deployment","typeLabel":{"zh":"落地部署","en":"Deployment"},"why":{"zh":"计算机视觉判罚在顶级赛事中完全取代人类司线","en":"Computer-vision calls fully replace human line judges at a Grand Slam"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"计算机视觉判罚在顶级赛事中完全取代人类司线","en":"Computer-vision calls fully replace human line judges at a Grand Slam"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"计算机视觉判罚在顶级赛事中完全取代人类司线","en":"Computer-vision calls fully replace human line judges at a Grand Slam"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-06-03","title":{"zh":"FIFA 公布 2026 世界杯技术：进阶半自动越位首次把明显越位直接推送给场上助理裁判，全部球员 3D 扫描生成 AI 数字替身；生成式 AI 分析助手 Football AI Pro 向 48 支球队平等开放（仅限赛前赛后）","en":"FIFA details World Cup 2026 tech: advanced semi-automated offside alerts go straight to on-field officials, AI-enabled 3D player avatars, and the generative-AI assistant Football AI Pro for all 48 teams (pre/post-match only)"},"url":"https://inside.fifa.com/news/offside-decisions-referee-body-cams-innovation-world-cup-2026","source":"FIFA","flags":["vendor"],"flagLabels":[{"zh":"厂商自述","en":"Vendor claim"}],"linkCheck":"ok","datePrecision":"day","type":"deployment","typeLabel":{"zh":"落地部署","en":"Deployment"},"why":{"zh":"AI：每座球场 16 台光学追踪摄像机，越位判定从“提示 VAR”前移到“直接提示场上裁判”（仅限位置越位，干扰判罚仍由人决定）；学科：首次让所有参赛队用同一套 AI 比赛分析工具，缩小强弱队的数据分析差距（效果为 FIFA 自述）","en":"AI: 16 tracking cameras per stadium; positional offside now flagged directly to on-field officials (interference still human). Field: first time all teams share the same AI match-analysis tool (FIFA’s own account)"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：每座球场 16 台光学追踪摄像机，越位判定从“提示 VAR”前移到“直接提示场上裁判”（仅限位置越位，干扰判罚仍由人决定）；学科：首次让所有参赛队用同一套 AI 比赛分析工具，缩小强弱队的数据分析差距（效果为 FIFA 自述）","en":"AI: 16 tracking cameras per stadium; positional offside now flagged directly to on-field officials (interference still human). Field: first time all teams share the same AI match-analysis tool (FIFA’s own account)"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：每座球场 16 台光学追踪摄像机，越位判定从“提示 VAR”前移到“直接提示场上裁判”（仅限位置越位，干扰判罚仍由人决定）；学科：首次让所有参赛队用同一套 AI 比赛分析工具，缩小强弱队的数据分析差距（效果为 FIFA 自述）","en":"AI: 16 tracking cameras per stadium; positional offside now flagged directly to on-field officials (interference still human). Field: first time all teams share the same AI match-analysis tool (FIFA’s own account)"}},"impactNotes":[],"verified":null,"addedIn":"v2-20261022audit"},{"date":"2026-08-26","title":{"zh":"爱尔兰四支职业橄榄球队 + 国家队 3 个赛季数据：五种机器学习模型预测下肢非接触伤病，最好的 CatBoost 也只有 AUC 0.66、精确率 1.3%，作者结论是“预测性能差”（Sports Medicine – Open）","en":"Multi-team, multi-season study of 129 elite Irish rugby players: ML injury-risk models perform poorly (best CatBoost ROC AUC 0.66, precision 0.013) (Sports Medicine – Open)"},"url":"https://doi.org/10.1186/s40798-026-01060-7","source":"Sports Medicine – Open","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"AI：GNSS 负荷、主观状态和体能筛查数据输入逻辑回归、SVM、随机森林、XGBoost、CatBoost；学科：这类多队多赛季的负面结果说明“AI 预测伤病”离可用还很远，是判断体育学仍处 L1 的重要依据","en":"AI: GNSS load, wellness and screening data across five model types. Field: a rare multi-team negative result showing AI injury prediction is far from usable — key evidence for keeping sports science at L1"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：GNSS 负荷、主观状态和体能筛查数据输入逻辑回归、SVM、随机森林、XGBoost、CatBoost；学科：这类多队多赛季的负面结果说明“AI 预测伤病”离可用还很远，是判断体育学仍处 L1 的重要依据","en":"AI: GNSS load, wellness and screening data across five model types. Field: a rare multi-team negative result showing AI injury prediction is far from usable — key evidence for keeping sports science at L1"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：GNSS 负荷、主观状态和体能筛查数据输入逻辑回归、SVM、随机森林、XGBoost、CatBoost；学科：这类多队多赛季的负面结果说明“AI 预测伤病”离可用还很远，是判断体育学仍处 L1 的重要依据","en":"AI: GNSS load, wellness and screening data across five model types. Field: a rare multi-team negative result showing AI injury prediction is far from usable — key evidence for keeping sports science at L1"}},"impactNotes":[],"verified":{"id":"10.1186/s40798-026-01060-7","title":"Machine Learning Model Development and Evaluation for Non-contact Lower Limb Injury Risk Prediction in Elite Male Rugby Union: A Multi-team, Multi-season Analysis","venue":"Sports Medicine - Open","date":"2026-08-26"},"addedIn":"v2-20261022audit"},{"date":"2026-09-11","title":{"zh":"比利时安德莱赫特俱乐部采用 MyGamePlan AI 对手分析平台","en":"RSC Anderlecht adopts MyGamePlan AI opponent analysis"},"url":"https://www.mygameplan.ai/blog/rsc-anderlecht-sign-with-mygameplan","source":"MyGamePlan","flags":["vendor"],"flagLabels":[{"zh":"厂商自述","en":"Vendor claim"}],"linkCheck":"ok","datePrecision":"day","type":"industry","typeLabel":{"zh":"产业/采用","en":"Industry"},"why":{"zh":"俱乐部层面采用","en":"Club-level adoption"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"俱乐部层面采用","en":"Club-level adoption"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"俱乐部层面采用","en":"Club-level adoption"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-09-29","title":{"zh":"国际奥委会与德勤发布《奥林匹克运动可信 AI 框架》：安全、数据保护、可靠、透明、问责、公平、可持续七项原则，附三步自评清单，征询了 200 多位奥林匹克运动负责人","en":"IOC launches its Trustworthy AI Framework for the Olympic Movement with Deloitte: seven principles and a three-step self-assessment checklist"},"url":"https://www.insidethegames.biz/articles/ioc-launches-trustworthy-ai-framework","source":"Inside The Games / IOC","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"policy","typeLabel":{"zh":"政策","en":"Policy"},"why":{"zh":"AI：把运动员数据保护、公平竞赛和 AI 使用透明度写成可操作的检查清单；学科：继 2024 年《奥林匹克 AI 议程》之后，国际奥委会首次给整个奥林匹克运动提供统一的 AI 治理工具（IOC 原文页面拒绝脚本访问，日期取 Inside The Games 报道）","en":"AI: turns athlete-data protection, fair play and transparency into a checklist. Field: first common AI-governance tool for the whole Olympic Movement after the 2024 Olympic AI Agenda"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：把运动员数据保护、公平竞赛和 AI 使用透明度写成可操作的检查清单；学科：继 2024 年《奥林匹克 AI 议程》之后，国际奥委会首次给整个奥林匹克运动提供统一的 AI 治理工具（IOC 原文页面拒绝脚本访问，日期取 Inside The Games 报道）","en":"AI: turns athlete-data protection, fair play and transparency into a checklist. Field: first common AI-governance tool for the whole Olympic Movement after the 2024 Olympic AI Agenda"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：把运动员数据保护、公平竞赛和 AI 使用透明度写成可操作的检查清单；学科：继 2024 年《奥林匹克 AI 议程》之后，国际奥委会首次给整个奥林匹克运动提供统一的 AI 治理工具（IOC 原文页面拒绝脚本访问，日期取 Inside The Games 报道）","en":"AI: turns athlete-data protection, fair play and transparency into a checklist. Field: first common AI-governance tool for the whole Olympic Movement after the 2024 Olympic AI Agenda"}},"impactNotes":[],"verified":null,"addedIn":"v2-20261022audit"},{"date":"2026-10-05","title":{"zh":"NBA 季前赛试用威尔逊内置传感器的“智能篮球”采集运动、旋转和轨迹数据，为将来实时判罚做准备（本季只做研究，不影响判罚）；裁判同时试用腕戴设备与回放中心沟通","en":"NBA tests Wilson sensor-equipped smart basketballs in preseason to gather data for future real-time officiating (research only), plus referee wrist wearables"},"url":"https://www.nba.com/news/nba-experiments-with-smart-basketballs-wrist-wearables","source":"NBA.com (AP)","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"deployment","typeLabel":{"zh":"落地部署","en":"Deployment"},"why":{"zh":"AI：与 Hawk-Eye 光学追踪配合，目标是自动判定出界、干扰球等客观判罚；学科：篮球判罚自动化进入实测阶段，但联盟明确本季不用于正式判罚","en":"AI: paired with Hawk-Eye tracking, aimed at automating objective calls such as out-of-bounds. Field: officiating automation reaches on-court testing, explicitly not used for calls this season"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：与 Hawk-Eye 光学追踪配合，目标是自动判定出界、干扰球等客观判罚；学科：篮球判罚自动化进入实测阶段，但联盟明确本季不用于正式判罚","en":"AI: paired with Hawk-Eye tracking, aimed at automating objective calls such as out-of-bounds. Field: officiating automation reaches on-court testing, explicitly not used for calls this season"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：与 Hawk-Eye 光学追踪配合，目标是自动判定出界、干扰球等客观判罚；学科：篮球判罚自动化进入实测阶段，但联盟明确本季不用于正式判罚","en":"AI: paired with Hawk-Eye tracking, aimed at automating objective calls such as out-of-bounds. Field: officiating automation reaches on-court testing, explicitly not used for calls this season"}},"impactNotes":[],"verified":null,"addedIn":"v2-20261022audit"},{"date":"2026-10-08","title":{"zh":"ABA 联赛（亚得里亚海篮球联赛）裁判部开始用 AI 分析每轮判罚：把裁判委员会对关键判罚的评估与官方技术统计、逐回合记录对照，生成面向球队和裁判的轮次报告","en":"ABA League refereeing department begins using AI to analyse officiating each round, linking key-play assessments with official stats and play-by-play"},"url":"https://www.aba-liga.com/news/55261","source":"ABA League","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"deployment","typeLabel":{"zh":"落地部署","en":"Deployment"},"why":{"zh":"AI：汇总最后两分钟犯规、关键判罚高频问题、程序性意见和改进重点；学科：AI 从“辅助判罚”延伸到“评估裁判”，报告不给个人打分、不作处罚","en":"AI: summarises last-two-minute fouls, recurring key-play themes and development priorities. Field: AI extends from assisting calls to reviewing officiating; no individual ratings or sanctions"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：汇总最后两分钟犯规、关键判罚高频问题、程序性意见和改进重点；学科：AI 从“辅助判罚”延伸到“评估裁判”，报告不给个人打分、不作处罚","en":"AI: summarises last-two-minute fouls, recurring key-play themes and development priorities. Field: AI extends from assisting calls to reviewing officiating; no individual ratings or sanctions"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：汇总最后两分钟犯规、关键判罚高频问题、程序性意见和改进重点；学科：AI 从“辅助判罚”延伸到“评估裁判”，报告不给个人打分、不作处罚","en":"AI: summarises last-two-minute fouls, recurring key-play themes and development priorities. Field: AI extends from assisting calls to reviewing officiating; no individual ratings or sanctions"}},"impactNotes":[],"verified":null,"addedIn":"v2-20261022audit"}],"papersRepos":[{"title":"TacticAI: an AI assistant for football tactics","url":"https://doi.org/10.1038/s41467-024-45965-x","date":"2024-03-19","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"Google DeepMind","url":"https://deepmind.google"}],"news":[{"id":"bd4f20eeee8d","date":"2026-10-19","title":"New enhanced digital experiences for 2026-27 season","url":"https://www.nba.com/news/nba-digital-experiences-2026-27","source":"NBA.com","lang":"en","kind":"news","summaryZh":"NBA 2026–27 赛季数字产品更新：NBA Blueprint 用 AI 为每场比赛生成赛前赛后报告，NBA App 新增 AI 辅助导航。","via":"manual-websearch-20261022","linkCheck":"ok"},{"id":"27098ea655a7","date":"2026-10-06","title":"Hawthorn Football Club drafts Google Cloud to improve Club analytics with AI","url":"https://blog.google/intl/en-au/company-news/technology/hawthorn-football-club-drafts-google-cloud-to-improve-club-analytics-with-ai/","source":"Google（澳大利亚博客）","lang":"en","kind":"news","summaryZh":"澳式足球霍桑俱乐部与 Google Cloud 合作，用 Gemini 把比赛、GPS、医疗和球探数据整合进对话式分析平台（厂商自述）。","via":"manual-websearch-20261022","linkCheck":"ok"},{"id":"16a889e8ce03","date":"2026-09-30","title":"Soccer-playing humanoid robots score a major robotics goal for coordination","url":"https://techxplore.com/news/2026-09-soccer-playing-humanoid-robots-score.html","source":"TechXplore 机器人","lang":"en","kind":"news","summaryZh":"TechXplore 机器人于2026-09-30发布的文章，涉及AI 机器人，原题《Soccer-playing humanoid robots score a major robotics goal for coordination》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"5f5d38071779","date":"2026-09-11","title":"RSC Anderlecht sign with MyGamePlan","url":"https://www.mygameplan.ai/blog/rsc-anderlecht-sign-with-mygameplan","source":"MyGamePlan","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"ok"},{"id":"2733c7ac8961","date":"2026-08-28","title":"First 11 vs 11 humanoid soccer game played at RoboCup 2026","url":"https://robohub.org/first-11-vs-11-humanoid-soccer-game-played-at-robocup-2026/","source":"Robohub","lang":"en","kind":"news","summaryZh":"Robohub于2026-08-28发布的文章，涉及AI 机器人、AI 创意媒体，原题《First 11 vs 11 humanoid soccer game played at RoboCup 2026》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"}],"review":{"lastReviewed":"2026-10-22","cycleDays":90,"nextReview":"2027-01-19"},"urls":{"zh":"/zh-hans/disciplines/sports-science/","en":"/en/disciplines/sports-science/"},"dimensions":{"method":{"score":2,"label":{"zh":"研究方法","en":"Research method"},"evidence":"追踪数据与计算机视觉分析在顶级俱乐部普遍","source":null,"weight":0.3},"discovery":{"score":1,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"TacticAI（Nat. Commun. 2024）是少数顶刊成果，属战术建议工具；2026-08 爱尔兰橄榄球多队多赛季研究显示机器学习伤病预测 AUC 仅 0.66（Sports Med Open），说明预测类应用仍不可用","source":null,"weight":0.35},"adoption":{"score":2,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"半自动越位技术用于世界杯；墨西哥联赛、安德莱赫特等采用 AI 分析平台","source":"https://www.mygameplan.ai/blog/rsc-anderlecht-sign-with-mygameplan","weight":0.25},"education":{"score":0,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到证据","source":null,"weight":0.1}},"compositeScore":1.45,"levelCapped":false,"levelChange":{"from":2,"to":1,"direction":"down","reason":{"zh":"下调为 L1：AI 在体育中主要是追踪、视频分析和判罚辅助（如 2022 世界杯半自动越位）；以 AI 为关键的研究成果只有 TacticAI（Nat. Commun. 2024）等少数，且属战术建议工具，尚不足以称“局部突破”。","en":"Lowered to L1: AI in sport is mainly tracking, video analysis and officiating aids (e.g. semi-automated offside at the 2022 World Cup); AI-key research is limited to a few tools such as TacticAI (Nat. Commun. 2024)."}},"biggestBreakthrough":{"date":"2024-03-19","title":{"zh":"TacticAI：足球战术 AI 助手（Nat. Commun.）","en":"TacticAI: AI assistant for football tactics (Nat. Commun.)"},"url":"https://www.nature.com/articles/s41467-024-45965-x","aiImpact":"major","fieldImpact":"major","why":{"zh":"与利物浦合作，角球战术建议在盲评中 90% 被专家偏好","en":"Liverpool collaboration; experts preferred suggestions 90% of the time"}},"keyPapers":[{"title":"TacticAI: an AI assistant for football tactics","titleZh":"TacticAI：足球战术 AI 助手（Nat. Commun.）","id":"10.1038/s41467-024-45965-x","url":"https://doi.org/10.1038/s41467-024-45965-x","venue":"Nature Communications","date":"2024-03-19","why":{"zh":"与利物浦合作，角球战术建议在盲评中 90% 被专家偏好","en":"Liverpool collaboration; experts preferred suggestions 90% of the time"},"fieldImpact":"major","aiImpact":"major"},{"title":"Machine Learning Model Development and Evaluation for Non-contact Lower Limb Injury Risk Prediction in Elite Male Rugby Union: A Multi-team, Multi-season Analysis","titleZh":"爱尔兰四支职业橄榄球队 + 国家队 3 个赛季数据：五种机器学习模型预测下肢非接触伤病，最好的 CatBoost 也只有 AUC 0.66、精确率 1.3%，作者结论是“预测性能差”（Sports Medicine – Open）","id":"10.1186/s40798-026-01060-7","url":"https://doi.org/10.1186/s40798-026-01060-7","venue":"Sports Medicine - Open","date":"2026-08-26","why":{"zh":"AI：GNSS 负荷、主观状态和体能筛查数据输入逻辑回归、SVM、随机森林、XGBoost、CatBoost；学科：这类多队多赛季的负面结果说明“AI 预测伤病”离可用还很远，是判断体育学仍处 L1 的重要依据","en":"AI: GNSS load, wellness and screening data across five model types. Field: a rare multi-team negative result showing AI injury prediction is far from usable — key evidence for keeping sports science at L1"},"fieldImpact":"major","aiImpact":"incremental"}],"latest_papers":[{"id":"doi:10.4324/9781003757054-4","title":"Explainable AI for Sport Scientists","titleZh":"面向运动科学研究者的可解释 AI","summaryZh":"讨论可解释 AI 在运动表现分析、伤病预防和运动员管理中的作用，把可解释性视为一个连续谱。","summaryEn":"A chapter on explainable AI for sport science in performance analysis, injury prevention and athlete management.","authors":["Michal Bozděch"],"date":"2026-10-20","url":"https://doi.org/10.4324/9781003757054-4","site":"OpenAlex","venue":"Routledge eBooks","category":null,"fetchedAt":"2026-10-22T10:10:09"},{"id":"doi:10.4324/9781003757054-9","title":"Cross-Platform AI Modeling in Sports Science","titleZh":"运动科学中的跨平台 AI 建模","summaryZh":"以职业网球为例，用神经网络回归根据相对表现指标预测总奖金，演示 AI 辅助的数据决策。","summaryEn":"Uses neural-network regression on relative performance indicators to model professional tennis prize money.","authors":["Michal Bozděch"],"date":"2026-10-20","url":"https://doi.org/10.4324/9781003757054-9","site":"OpenAlex","venue":"Routledge eBooks","category":null,"fetchedAt":"2026-10-22T10:10:09"},{"id":"doi:10.22158/asir.v10n4p1","title":"Predicting Passing-Play Outcomes in American Football: A Comparative Study of Traditional Machine Learning Algorithms and Fine-Tuned Large Language Models","titleZh":"预测美式橄榄球传球进攻结果：传统机器学习与微调大模型比较","summaryZh":"基于 100 场 NFL 比赛，比较 SVM、XGBoost 等与 LoRA 微调的 Llama 3 预测传球码数和达阵，SVM 平均绝对误差约 7 码最佳。","summaryEn":"Compares classic ML and a LoRA-tuned Llama 3 on 100 NFL games; SVM best (MAE ≈ 7 yards).","authors":["Tianjian (Jason) Wang"],"date":"2026-10-19","url":"https://doi.org/10.22158/asir.v10n4p1","site":"OpenAlex","venue":"Applied Science and Innovative Research","category":null,"fetchedAt":"2026-10-22T10:10:09"},{"id":"doi:10.3389/fpsyg.2026.1951652","title":"AI feedback does not motivate by itself: a motivational-meaning model for university physical education","titleZh":"AI 反馈本身并不激励：大学体育课的动机-意义模型","summaryZh":"提出模型说明 AI 把动作转为评分和纠正提示后，学习者如何理解这种反馈取决于教学框架和社会情境。","summaryEn":"A model of how university PE learners interpret AI movement feedback.","authors":["Xiangyu Wang","Yongwei Wang"],"date":"2026-10-19","url":"https://doi.org/10.3389/fpsyg.2026.1951652","site":"OpenAlex","venue":"Frontiers in Psychology","category":null,"fetchedAt":"2026-10-22T10:10:09"},{"id":"doi:10.3390/jfmk11040431","title":"Determining the Level of Reaction to a Moving Object in Combat Sports Athletes Using a Composite Index and Machine Learning: A Cross-Sectional Methodological Study","titleZh":"用复合指数和机器学习评估格斗运动员对移动物体的反应水平","summaryZh":"对 312 名 10–25 岁格斗运动员建立复合指数并开发机器学习自动分类模型。","summaryEn":"Composite index and ML classification of moving-object reaction in 312 combat-sport athletes.","authors":["Vyacheslav Romanenko","Y. Tropin","Rashid Matkarimov","Georgiy Korobeynikov"],"date":"2026-10-18","url":"https://doi.org/10.3390/jfmk11040431","site":"OpenAlex","venue":"Journal of Functional Morphology and Kinesiology","category":null,"fetchedAt":"2026-10-22T10:10:09"},{"id":"doi:10.1145/3841455.3841534","title":"Event Detection in Table Tennis Videos using 2D Keypoints","titleZh":"基于二维关键点的乒乓球视频事件检测","summaryZh":"提出 EventNet 两阶段流程，用球员姿态、球桌和球的关键点，在长视频中逐帧检测击球等关键事件。","summaryEn":"EventNet detects frame-accurate table-tennis events from 2D keypoints.","authors":["Rainer W. Lienhart","Daniel Kienzle","Shin’ichi Satoh","Anastasiia Bilinska"],"date":"2026-10-17","url":"https://doi.org/10.1145/3841455.3841534","site":"OpenAlex","venue":"?","category":null,"fetchedAt":"2026-10-22T10:10:09"},{"id":"doi:10.5281/zenodo.23432801","title":"Machine Learning for Injury Risk Prediction and Prevention in Athletics: A Data Driven Framework Integrating Wearable Sensor Metrics, Neuromuscular Screening, and Ensemble Learning for Prospective Injury Surveillance in Collegiate Athletes","titleZh":"机器学习预测田径运动损伤风险","summaryZh":"整合训练负荷、神经肌肉筛查和可穿戴数据，用集成学习预测大学运动员非接触性肌肉骨骼损伤。","summaryEn":"Ensemble ML with wearable and screening data to predict non-contact injuries in collegiate athletes.","authors":["Sambit Bhattacharjee","Sagnik Bhattacharjee","Pulen Das"],"date":"2026-10-17","url":"https://doi.org/10.5281/zenodo.23432801","site":"OpenAlex","venue":"Zenodo (CERN European Organization for Nuclear Research)","category":null,"fetchedAt":"2026-10-22T10:10:09"},{"id":"doi:10.1038/s41598-026-76341-y","title":"Exploring foot pressure patterns for skiing experience classification in a simulator-based environment","titleZh":"在模拟器环境中用足底压力模式分类滑雪水平","summaryZh":"用基于 YOLO11 的分类器从 27 名参与者的足底压力图像判别初学、新手和中级滑雪水平。","summaryEn":"A YOLO11 classifier infers skiing proficiency from plantar pressure in a simulator (27 participants).","authors":["Shigeharu Ono","Ryosuke Atsumi","Hideaki Kanai","Hideki Koike"],"date":"2026-10-16","url":"https://doi.org/10.1038/s41598-026-76341-y","site":"OpenAlex","venue":"Scientific Reports","category":null,"fetchedAt":"2026-10-22T10:10:09"}]},{"id":"history","slug":"history","category":"humanities","categoryName":{"zh":"人文","en":"Humanities"},"name":{"zh":"历史学","en":"History"},"classification":{"oecdFos":"6.1","moe":"0602 中国史 / 0603 世界史"},"level":1,"levelKey":"assistive","levelLabel":{"zh":"辅助起步","en":"Assistive"},"trend":"up","reason":{"zh":"手写文字识别（如 Transkribus）大幅加快档案转录；2026 年 Schmidt Sciences 资助 1000 万美元的 AI 与人文研究计划，但历史解释仍完全由人完成。","en":"Handwritten text recognition (e.g., Transkribus) speeds up archive transcription; Schmidt Sciences funded a $10M AI-and-humanities programme in 2026, but historical interpretation remains human."},"reasonSources":[{"url":"https://www.transkribus.org/","linkCheck":"ok"},{"url":"https://www.schmidtsciences.org/havi-2026-announcement/","linkCheck":"ok"}],"directions":{"zh":["手写档案识别","历史文献检索与问答","数字人文"],"en":["Handwritten archive recognition","Historical retrieval & QA","Digital humanities"]},"stats":{"milestones":6,"milestonesSince2025":4,"milestones2026":2,"milestonesLast6Months":1,"news":3,"latestProgress":"2026-10-20","latestMilestone":"2026-10-14","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":1,"leaps":{"ai":0,"field":0},"latestPapers":4,"keyPapers":3},"milestones":[{"date":"2019-09-09","title":{"zh":"用手写文字识别改造档案研究（Journal of Documentation，Transkribus）","en":"Transforming scholarship in the archives through handwritten text recognition (J. Documentation)"},"url":"https://doi.org/10.1108/JD-07-2018-0114","source":"Journal of Documentation","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"HTR 平台 Transkribus 的奠基性论文","en":"Foundational Transkribus paper"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"HTR 平台 Transkribus 的奠基性论文","en":"Foundational Transkribus paper"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"HTR 平台 Transkribus 的奠基性论文","en":"Foundational Transkribus paper"}},"impactNotes":[],"verified":{"id":"10.1108/JD-07-2018-0114","title":"Transforming scholarship in the archives through handwritten text recognition","venue":"Journal of Documentation","date":"2019-09-09"},"addedIn":"v2"},{"date":"2023-08-24","title":{"zh":"American Stories：用深度学习版面识别与 OCR 从美国国会图书馆 2000 万页历史报纸中提取结构化文本","en":"American Stories: structured text from ~20M pages of historical US newspapers via deep-learning layout + OCR (arXiv 2308.12477)"},"url":"https://arxiv.org/abs/2308.12477","source":"arXiv","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok-crossref","datePrecision":"day","type":"dataset","typeLabel":{"zh":"数据集","en":"Dataset"},"why":{"zh":"把大规模历史报纸变成可计算文本，支撑定量史学","en":"Makes historical newspapers computable at scale"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"把大规模历史报纸变成可计算文本，支撑定量史学","en":"Makes historical newspapers computable at scale"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"把大规模历史报纸变成可计算文本，支撑定量史学","en":"Makes historical newspapers computable at scale"}},"impactNotes":[],"verified":{"id":"arXiv:2308.12477","title":"American Stories: A Large-Scale Structured Text Dataset of Historical U.S. Newspapers","venue":"arXiv","date":"2023-08-24"},"addedIn":"v2"},{"date":"2025-01-20","title":{"zh":"READ-COOP 与 Transkribus：社区共享基础设施，约 9000 万张历史文本图像已识别（Open Research Europe）","en":"READ-COOP and Transkribus: shared infrastructure for automated text recognition (Open Research Europe)"},"url":"https://doi.org/10.12688/openreseurope.18747.1","source":"Open Research Europe","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"HTR 达到档案馆规模","en":"HTR at archive scale"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"HTR 达到档案馆规模","en":"HTR at archive scale"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"HTR 达到档案馆规模","en":"HTR at archive scale"}},"impactNotes":[],"verified":{"id":"10.12688/openreseurope.18747.1","title":"The artificial intelligence cooperative: READ-COOP, Transkribus, and the benefits of shared community infrastructure for automated text recognition","venue":"Open Research Europe","date":"2025-01-20"},"addedIn":"v2"},{"date":"2025-04","title":{"zh":"瑞典国家档案馆用 Swedish Lion HTR 模型让 120 万页手写档案可检索","en":"Swedish National Archives makes 1.2M handwritten pages searchable with its HTR model"},"url":"https://riksarkivet.se/inlagg/riksarkivet-gor-en-miljon-handskrivna-dokument-sokbara","source":"Riksarkivet","flags":["tbv"],"flagLabels":[{"zh":"日期或细节待核实","en":"Date/details to be verified"}],"linkCheck":"ok","datePrecision":"month","type":"deployment","typeLabel":{"zh":"落地部署","en":"Deployment"},"why":{"zh":"国家档案馆级别的 AI 转写部署","en":"National-archive-scale deployment"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"国家档案馆级别的 AI 转写部署","en":"National-archive-scale deployment"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"国家档案馆级别的 AI 转写部署","en":"National-archive-scale deployment"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-01-29","title":{"zh":"CoMMA 项目四个月用 AI 转写 32763 部中世纪手稿","en":"CoMMA transcribes 32,763 medieval manuscripts with AI in four months"},"url":"https://www.medievalists.net/2026/01/32000-medieval-manuscripts-transcribed-using-ai/","source":"Medievalists.net","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"dataset","typeLabel":{"zh":"数据集","en":"Dataset"},"why":{"zh":"中世纪研究史料获取规模跃升","en":"Step change in access to medieval sources"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"中世纪研究史料获取规模跃升","en":"Step change in access to medieval sources"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"中世纪研究史料获取规模跃升","en":"Step change in access to medieval sources"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-10-14","title":{"zh":"Schmidt Sciences 资助 1000 万美元 AI 与人文研究计划（19 个项目）","en":"Schmidt Sciences awards $10M to AI & humanities programme (19 projects)"},"url":"https://www.schmidtsciences.org/havi-2026-announcement/","source":"Schmidt Sciences","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"policy","typeLabel":{"zh":"政策","en":"Policy"},"why":{"zh":"资助 19 个 AI 与人文研究项目","en":"Funds 19 AI-and-humanities projects"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"资助 19 个 AI 与人文研究项目","en":"Funds 19 AI-and-humanities projects"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"资助 19 个 AI 与人文研究项目","en":"Funds 19 AI-and-humanities projects"}},"impactNotes":[],"verified":null,"addedIn":"v1"}],"papersRepos":[{"title":"Transforming scholarship in the archives through handwritten text recognition","url":"https://doi.org/10.1108/JD-07-2018-0114","date":"2019-09-09","kind":"paper","linkCheck":"ok-crossref"},{"title":"American Stories: A Large-Scale Structured Text Dataset of Historical U.S. Newspapers","url":"https://arxiv.org/abs/2308.12477","date":"2023-08-24","kind":"paper","linkCheck":"ok-crossref"},{"title":"The artificial intelligence cooperative: READ-COOP, Transkribus, and the benefits of shared community infrastructure for automated text recognition","url":"https://doi.org/10.12688/openreseurope.18747.1","date":"2025-01-20","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"Transkribus / READ-COOP","url":"https://www.transkribus.org/"},{"name":"Schmidt Sciences","url":"https://www.schmidtsciences.org"}],"news":[{"id":"99e0b4d799dd","date":"2026-10-20","title":"European National Archives meet in Dublin Castle to discuss the impact of AI and New Technologies on state archives and records","url":"https://irish-presidency.consilium.europa.eu/en/news/european-national-archives-meet-in-dublin-castle-to-discuss-the-impact-of-ai-and-new-technologies-on-state-archives-and-records/","source":"Irish Presidency of the Council of the EU","lang":"en","kind":"news","summaryZh":"欧盟各成员国国家档案馆在都柏林城堡开会，讨论 AI 对国家档案与记录工作的影响，并研究设立欧盟 AI 观察站/协作技术实验室以保障数字档案真实性。","via":"manual-websearch-20261022","linkCheck":"ok"},{"id":"7a5a95d26ba6","date":"2026-10","title":"Schmidt Sciences awards $10 million to one of the world’s largest AI and humanities research programs","url":"https://www.schmidtsciences.org/havi-2026-announcement/","source":"Schmidt Sciences","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"ok"},{"id":"384a0e6d528f","date":"2026-01-29","title":"32,000 medieval manuscripts transcribed using AI","url":"https://www.medievalists.net/2026/01/32000-medieval-manuscripts-transcribed-using-ai/","source":"Medievalists.net","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"ok"}],"review":{"lastReviewed":"2026-10-21","cycleDays":90,"nextReview":"2027-01-19"},"urls":{"zh":"/zh-hans/disciplines/history/","en":"/en/disciplines/history/"},"dimensions":{"method":{"score":2,"label":{"zh":"研究方法","en":"Research method"},"evidence":"手写文字识别（HTR）已是档案数字化标准工具","source":"https://doi.org/10.12688/openreseurope.18747.1","weight":0.3},"discovery":{"score":1,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"AI 加速了史料获取，历史解释仍由人完成","source":null,"weight":0.35},"adoption":{"score":1,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"大型档案馆采用，历史学者个人使用率无数据","source":null,"weight":0.25},"education":{"score":0,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到证据","source":null,"weight":0.1}},"compositeScore":1.2,"levelCapped":false,"levelChange":{"from":1,"to":1,"direction":"same","reason":null},"biggestBreakthrough":{"date":"2026-10-14","title":{"zh":"Schmidt Sciences 资助 1000 万美元 AI 与人文研究计划（19 个项目）","en":"Schmidt Sciences awards $10M to AI & humanities programme (19 projects)"},"url":"https://www.schmidtsciences.org/havi-2026-announcement/","aiImpact":"incremental","fieldImpact":"major","why":{"zh":"资助 19 个 AI 与人文研究项目","en":"Funds 19 AI-and-humanities projects"}},"keyPapers":[{"title":"Transforming scholarship in the archives through handwritten text recognition","titleZh":"用手写文字识别改造档案研究（Journal of Documentation，Transkribus）","id":"10.1108/JD-07-2018-0114","url":"https://doi.org/10.1108/JD-07-2018-0114","venue":"Journal of Documentation","date":"2019-09-09","why":{"zh":"HTR 平台 Transkribus 的奠基性论文","en":"Foundational Transkribus paper"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"American Stories: A Large-Scale Structured Text Dataset of Historical U.S. Newspapers","titleZh":"American Stories：用深度学习版面识别与 OCR 从美国国会图书馆 2000 万页历史报纸中提取结构化文本","id":"arXiv:2308.12477","url":"https://arxiv.org/abs/2308.12477","venue":"arXiv","date":"2023-08-24","why":{"zh":"把大规模历史报纸变成可计算文本，支撑定量史学","en":"Makes historical newspapers computable at scale"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"The artificial intelligence cooperative: READ-COOP, Transkribus, and the benefits of shared community infrastructure for automated text recognition","titleZh":"READ-COOP 与 Transkribus：社区共享基础设施，约 9000 万张历史文本图像已识别（Open Research Europe）","id":"10.12688/openreseurope.18747.1","url":"https://doi.org/10.12688/openreseurope.18747.1","venue":"Open Research Europe","date":"2025-01-20","why":{"zh":"HTR 达到档案馆规模","en":"HTR at archive scale"},"fieldImpact":"major","aiImpact":"incremental"}],"latest_papers":[{"id":"doi:10.1017/s2058631026101329","title":"Humanist learning and artificial intelligences: critical literacies for digital worlds","titleZh":"人文学习与人工智能：数字世界的批判素养","summaryZh":"讨论生成式 AI 对古典学、历史学和考古学的方法与教学的影响，呼吁在学科内部批判性反思 AI 的使用。","summaryEn":"Calls for critical reflection on generative AI in classics, history and archaeology teaching and methods.","authors":["Christine L. Johnston"],"date":"2026-10-19","url":"https://doi.org/10.1017/s2058631026101329","site":"OpenAlex","venue":"˜The œjournal of classics teaching","category":null,"fetchedAt":"2026-10-22T10:10:13"},{"id":"arXiv:2610.25440","title":"Design Considerations in Low-Resource Archival OCR: A Case Study in Lakota","titleZh":"低资源档案 OCR 的设计考量：以拉科塔语为例","summaryZh":"用 35 页转写数据微调小型视觉语言模型做 OCR 后校正，使双语档案的字符错误率从 17.1% 降至 3.9%。","summaryEn":"Fine-tuning small VLMs for OCR post-correction on 35 pages cuts Lakota archival CER from 17.1% to 3.9%.","authors":["Téa Wright","Laura X. Ma","Claudia Iron Hawk","Hannah J. Haynie"],"date":"2026-10-18","url":"https://arxiv.org/abs/2610.25440","site":"arXiv","venue":null,"category":"cs.CL","fetchedAt":"2026-10-22T10:10:13"},{"id":"arXiv:2610.25235","title":"Evaluating Cloud and Self-Hosted OCR Systems for Arabic, English, and French Documents: Bounding Box Detection, Text Recognition, and Mathematical Expression Recognition","titleZh":"评估阿拉伯语、英语和法语文档的云端与自托管 OCR 系统","summaryZh":"在 315 页九类文档上比较多种 OCR 系统的版面检测、文字识别和数学公式识别，为多语言档案选型提供依据。","summaryEn":"Compares cloud and self-hosted OCR on Arabic, English and French documents for multilingual archives.","authors":["Jumana Alsubhi","Ahmed Alhusayni","Abdulrahman Gharawi","Israa Hamdine"],"date":"2026-10-18","url":"https://arxiv.org/abs/2610.25235","site":"arXiv","venue":null,"category":"cs.CV","fetchedAt":"2026-10-22T10:10:13"},{"id":"doi:10.17485/ijst/v19i35.970","title":"Enhanced Script Identification for Multilingual Document Analysis: A Dataset and Machine Learning Approach for Kashmiri and English","titleZh":"多语言文档分析的文字识别增强：克什米尔语与英语数据集及机器学习方法","summaryZh":"构建 4423 张克什米尔语（波斯-阿拉伯字母）与英语印刷文本图像数据集，用 HOG 特征和四种分类器识别文字类型。","summaryEn":"A 4,423-image Kashmiri/English script dataset with HOG features and four classifiers.","authors":["Rafia Amin","Rumaan Bashir"],"date":"2026-10-18","url":"https://doi.org/10.17485/ijst/v19i35.970","site":"OpenAlex","venue":"Indian Journal of Science and Technology","category":null,"fetchedAt":"2026-10-22T10:10:13"}]},{"id":"archaeology-classics","slug":"archaeology-classics","category":"humanities","categoryName":{"zh":"人文","en":"Humanities"},"name":{"zh":"考古学与古典文献","en":"Archaeology & classics"},"classification":{"oecdFos":"6.1","moe":"0601 考古学"},"level":2,"levelKey":"breakthrough","levelLabel":{"zh":"局部突破","en":"Local breakthroughs"},"trend":"up","reason":{"zh":"Ithaca（Nature 2022）与 Aeneas（Nature 2025）修复、断代古代铭文；维苏威挑战赛用 AI 读出碳化的赫库兰尼姆纸草（2024 年大奖，2026 年读出 PHerc. 1667 约 1.5 米希腊文）；AI 辅助在纳斯卡发现 303 处新地画（PNAS 2024）。","en":"Ithaca (Nature 2022) and Aeneas (Nature 2025) restore and date inscriptions; the Vesuvius Challenge read carbonised Herculaneum papyri (2024 prize; ~1.5 m of Greek from PHerc. 1667 in 2026); AI helped find 303 new Nazca geoglyphs (PNAS 2024)."},"reasonSources":[{"url":"https://www.nature.com/articles/s41586-022-04448-z","linkCheck":"ok"},{"url":"https://www.nature.com/articles/s41586-025-09292-5","linkCheck":"ok"},{"url":"https://scrollprize.org/grandprize","linkCheck":"ok"},{"url":"https://www.pnas.org/doi/10.1073/pnas.2407652121","linkCheck":"ok-crossref"}],"directions":{"zh":["古文字修复与断代","虚拟展开与墨迹检测","遥感遗址发现","器物拼合"],"en":["Inscription restoration & dating","Virtual unwrapping & ink detection","Remote-sensing site discovery","Artefact refitting"]},"stats":{"milestones":7,"milestonesSince2025":4,"milestones2026":3,"milestonesLast6Months":3,"news":5,"latestProgress":"2026-10-15","latestMilestone":"2026-10-10","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":0,"leaps":{"ai":0,"field":2},"latestPapers":6,"keyPapers":4},"milestones":[{"date":"2022-03-09","title":{"zh":"Ithaca 修复并断代古希腊铭文（Nature）","en":"Ithaca restores and dates Greek inscriptions (Nature)"},"url":"https://www.nature.com/articles/s41586-022-04448-z","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"修复古希腊铭文准确率 62%，与历史学家合作时提升到 72%","en":"62% restoration accuracy, 72% with historians"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"修复古希腊铭文准确率 62%，与历史学家合作时提升到 72%","en":"62% restoration accuracy, 72% with historians"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"修复古希腊铭文准确率 62%，与历史学家合作时提升到 72%","en":"62% restoration accuracy, 72% with historians"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-022-04448-z","title":"Restoring and attributing ancient texts using deep neural networks","venue":"Nature","date":"2022-03-09"},"addedIn":"v1"},{"date":"2024-02-05","title":{"zh":"维苏威挑战赛大奖：AI 读出赫库兰尼姆纸草段落","en":"Vesuvius Challenge Grand Prize: AI reads Herculaneum papyrus passages"},"url":"https://scrollprize.org/grandprize","source":"Vesuvius Challenge","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"competition","typeLabel":{"zh":"竞赛","en":"Competition"},"why":{"zh":"AI 首次从未展开的碳化卷轴读出成段文字","en":"First passages read from an unopened carbonised scroll"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI 首次从未展开的碳化卷轴读出成段文字","en":"First passages read from an unopened carbonised scroll"}},"fieldImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"AI 首次从未展开的碳化卷轴读出成段文字","en":"First passages read from an unopened carbonised scroll"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2024-09-23","title":{"zh":"AI 辅助在纳斯卡发现 303 处新地画（PNAS）","en":"AI helps find 303 new Nazca geoglyphs (PNAS)"},"url":"https://www.pnas.org/doi/10.1073/pnas.2407652121","source":"PNAS","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"AI 辅助数月内发现 303 处新地画，几乎使已知数量翻倍","en":"303 new geoglyphs, nearly doubling the known count"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI 辅助数月内发现 303 处新地画，几乎使已知数量翻倍","en":"303 new geoglyphs, nearly doubling the known count"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI 辅助数月内发现 303 处新地画，几乎使已知数量翻倍","en":"303 new geoglyphs, nearly doubling the known count"}},"impactNotes":[],"verified":{"id":"10.1073/pnas.2407652121","title":"AI-accelerated Nazca survey nearly doubles the number of known figurative geoglyphs and sheds light on their purpose","venue":"Proceedings of the National Academy of Sciences","date":"2024-09-23"},"addedIn":"v1"},{"date":"2025-07-23","title":{"zh":"Aeneas：拉丁铭文语境化与修复（Nature）","en":"Aeneas: contextualising and restoring Latin inscriptions (Nature)"},"url":"https://www.nature.com/articles/s41586-025-09292-5","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"拉丁铭文语境化、修复与断代","en":"Latin inscription contextualisation"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"拉丁铭文语境化、修复与断代","en":"Latin inscription contextualisation"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"拉丁铭文语境化、修复与断代","en":"Latin inscription contextualisation"}},"impactNotes":[],"verified":{"id":"10.1038/s41586-025-09292-5","title":"Contextualizing ancient texts with generative neural networks","venue":"Nature","date":"2025-07-23"},"addedIn":"v1"},{"date":"2026-06-25","title":{"zh":"维苏威挑战赛：首次完整读出一整卷赫库兰尼姆纸草（PHerc. 1667，约 20 栏、近 1.5 米）","en":"Vesuvius Challenge: first entire Herculaneum scroll read (PHerc. 1667, ~20 columns, ~1.5 m)"},"url":"https://scrollprize.org/firstscroll","source":"Vesuvius Challenge","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"discovery","typeLabel":{"zh":"科学发现","en":"Discovery"},"why":{"zh":"近 2000 年前被火山碳化、无法展开的卷轴整卷被读出，开启大规模阅读古代失传文本的可能","en":"A carbonised 2,000-year-old scroll read in full without opening"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"近 2000 年前被火山碳化、无法展开的卷轴整卷被读出，开启大规模阅读古代失传文本的可能","en":"A carbonised 2,000-year-old scroll read in full without opening"}},"fieldImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"近 2000 年前被火山碳化、无法展开的卷轴整卷被读出，开启大规模阅读古代失传文本的可能","en":"A carbonised 2,000-year-old scroll read in full without opening"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-06-26","title":{"zh":"AI 与 X 射线读出 PHerc. 1667 纸草约 1.5 米希腊文","en":"AI and X-rays recover ~1.5 m of Greek text from PHerc. 1667"},"url":"https://www.cnn.com/2026/06/26/science/papyrus-scroll-vesuvius-ai-scli-intl","source":"CNN","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"discovery","typeLabel":{"zh":"科学发现","en":"Discovery"},"why":{"zh":"媒体报道同一成果","en":"Media coverage of the same result"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"媒体报道同一成果","en":"Media coverage of the same result"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"媒体报道同一成果","en":"Media coverage of the same result"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-10-10","title":{"zh":"PNAS Nexus：在一处遗址训练的哨兵-2 作物痕迹识别模型可迁移到其他欧洲遗址，迁移效果取决于植被光谱几何相似度","en":"PNAS Nexus: Sentinel-2 cropmark detection models trained at one archaeological site transfer to others, governed by spectral-geometric similarity"},"url":"https://doi.org/10.1093/pnasnexus/pgag336","source":"PNAS Nexus","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"AI：用受约束的 SVM/XGBoost 和方向强度分数解释跨遗址迁移；学科：挑战“作物痕迹探测高度依赖单一遗址”的假设，为用全球卫星数据做区域性考古普查提供依据（三个遗址的初步证据）","en":"AI: constrained SVM/XGBoost with a directional strength score explains transfer. Field: challenges the assumption that cropmark detection is site-specific, supporting regional satellite prospection (three sites)"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：用受约束的 SVM/XGBoost 和方向强度分数解释跨遗址迁移；学科：挑战“作物痕迹探测高度依赖单一遗址”的假设，为用全球卫星数据做区域性考古普查提供依据（三个遗址的初步证据）","en":"AI: constrained SVM/XGBoost with a directional strength score explains transfer. Field: challenges the assumption that cropmark detection is site-specific, supporting regional satellite prospection (three sites)"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：用受约束的 SVM/XGBoost 和方向强度分数解释跨遗址迁移；学科：挑战“作物痕迹探测高度依赖单一遗址”的假设，为用全球卫星数据做区域性考古普查提供依据（三个遗址的初步证据）","en":"AI: constrained SVM/XGBoost with a directional strength score explains transfer. Field: challenges the assumption that cropmark detection is site-specific, supporting regional satellite prospection (three sites)"}},"impactNotes":[],"verified":{"id":"10.1093/pnasnexus/pgag336","title":"Beyond site-specific archaeological prospection: Spectral-geometry reveals cross-site cropmark signals in satellite data","venue":"PNAS Nexus","date":"2026-10-10"},"addedIn":"v2-20261022pm"}],"papersRepos":[{"title":"google-deepmind/ithaca","url":"https://github.com/google-deepmind/ithaca","date":null,"kind":"repo","linkCheck":"ok"},{"title":"PyPottery (arXiv 2610.02072)","url":"https://arxiv.org/abs/2610.02072","date":"2026-10","kind":"paper","linkCheck":"ok"},{"title":"Restoring and attributing ancient texts using deep neural networks","url":"https://doi.org/10.1038/s41586-022-04448-z","date":"2022-03-09","kind":"paper","linkCheck":"ok-crossref"},{"title":"AI-accelerated Nazca survey nearly doubles the number of known figurative geoglyphs and sheds light on their purpose","url":"https://doi.org/10.1073/pnas.2407652121","date":"2024-09-23","kind":"paper","linkCheck":"ok-crossref"},{"title":"Contextualizing ancient texts with generative neural networks","url":"https://doi.org/10.1038/s41586-025-09292-5","date":"2025-07-23","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"Vesuvius Challenge","url":"https://scrollprize.org"},{"name":"Google DeepMind","url":"https://deepmind.google"}],"news":[{"id":"8d8d27afc7bf","date":"2026-10-15","title":"Scientists decipher charred ancient scrolls buried by Vesuvius","url":"https://www.irishtimes.com/science/2026/10/15/scientists-decipher-charred-ancient-scrolls-buried-by-vesuvius/","source":"The Irish Times","lang":"en","kind":"news","summaryZh":"《爱尔兰时报》科普：借助 CT 扫描、虚拟展开和墨迹识别，学者读出赫库兰尼姆碳化纸草，并据此定位柏拉图墓的位置。","via":"manual-websearch-20261022","linkCheck":"ok"},{"id":"6ea25e81ca4e","date":"2026-10","title":"NYU Researchers Are Building AI to Piece Together Broken Archaeological Finds","url":"https://www.nyu.edu/about/news-publications/news/2026/october/nyu-researchers-are-building-ai-to-piece-together-broken-archaeo.html","source":"NYU","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"ok"},{"id":"be759ff6f859","date":"2026-10","title":"Bringing ancient Mesopotamia into the AI age","url":"https://content.gmu.edu/news/2026-10/bringing-ancient-mesopotamia-ai-age","source":"George Mason University","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"ok"},{"id":"b87ba284ce3a","date":"2026-06-26","title":"Papyrus scroll burnt to a crisp during Vesuvius eruption deciphered with help of AI","url":"https://www.cnn.com/2026/06/26/science/papyrus-scroll-vesuvius-ai-scli-intl","source":"CNN","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"ok"},{"id":"7157eeceef54","date":"2026-06-25","title":"First entire Herculaneum scroll read","url":"https://scrollprize.org/firstscroll","source":"Vesuvius Challenge","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"ok"}],"review":{"lastReviewed":"2026-10-22","cycleDays":90,"nextReview":"2027-01-19"},"urls":{"zh":"/zh-hans/disciplines/archaeology-classics/","en":"/en/disciplines/archaeology-classics/"},"dimensions":{"method":{"score":2,"label":{"zh":"研究方法","en":"Research method"},"evidence":"AI 用于铭文修复、遥感考古和碳化纸草虚拟展开","source":"https://www.nature.com/articles/s41586-025-09292-5","weight":0.3},"discovery":{"score":3,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"整卷赫库兰尼姆纸草被读出（2026-06），纳斯卡 303 处新地画（PNAS 2024）","source":"https://scrollprize.org/firstscroll","weight":0.35},"adoption":{"score":1,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"少数团队使用","source":null,"weight":0.25},"education":{"score":0,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到证据","source":null,"weight":0.1}},"compositeScore":1.9,"levelCapped":false,"levelChange":{"from":2,"to":2,"direction":"same","reason":null},"biggestBreakthrough":{"date":"2026-06-25","title":{"zh":"维苏威挑战赛：首次完整读出一整卷赫库兰尼姆纸草（PHerc. 1667，约 20 栏、近 1.5 米）","en":"Vesuvius Challenge: first entire Herculaneum scroll read (PHerc. 1667, ~20 columns, ~1.5 m)"},"url":"https://scrollprize.org/firstscroll","aiImpact":"major","fieldImpact":"leap","why":{"zh":"近 2000 年前被火山碳化、无法展开的卷轴整卷被读出，开启大规模阅读古代失传文本的可能","en":"A carbonised 2,000-year-old scroll read in full without opening"}},"keyPapers":[{"title":"Restoring and attributing ancient texts using deep neural networks","titleZh":"Ithaca 修复并断代古希腊铭文（Nature）","id":"10.1038/s41586-022-04448-z","url":"https://doi.org/10.1038/s41586-022-04448-z","venue":"Nature","date":"2022-03-09","why":{"zh":"修复古希腊铭文准确率 62%，与历史学家合作时提升到 72%","en":"62% restoration accuracy, 72% with historians"},"fieldImpact":"major","aiImpact":"major"},{"title":"AI-accelerated Nazca survey nearly doubles the number of known figurative geoglyphs and sheds light on their purpose","titleZh":"AI 辅助在纳斯卡发现 303 处新地画（PNAS）","id":"10.1073/pnas.2407652121","url":"https://doi.org/10.1073/pnas.2407652121","venue":"Proceedings of the National Academy of Sciences","date":"2024-09-23","why":{"zh":"AI 辅助数月内发现 303 处新地画，几乎使已知数量翻倍","en":"303 new geoglyphs, nearly doubling the known count"},"fieldImpact":"major","aiImpact":"major"},{"title":"Contextualizing ancient texts with generative neural networks","titleZh":"Aeneas：拉丁铭文语境化与修复（Nature）","id":"10.1038/s41586-025-09292-5","url":"https://doi.org/10.1038/s41586-025-09292-5","venue":"Nature","date":"2025-07-23","why":{"zh":"拉丁铭文语境化、修复与断代","en":"Latin inscription contextualisation"},"fieldImpact":"major","aiImpact":"major"},{"title":"Beyond site-specific archaeological prospection: Spectral-geometry reveals cross-site cropmark signals in satellite data","titleZh":"PNAS Nexus：在一处遗址训练的哨兵-2 作物痕迹识别模型可迁移到其他欧洲遗址，迁移效果取决于植被光谱几何相似度","id":"10.1093/pnasnexus/pgag336","url":"https://doi.org/10.1093/pnasnexus/pgag336","venue":"PNAS Nexus","date":"2026-10-10","why":{"zh":"AI：用受约束的 SVM/XGBoost 和方向强度分数解释跨遗址迁移；学科：挑战“作物痕迹探测高度依赖单一遗址”的假设，为用全球卫星数据做区域性考古普查提供依据（三个遗址的初步证据）","en":"AI: constrained SVM/XGBoost with a directional strength score explains transfer. Field: challenges the assumption that cropmark detection is site-specific, supporting regional satellite prospection (three sites)"},"fieldImpact":"incremental","aiImpact":"incremental"}],"latest_papers":[{"id":"doi:10.4995/var.2027.25826","title":"Design and cultural heritage in architecture in the age of artificial intelligence: generative models, HBIM, and evidence-based digital reconstruction","titleZh":"AI 时代建筑设计与文化遗产：生成模型、HBIM 与基于证据的数字重建","summaryZh":"强调生成式 AI 的结果只有经历史证据、空间与材料约束和保护要求验证后，才能成为建筑与遗产知识。","summaryEn":"Argues generative-AI outputs become heritage knowledge only when validated against historical evidence and conservation constraints in HBIM.","authors":["Ömer Faruk Pamak","Cemile Feyzan Şimşek"],"date":"2026-10-19","url":"https://doi.org/10.4995/var.2027.25826","site":"OpenAlex","venue":"Virtual Archaeology Review","category":null,"fetchedAt":"2026-10-22T10:10:26"},{"id":"doi:10.1038/s40494-026-03053-x","title":"An evidence-informed fine-tuning framework for traceable generative redesign of architectural heritage ornaments","titleZh":"建筑遗产纹样可溯源生成式再设计的证据驱动微调框架","summaryZh":"把经过来源审核的史料证据转为可计算的条件，约束和评估遗产纹样的生成式再设计。","summaryEn":"Evidence-informed fine-tuning converts audited heritage evidence into conditions for generative ornament redesign.","authors":["Jiale Zhang","Hanbin Luo","Jie Xu"],"date":"2026-10-17","url":"https://doi.org/10.1038/s40494-026-03053-x","site":"OpenAlex","venue":"npj Heritage Science","category":null,"fetchedAt":"2026-10-22T10:10:26"},{"id":"doi:10.31127/tuje.1967989","title":"Reconstructing Lost Urban Memory with Generative AI: An Oral-History Prompt-Engineering Framework","titleZh":"用生成式 AI 重建消失的城市记忆：口述史提示工程框架","summaryZh":"把 5 位原居民的口述史转为结构化提示，用生成式图像模型重现土耳其开塞利城堡外已消失街区的景象。","summaryEn":"Turns oral histories into prompts for generative AI to visualise a lost Kayseri neighbourhood.","authors":["Özlem Kevseroğlu","İrem Yaşar","Vildan Yıldırım"],"date":"2026-10-16","url":"https://doi.org/10.31127/tuje.1967989","site":"OpenAlex","venue":"Turkish Journal of Engineering","category":null,"fetchedAt":"2026-10-22T10:10:26"},{"id":"doi:10.1038/s40494-026-03055-9","title":"Digital preservation and automated roof damage detection for traditional village heritage in Guangdong using UAV photogrammetry and deep learning","titleZh":"用无人机摄影测量和深度学习对广东传统村落遗产做数字保存和屋顶损毁自动检测","summaryZh":"在 5 个广东传统村落采集 12762 张影像，基于 41213 个标注实例训练 YOLOv8 模型，mAP50 为 0.832。","summaryEn":"UAV photogrammetry plus YOLOv8 (mAP50 0.832) detects roof damage in five Guangdong heritage villages.","authors":["ZHU Zhengyang","Chun-Ming Hsieh","Xiaodong Liu","LI Xiangsi"],"date":"2026-10-16","url":"https://doi.org/10.1038/s40494-026-03055-9","site":"OpenAlex","venue":"npj Heritage Science","category":null,"fetchedAt":"2026-10-22T10:10:26"},{"id":"doi:10.1038/s40494-026-03060-y","title":"From fragmented narratives to spatio-narrative networks: a knowledge graph approach to Tianjin Grand Canal heritage","titleZh":"从碎片化叙事到时空叙事网络：天津大运河遗产知识图谱","summaryZh":"借助大模型辅助抽取，从多源文本为天津段大运河 98 处遗产点构建叙事知识图谱并做网络与空间分析。","summaryEn":"LLM-assisted extraction builds a narrative knowledge graph of 98 Grand Canal heritage sites in Tianjin.","authors":["Bolun Zhang","Jingwen Tian","Lihao Meng","Jiaying Li"],"date":"2026-10-16","url":"https://doi.org/10.1038/s40494-026-03060-y","site":"OpenAlex","venue":"npj Heritage Science","category":null,"fetchedAt":"2026-10-22T10:10:26"},{"id":"arXiv:2610.09439","title":"InscriptionOCR: A Dataset and Method for Understanding Inscriptions","titleZh":"InscriptionOCR：理解古代铭文的数据集与方法","summaryZh":"端到端 AI 框架处理石刻阿育王婆罗米文铭文：图像修复、文字识别、转写和神经机器翻译。","summaryEn":"End-to-end AI pipeline for Ashokan Brahmi stone inscriptions: restoration, OCR, transliteration and translation.","authors":["Jaidev Sanjay Khalane","Akbar Ali","V. N. Prabhakar","Shanmuganathan Raman"],"date":"2026-10-07","url":"https://arxiv.org/abs/2610.09439","site":"arXiv","venue":null,"category":"cs.CV","fetchedAt":"2026-10-22T10:10:26"}]},{"id":"philosophy","slug":"philosophy","category":"humanities","categoryName":{"zh":"人文","en":"Humanities"},"name":{"zh":"哲学","en":"Philosophy"},"classification":{"oecdFos":"6.3","moe":"0101 哲学"},"level":0,"levelKey":"not-started","levelLabel":{"zh":"尚未改造","en":"Not yet transformed"},"trend":"flat","reason":{"zh":"下调为 L0（尚未改造）：AI 在哲学中主要是研究对象；作为研究工具只有零星的文献综述尝试和实验哲学中的大模型实验，Erkenntnis 2026 论文认为大模型“直觉”还不能替代人类直觉。综合得分低于 0.75。","en":"Lowered to L0 (not yet transformed): AI is mainly an object of philosophical study; tool use is limited to sporadic literature-review attempts and LLM experiments in experimental philosophy. Composite score below 0.75."},"reasonSources":[{"url":"https://link.springer.com/article/10.1007/s10670-026-01136-6","linkCheck":"ok"}],"directions":{"zh":["文献综述辅助","实验哲学中的大模型","论证分析"],"en":["Literature-review aids","LLMs in experimental philosophy","Argument analysis"]},"stats":{"milestones":7,"milestonesSince2025":3,"milestones2026":3,"milestonesLast6Months":2,"news":4,"latestProgress":"2026-10-19","latestMilestone":"2026-10-19","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":0,"leaps":{"ai":0,"field":0},"latestPapers":7,"keyPapers":6},"milestones":[{"date":"2023-03-04","title":{"zh":"Chalmers：大语言模型可能有意识吗？（Boston Review 2023 / arXiv 2303.07103）","en":"Chalmers: Could a Large Language Model be Conscious? (arXiv 2303.07103)"},"url":"https://arxiv.org/abs/2303.07103","source":"arXiv / Boston Review","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"意识哲学主流学者系统评估 LLM，使 AI 意识成为正式议题","en":"Leading philosopher of mind makes LLM consciousness a mainstream question"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"意识哲学主流学者系统评估 LLM，使 AI 意识成为正式议题","en":"Leading philosopher of mind makes LLM consciousness a mainstream question"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"意识哲学主流学者系统评估 LLM，使 AI 意识成为正式议题","en":"Leading philosopher of mind makes LLM consciousness a mainstream question"}},"impactNotes":[],"verified":{"id":"arXiv:2303.07103","title":"Could a Large Language Model be Conscious?","venue":"arXiv / Boston Review, August 9, 2023","date":"2023-03-04"},"addedIn":"v2"},{"date":"2023-07-12","title":{"zh":"用哲学家丹尼特的著作微调 GPT-3，专家难以分辨机器与本人的回答（Mind & Language）","en":"Creating a large language model of a philosopher (Mind & Language)"},"url":"https://doi.org/10.1111/mila.12466","source":"Mind & Language","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"引发对 AI 能否“做哲学”的讨论","en":"Sparked debate on whether AI can do philosophy"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"引发对 AI 能否“做哲学”的讨论","en":"Sparked debate on whether AI can do philosophy"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"引发对 AI 能否“做哲学”的讨论","en":"Sparked debate on whether AI can do philosophy"}},"impactNotes":[],"verified":{"id":"10.1111/mila.12466","title":"Creating a large language model of a philosopher","venue":"Mind &amp; Language","date":"2023-07-12"},"addedIn":"v2"},{"date":"2023-08-17","title":{"zh":"Butlin、Long 等 19 位学者：从意识科学理论推导 AI 意识指标（arXiv 2308.08708）","en":"Butlin, Long et al.: Consciousness in AI — indicator properties from the science of consciousness (arXiv 2308.08708)"},"url":"https://arxiv.org/abs/2308.08708","source":"arXiv","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"把哲学与神经科学理论转化为可检验的指标清单","en":"Turns theories into a checklist of testable indicators"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"把哲学与神经科学理论转化为可检验的指标清单","en":"Turns theories into a checklist of testable indicators"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"把哲学与神经科学理论转化为可检验的指标清单","en":"Turns theories into a checklist of testable indicators"}},"impactNotes":[],"verified":{"id":"arXiv:2308.08708","title":"Consciousness in Artificial Intelligence: Insights from the Science of Consciousness","venue":"arXiv","date":"2023-08-17"},"addedIn":"v2"},{"date":"2024-11-04","title":{"zh":"Long、Sebo、Butlin、Chalmers 等：认真对待 AI 福利（arXiv 2411.00986）","en":"Long, Sebo, Butlin, Chalmers et al.: Taking AI Welfare Seriously (arXiv 2411.00986)"},"url":"https://arxiv.org/abs/2411.00986","source":"arXiv","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"把 AI 道德地位从思想实验推进到机构政策建议","en":"Moves AI moral status from thought experiment to policy recommendations"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"把 AI 道德地位从思想实验推进到机构政策建议","en":"Moves AI moral status from thought experiment to policy recommendations"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"把 AI 道德地位从思想实验推进到机构政策建议","en":"Moves AI moral status from thought experiment to policy recommendations"}},"impactNotes":[],"verified":{"id":"arXiv:2411.00986","title":"Taking AI Welfare Seriously","venue":"arXiv","date":"2024-11-04"},"addedIn":"v2"},{"date":"2026-02-12","title":{"zh":"Daily Nous：AI 能否写出有用的哲学文献综述","en":"Daily Nous: can AI write a useful philosophical literature review?"},"url":"https://dailynous.com/2026/02/12/can-ai-write-a-useful-philosophical-literature-review-guest-post/","source":"Daily Nous","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"industry","typeLabel":{"zh":"产业/采用","en":"Industry"},"why":{"zh":"哲学界讨论 AI 写文献综述的可用性","en":"Debate on AI literature reviews"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"哲学界讨论 AI 写文献综述的可用性","en":"Debate on AI literature reviews"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"哲学界讨论 AI 写文献综述的可用性","en":"Debate on AI literature reviews"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-07-10","title":{"zh":"Erkenntnis：人工直觉的哲学意义","en":"Erkenntnis: philosophical significance of artificial intuitions"},"url":"https://link.springer.com/article/10.1007/s10670-026-01136-6","source":"Erkenntnis","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"论证大模型“直觉”尚不能替代人类直觉","en":"Argues LLM intuitions cannot yet replace human ones"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"论证大模型“直觉”尚不能替代人类直觉","en":"Argues LLM intuitions cannot yet replace human ones"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"论证大模型“直觉”尚不能替代人类直觉","en":"Argues LLM intuitions cannot yet replace human ones"}},"impactNotes":[],"verified":{"id":"10.1007/s10670-026-01136-6","title":"Philosophical Significance of Artificial Intuitions","venue":"Erkenntnis","date":"2026-07-10"},"addedIn":"v1"},{"date":"2026-10-19","title":{"zh":"Seth、Block、Tononi 在 Nature 撰文：AI 没有意识、也许永远不会有的三个理由，Chalmers 等公开回应","en":"Seth, Block & Tononi in Nature: three reasons why AI is not conscious and perhaps never will be"},"url":"https://doi.org/10.1038/d41586-026-03313-7","source":"Nature","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"AI：区分智能与主观体验，认为现有发展路径难以产生意识；学科：意识哲学与神经科学的主流学者在顶刊正面交锋，AI 意识成为学界公共议题","en":"AI: separates intelligence from experience; Field: leading philosophers and neuroscientists debate AI consciousness in a top journal"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：区分智能与主观体验，认为现有发展路径难以产生意识；学科：意识哲学与神经科学的主流学者在顶刊正面交锋，AI 意识成为学界公共议题","en":"AI: separates intelligence from experience; Field: leading philosophers and neuroscientists debate AI consciousness in a top journal"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：区分智能与主观体验，认为现有发展路径难以产生意识；学科：意识哲学与神经科学的主流学者在顶刊正面交锋，AI 意识成为学界公共议题","en":"AI: separates intelligence from experience; Field: leading philosophers and neuroscientists debate AI consciousness in a top journal"}},"impactNotes":[],"verified":{"id":"10.1038/d41586-026-03313-7","title":"Three reasons why AI is not conscious and perhaps never will be","venue":"Nature","date":"2026-10-19"},"addedIn":"v2"}],"papersRepos":[{"title":"AI-4-Phi/PhilLit","url":"https://github.com/ai-4-phi/phillit","date":null,"kind":"repo","linkCheck":"ok"},{"title":"Could a Large Language Model be Conscious?","url":"https://arxiv.org/abs/2303.07103","date":"2023-03-04","kind":"paper","linkCheck":"ok-crossref"},{"title":"Consciousness in Artificial Intelligence: Insights from the Science of Consciousness","url":"https://arxiv.org/abs/2308.08708","date":"2023-08-17","kind":"paper","linkCheck":"ok-crossref"},{"title":"Taking AI Welfare Seriously","url":"https://arxiv.org/abs/2411.00986","date":"2024-11-04","kind":"paper","linkCheck":"ok-crossref"},{"title":"Three reasons why AI is not conscious and perhaps never will be","url":"https://doi.org/10.1038/d41586-026-03313-7","date":"2026-10-19","kind":"paper","linkCheck":"ok-crossref"},{"title":"Creating a large language model of a philosopher","url":"https://doi.org/10.1111/mila.12466","date":"2023-07-12","kind":"paper","linkCheck":"ok-crossref"},{"title":"Philosophical Significance of Artificial Intuitions","url":"https://doi.org/10.1007/s10670-026-01136-6","date":"2026-07-10","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"PhilPapers","url":"https://philpapers.org"}],"news":[{"id":"1dff0995c089","date":"2026-10-19","title":"New Philosophy of AI Initiatives","url":"https://dailynous.com/2026/10/19/new-philosophy-of-ai-initiatives/","source":"Daily Nous","lang":"en","kind":"news","summaryZh":"Daily Nous：伦敦政经学院（LSE）哲学系获 530 万美元捐赠成立“全球优先事项研究组”，多伦多大学也推出 AI 哲学研究新计划。","via":"pool-v2-20261022","linkCheck":"ok"},{"id":"cf3b0a2d9b5a","date":"2026-10-18","title":"The Alignment Problem: Computer Science Meets Analytic Philosophy","url":"https://www.alignmentforum.org/posts/Emp7awtxsYn8MusNq/the-alignment-problem-computer-science-meets-analytic","source":"Alignment Forum","lang":"en","kind":"news","summaryZh":"Alignment Forum于2026-10-18发布的文章，涉及AI 安全、AI 数学，原题《The Alignment Problem: Computer Science Meets Analytic Philosophy》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"b9dcdda324f2","date":"2026-10-13","title":"Google DeepMind’s New Resident Philosopher Tackles AI Ethics","url":"https://spectrum.ieee.org/ai-ethics-deepmind-philosopher-shevlin","source":"IEEE Spectrum：AI","lang":"en","kind":"news","summaryZh":"IEEE Spectrum：AI于2026-10-13发布的文章，涉及AI 综合、AI 教育，原题《Google DeepMind’s New Resident Philosopher Tackles AI Ethics》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"c7d14d1fccfc","date":"2026-02-12","title":"Can AI Write a Useful Philosophical Literature Review?","url":"https://dailynous.com/2026/02/12/can-ai-write-a-useful-philosophical-literature-review-guest-post/","source":"Daily Nous","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"ok"}],"review":{"lastReviewed":"2026-10-21","cycleDays":180,"nextReview":"2027-04-19"},"urls":{"zh":"/zh-hans/disciplines/philosophy/","en":"/en/disciplines/philosophy/"},"dimensions":{"method":{"score":1,"label":{"zh":"研究方法","en":"Research method"},"evidence":"少量实验哲学和文献综述工作使用大模型","source":"https://dailynous.com/2026/02/12/can-ai-write-a-useful-philosophical-literature-review-guest-post/","weight":0.3},"discovery":{"score":0,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"无","source":null,"weight":0.35},"adoption":{"score":0,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"缺乏使用证据","source":null,"weight":0.25},"education":{"score":0,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到证据","source":null,"weight":0.1}},"compositeScore":0.3,"levelCapped":false,"levelChange":{"from":1,"to":0,"direction":"down","reason":{"zh":"下调为 L0（尚未改造）：AI 在哲学中主要是研究对象；作为研究工具只有零星的文献综述尝试和实验哲学中的大模型实验，Erkenntnis 2026 论文认为大模型“直觉”还不能替代人类直觉。综合得分低于 0.75。","en":"Lowered to L0 (not yet transformed): AI is mainly an object of philosophical study; tool use is limited to sporadic literature-review attempts and LLM experiments in experimental philosophy. Composite score below 0.75."}},"biggestBreakthrough":{"date":"2026-10-19","title":{"zh":"Seth、Block、Tononi 在 Nature 撰文：AI 没有意识、也许永远不会有的三个理由，Chalmers 等公开回应","en":"Seth, Block & Tononi in Nature: three reasons why AI is not conscious and perhaps never will be"},"url":"https://doi.org/10.1038/d41586-026-03313-7","aiImpact":"incremental","fieldImpact":"major","why":{"zh":"AI：区分智能与主观体验，认为现有发展路径难以产生意识；学科：意识哲学与神经科学的主流学者在顶刊正面交锋，AI 意识成为学界公共议题","en":"AI: separates intelligence from experience; Field: leading philosophers and neuroscientists debate AI consciousness in a top journal"}},"keyPapers":[{"title":"Could a Large Language Model be Conscious?","titleZh":"Chalmers：大语言模型可能有意识吗？（Boston Review 2023 / arXiv 2303.07103）","id":"arXiv:2303.07103","url":"https://arxiv.org/abs/2303.07103","venue":"arXiv / Boston Review, August 9, 2023","date":"2023-03-04","why":{"zh":"意识哲学主流学者系统评估 LLM，使 AI 意识成为正式议题","en":"Leading philosopher of mind makes LLM consciousness a mainstream question"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Consciousness in Artificial Intelligence: Insights from the Science of Consciousness","titleZh":"Butlin、Long 等 19 位学者：从意识科学理论推导 AI 意识指标（arXiv 2308.08708）","id":"arXiv:2308.08708","url":"https://arxiv.org/abs/2308.08708","venue":"arXiv","date":"2023-08-17","why":{"zh":"把哲学与神经科学理论转化为可检验的指标清单","en":"Turns theories into a checklist of testable indicators"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Taking AI Welfare Seriously","titleZh":"Long、Sebo、Butlin、Chalmers 等：认真对待 AI 福利（arXiv 2411.00986）","id":"arXiv:2411.00986","url":"https://arxiv.org/abs/2411.00986","venue":"arXiv","date":"2024-11-04","why":{"zh":"把 AI 道德地位从思想实验推进到机构政策建议","en":"Moves AI moral status from thought experiment to policy recommendations"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Three reasons why AI is not conscious and perhaps never will be","titleZh":"Seth、Block、Tononi 在 Nature 撰文：AI 没有意识、也许永远不会有的三个理由，Chalmers 等公开回应","id":"10.1038/d41586-026-03313-7","url":"https://doi.org/10.1038/d41586-026-03313-7","venue":"Nature","date":"2026-10-19","why":{"zh":"AI：区分智能与主观体验，认为现有发展路径难以产生意识；学科：意识哲学与神经科学的主流学者在顶刊正面交锋，AI 意识成为学界公共议题","en":"AI: separates intelligence from experience; Field: leading philosophers and neuroscientists debate AI consciousness in a top journal"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Creating a large language model of a philosopher","titleZh":"用哲学家丹尼特的著作微调 GPT-3，专家难以分辨机器与本人的回答（Mind & Language）","id":"10.1111/mila.12466","url":"https://doi.org/10.1111/mila.12466","venue":"Mind &amp; Language","date":"2023-07-12","why":{"zh":"引发对 AI 能否“做哲学”的讨论","en":"Sparked debate on whether AI can do philosophy"},"fieldImpact":"incremental","aiImpact":"incremental"},{"title":"Philosophical Significance of Artificial Intuitions","titleZh":"Erkenntnis：人工直觉的哲学意义","id":"10.1007/s10670-026-01136-6","url":"https://doi.org/10.1007/s10670-026-01136-6","venue":"Erkenntnis","date":"2026-07-10","why":{"zh":"论证大模型“直觉”尚不能替代人类直觉","en":"Argues LLM intuitions cannot yet replace human ones"},"fieldImpact":"incremental","aiImpact":"incremental"}],"latest_papers":[{"id":"arXiv:2610.25709","title":"Causal Humility: Governing the Authority of Fairness Claims Under Causal Uncertainty","titleZh":"因果谦逊：在因果不确定下治理公平性主张的权威","summaryZh":"结合因果公平研究和认识论，提出“因果谦逊”框架，处理认识不确定性与因果结论权威之间的关系。","summaryEn":"“Causal humility”: a governance framework for fairness claims under causal uncertainty.","authors":["Khadija Zanna","Rodrigo Ferreira","Frederick L. Oswald","Akane Sano"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.25709","site":"arXiv","venue":null,"category":"cs.CY","fetchedAt":"2026-10-22T10:10:36"},{"id":"doi:10.1007/s43681-026-01451-2","title":"AI sycophancy, self-interpretation, and conversational contentment","titleZh":"AI 谄媚、自我诠释与对话中的满足","summaryZh":"反驳“AI 谄媚只是为保护用户、危害主要在认识层面”的观点，指出在帮助用户自我诠释时谄媚也可能增进知识。","summaryEn":"Argues AI sycophancy can aid users’ self-interpretation rather than only harming knowledge.","authors":["Noam Tiran"],"date":"2026-10-16","url":"https://doi.org/10.1007/s43681-026-01451-2","site":"OpenAlex","venue":"AI and Ethics","category":null,"fetchedAt":"2026-10-22T10:10:36"},{"id":"doi:10.47989/ir31365045","title":"The information professional as an ethical intermediary: information quality criteria and the ethics of generative AI","titleZh":"作为伦理中介的信息专业人员：信息质量标准与生成式 AI 伦理","summaryZh":"范围综述 644 条记录、保留 120 篇论文，以 10 项信息质量标准分析生成式 AI 的伦理问题。","summaryEn":"Scoping review (120 of 644 records) of GenAI ethics against ten information-quality criteria.","authors":["T. D. Wilson","Elena Macevičiūtė"],"date":"2026-10-15","url":"https://doi.org/10.47989/ir31365045","site":"OpenAlex","venue":"Information Research","category":null,"fetchedAt":"2026-10-22T10:10:36"},{"id":"arXiv:2610.19427","title":"How Do LLMs Understand Legal Philosophy? Diagnosing Legal-Philosophical Reasoning in LLMs","titleZh":"大模型如何理解法哲学？诊断法哲学推理能力","summaryZh":"提出 LePhilBench，评估大模型理解法哲学并把它应用于法律推理的能力。","summaryEn":"LePhilBench evaluates whether LLMs understand legal philosophy and apply it in legal reasoning.","authors":["Shuyuan Zheng","Yun-seo Lee","Xing Gao","Jialu Wu"],"date":"2026-10-14","url":"https://arxiv.org/abs/2610.19427","site":"arXiv","venue":null,"category":"cs.CL","fetchedAt":"2026-10-22T10:10:36"},{"id":"arXiv:2610.09053","title":"Justice After Identity: Large Language Models and the View from Everywhere","titleZh":"身份之后的正义：大模型与“来自各处的视角”","summaryZh":"讨论大模型没有单一阶级、种族或经历，能否为罗尔斯“原初状态”思想实验提供新的可能。","summaryEn":"Explores whether identity-less LLMs offer a new route to Rawls’s original position.","authors":["W. Russell Neuman"],"date":"2026-10-06","url":"https://arxiv.org/abs/2610.09053","site":"arXiv","venue":null,"category":"cs.AI","fetchedAt":"2026-10-22T10:10:36"},{"id":"arXiv:2610.07847","title":"OMIT the Action: Measuring Framing-Invariant Omission Bias under Philosophical Disagreement","titleZh":"OMIT：在哲学分歧下测量框架不变的“不作为偏差”","summaryZh":"借助功利主义、义务论、德性伦理、关怀伦理和契约论五视角的人设小组构建 218 组对照情境，发现 8 个模型普遍存在不作为偏差，且同系列模型越大偏差越小。","summaryEn":"A 218-scenario benchmark built from five ethical perspectives finds pervasive omission bias in LLMs.","authors":["Sihyeon Lee","Jihun Song","Chanwoo Kim","Jiwoo Kum"],"date":"2026-10-06","url":"https://arxiv.org/abs/2610.07847","site":"arXiv","venue":null,"category":"cs.CL","fetchedAt":"2026-10-22T10:10:36"},{"id":"doi:10.1080/0020174x.2026.2745114","title":"Talking with language models","titleZh":"与语言模型交谈","summaryZh":"论证与大模型的交互并不是真正的对话，提出“人工制品立场”：大模型产出候选文本，由用户解释、评估并采纳。","summaryEn":"Argues LLM exchanges are not conversations and proposes an “artifactual stance”.","authors":["James Ravi Kirkpatrick","Alexandru Radulescu","Rachel Sterken"],"date":"2026-10-19","url":"https://doi.org/10.1080/0020174x.2026.2745114","site":"OpenAlex","venue":"Inquiry","category":null,"fetchedAt":"2026-10-22T10:10:36"}]},{"id":"religious-studies","slug":"religious-studies","category":"humanities","categoryName":{"zh":"人文","en":"Humanities"},"name":{"zh":"宗教学","en":"Religious studies"},"classification":{"oecdFos":"6.3","moe":"010107 宗教学"},"level":0,"levelKey":"not-started","levelLabel":{"zh":"尚未改造","en":"Not yet transformed"},"trend":"flat","reason":{"zh":"下调为 L0（尚未改造）：宗教学中 AI 主要是研究对象和伦理议题（梵蒂冈《罗马人工智能伦理倡议》2020、教宗通谕 Magnifica humanitas 2026）；作为研究工具只有零星的经文 OCR、对齐研究。综合得分低于 0.75。","en":"Lowered to L0: in religious studies AI is mainly an object of study and an ethical topic (Rome Call 2020, encyclical Magnifica humanitas 2026); tool use is sporadic. Composite below 0.75."},"reasonSources":[{"url":"https://www.romecall.org/the-first-signatories/","linkCheck":"ok"},{"url":"https://www.vaticannews.va/en/pope/news/2026-05/pope-leo-xiv-encyclical-magnifica-humanitas-ai.html","linkCheck":"ok"}],"directions":{"zh":["经文数字化与对齐","宗教文本翻译","AI 与宗教实践（研究对象）"],"en":["Scripture digitisation & alignment","Religious text translation","AI & religious practice (as subject)"]},"stats":{"milestones":7,"milestonesSince2025":5,"milestones2026":4,"milestonesLast6Months":3,"news":6,"latestProgress":"2026-10-21","latestMilestone":"2026-10-20","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":1,"leaps":{"ai":0,"field":0},"latestPapers":8,"keyPapers":0},"milestones":[{"date":"2020-02-28","title":{"zh":"梵蒂冈宗座生命科学院与微软、IBM、粮农组织签署《罗马人工智能伦理倡议》","en":"Rome Call for AI Ethics signed by the Pontifical Academy for Life, Microsoft, IBM and FAO"},"url":"https://www.romecall.org/the-first-signatories/","source":"Rome Call","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"policy","typeLabel":{"zh":"政策","en":"Policy"},"why":{"zh":"宗教机构首次与科技企业共同发布 AI 伦理原则","en":"First joint AI-ethics charter by a religious body and tech firms"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"宗教机构首次与科技企业共同发布 AI 伦理原则","en":"First joint AI-ethics charter by a religious body and tech firms"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"宗教机构首次与科技企业共同发布 AI 伦理原则","en":"First joint AI-ethics charter by a religious body and tech firms"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2024-11-21","title":{"zh":"瑞士卢塞恩圣彼得教堂在告解室设置“AI 耶稣”艺术装置，约 900 次对话被记录分析","en":"Lucerne’s St Peter’s Chapel installs “Deus in Machina” AI Jesus in a confessional"},"url":"https://www.theguardian.com/technology/2024/nov/21/deus-in-machina-swiss-church-installs-ai-powered-jesus","source":"The Guardian","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"deployment","typeLabel":{"zh":"落地部署","en":"Deployment"},"why":{"zh":"宗教场所首次公开实验 AI 灵性对话，引发神学争论","en":"First public church experiment with AI spiritual conversation"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"宗教场所首次公开实验 AI 灵性对话，引发神学争论","en":"First public church experiment with AI spiritual conversation"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"宗教场所首次公开实验 AI 灵性对话，引发神学争论","en":"First public church experiment with AI spiritual conversation"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2025-01-28","title":{"zh":"梵蒂冈信理部与文化教育部发布《古老而新颖》(Antiqua et nova) 说明，系统阐述 AI 与人类智能的关系","en":"Vatican note Antiqua et nova on artificial and human intelligence"},"url":"https://www.vatican.va/roman_curia/congregations/cfaith/documents/rc_ddf_doc_20250128_antiqua-et-nova_en.html","source":"Holy See","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"policy","typeLabel":{"zh":"政策","en":"Policy"},"why":{"zh":"天主教会首份系统性 AI 神学-伦理文件，为 2026 年通谕奠基","en":"First systematic Catholic doctrinal text on AI"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"天主教会首份系统性 AI 神学-伦理文件，为 2026 年通谕奠基","en":"First systematic Catholic doctrinal text on AI"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"天主教会首份系统性 AI 神学-伦理文件，为 2026 年通谕奠基","en":"First systematic Catholic doctrinal text on AI"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-01","title":{"zh":"ALMA：古代文本自动对齐（Computational Humanities Research）","en":"ALMA: automated alignment of ancient texts (Computational Humanities Research)"},"url":"https://www.cambridge.org/core/journals/computational-humanities-research/article/alma-automated-alignment-of-ancient-texts-using-linguistic-and-semantic-analysis/4267FDB026BD3433D4BAEC2ED254778E","source":"Cambridge Core","flags":["tbv"],"flagLabels":[{"zh":"日期或细节待核实","en":"Date/details to be verified"}],"linkCheck":"ok","datePrecision":"month","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"古代文本自动对齐工具","en":"Ancient text alignment tool"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"古代文本自动对齐工具","en":"Ancient text alignment tool"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"古代文本自动对齐工具","en":"Ancient text alignment tool"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-05-25","title":{"zh":"教宗良十四世发布通谕 Magnifica humanitas：AI 必须服务人类而非集中权力","en":"Pope Leo XIV’s encyclical Magnifica humanitas: AI must serve humanity"},"url":"https://www.vaticannews.va/en/pope/news/2026-05/pope-leo-xiv-encyclical-magnifica-humanitas-ai.html","source":"Vatican News","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"policy","typeLabel":{"zh":"政策","en":"Policy"},"why":{"zh":"天主教会最高级别的 AI 教导文件（是宗教对 AI 的回应，不代表 AI 改造宗教学研究）","en":"Highest-level Catholic teaching on AI"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"天主教会最高级别的 AI 教导文件（是宗教对 AI 的回应，不代表 AI 改造宗教学研究）","en":"Highest-level Catholic teaching on AI"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"天主教会最高级别的 AI 教导文件（是宗教对 AI 的回应，不代表 AI 改造宗教学研究）","en":"Highest-level Catholic teaching on AI"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-10-16","title":{"zh":"美国天主教主教团成立 AI 工作组","en":"US bishops announce AI task force members"},"url":"https://www.usccb.org/news/2026/members-and-consultants-us-bishops-task-force-artificial-intelligence-announced","source":"USCCB","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"policy","typeLabel":{"zh":"政策","en":"Policy"},"why":{"zh":"宗教机构的 AI 治理行动","en":"Religious AI governance"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"宗教机构的 AI 治理行动","en":"Religious AI governance"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"宗教机构的 AI 治理行动","en":"Religious AI governance"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-10-20","title":{"zh":"OpenAI 分析：ChatGPT 每周约 7000 万次宗教相关请求，45% 关于伊斯兰教、34% 关于基督教，随斋月、复活节等节期波动","en":"OpenAI analysis: ChatGPT receives ~70M religion-related requests a week, 45% about Islam and 34% about Christianity"},"url":"https://www.christianpost.com/news/chatgpt-gets-70m-requests-about-religion-weekly-most-about-islam.html","source":"The Christian Post","flags":["vendor"],"flagLabels":[{"zh":"厂商自述","en":"Vendor claim"}],"linkCheck":"ok","datePrecision":"day","type":"industry","typeLabel":{"zh":"产业/采用","en":"Industry"},"why":{"zh":"AI：基于 2025-09 至 2026-09 的使用数据；学科：首次给出大模型作为宗教知识与灵修咨询入口的规模数据，引出经典诠释权威的问题（数据为 OpenAI 自述，经《基督邮报》报道）","en":"AI: based on Sept 2025–Sept 2026 usage. Field: first scale data on LLMs as a gateway to religious knowledge, raising questions of interpretive authority (OpenAI figures via The Christian Post)"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：基于 2025-09 至 2026-09 的使用数据；学科：首次给出大模型作为宗教知识与灵修咨询入口的规模数据，引出经典诠释权威的问题（数据为 OpenAI 自述，经《基督邮报》报道）","en":"AI: based on Sept 2025–Sept 2026 usage. Field: first scale data on LLMs as a gateway to religious knowledge, raising questions of interpretive authority (OpenAI figures via The Christian Post)"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：基于 2025-09 至 2026-09 的使用数据；学科：首次给出大模型作为宗教知识与灵修咨询入口的规模数据，引出经典诠释权威的问题（数据为 OpenAI 自述，经《基督邮报》报道）","en":"AI: based on Sept 2025–Sept 2026 usage. Field: first scale data on LLMs as a gateway to religious knowledge, raising questions of interpretive authority (OpenAI figures via The Christian Post)"}},"impactNotes":[],"verified":null,"addedIn":"v2-20261022pm"}],"papersRepos":[],"orgsTools":[],"news":[{"id":"e18401ea21de","date":"2026-10-21","title":"Cardinal Parolin: The law must govern AI and eliminate inequalities","url":"https://www.vaticannews.va/en/vatican-city/news/2026-10/cardinal-parolin-gregorian-uni-artificial-intelligence-law.html","source":"Vatican News","lang":"en","kind":"news","summaryZh":"梵蒂冈国务卿帕罗林在宗座额我略大学教会法学院 150 周年会议上表示，法律必须规范 AI 并消除不平等。","via":"manual-websearch-20261022","linkCheck":"ok"},{"id":"dd181ebda55c","date":"2026-10-20","title":"AI, migration and Church renewal on bishops’ fall meeting agenda","url":"https://www.osvnews.com/ai-migration-and-church-renewal-on-bishops-fall-meeting-agenda/","source":"OSV News","lang":"en","kind":"news","summaryZh":"美国天主教主教团（USCCB）宣布，11 月 16–19 日巴尔的摩全体会议的议题将包括人工智能、移民与教会更新。","via":"manual-websearch-20261022","linkCheck":"ok"},{"id":"7b574550530b","date":"2026-10-19","title":"COMECE Assembly: EU bishops call for AI to remain at the service of the human person","url":"https://www.comece.eu/declaration-comece-assembly-eu-bishops-call-for-ai-to-remain-at-the-service-of-the-human-person/","source":"COMECE","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"unchecked"},{"id":"a504ab8e1d81","date":"2026-10-16","title":"Members and Consultants for U.S. Bishops’ Task Force on Artificial Intelligence Announced","url":"https://www.usccb.org/news/2026/members-and-consultants-us-bishops-task-force-artificial-intelligence-announced","source":"USCCB","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"ok"},{"id":"c929b9c10bd1","date":"2026-10-08","title":"Dr. Ruffini urges Church communicators to place human dignity at heart of AI","url":"https://www.vaticannews.va/en/vatican-city/news/2026-10/philippines-cebu-church-communicators-paolo-ruffini.html","source":"Vatican News","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"unchecked"},{"id":"3abdea4bb276","date":"2026-05-25","title":"Pope Leo’s ‘Magnifica humanitas’: AI must serve humanity not concentrate power","url":"https://www.vaticannews.va/en/pope/news/2026-05/pope-leo-xiv-encyclical-magnifica-humanitas-ai.html","source":"Vatican News","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"ok"}],"review":{"lastReviewed":"2026-10-22","cycleDays":180,"nextReview":"2027-04-19"},"urls":{"zh":"/zh-hans/disciplines/religious-studies/","en":"/en/disciplines/religious-studies/"},"dimensions":{"method":{"score":1,"label":{"zh":"研究方法","en":"Research method"},"evidence":"少量经文 OCR、翻译与文本对齐研究","source":"https://www.cambridge.org/core/journals/computational-humanities-research","weight":0.3},"discovery":{"score":0,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"无","source":null,"weight":0.35},"adoption":{"score":0,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"缺乏使用证据","source":null,"weight":0.25},"education":{"score":0,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到证据","source":null,"weight":0.1}},"compositeScore":0.3,"levelCapped":false,"levelChange":{"from":1,"to":0,"direction":"down","reason":{"zh":"下调为 L0（尚未改造）：宗教学中 AI 主要是研究对象和伦理议题（梵蒂冈《罗马人工智能伦理倡议》2020、教宗通谕 Magnifica humanitas 2026）；作为研究工具只有零星的经文 OCR、对齐研究。综合得分低于 0.75。","en":"Lowered to L0: in religious studies AI is mainly an object of study and an ethical topic (Rome Call 2020, encyclical Magnifica humanitas 2026); tool use is sporadic. Composite below 0.75."}},"biggestBreakthrough":{"date":"2026-10-20","title":{"zh":"OpenAI 分析：ChatGPT 每周约 7000 万次宗教相关请求，45% 关于伊斯兰教、34% 关于基督教，随斋月、复活节等节期波动","en":"OpenAI analysis: ChatGPT receives ~70M religion-related requests a week, 45% about Islam and 34% about Christianity"},"url":"https://www.christianpost.com/news/chatgpt-gets-70m-requests-about-religion-weekly-most-about-islam.html","aiImpact":"incremental","fieldImpact":"major","why":{"zh":"AI：基于 2025-09 至 2026-09 的使用数据；学科：首次给出大模型作为宗教知识与灵修咨询入口的规模数据，引出经典诠释权威的问题（数据为 OpenAI 自述，经《基督邮报》报道）","en":"AI: based on Sept 2025–Sept 2026 usage. Field: first scale data on LLMs as a gateway to religious knowledge, raising questions of interpretive authority (OpenAI figures via The Christian Post)"}},"keyPapers":[],"latest_papers":[{"id":"doi:10.1163/21659214-bja10168","title":"Technology Meets Theology: Understanding AI-Driven Wellness in Muslim Prayer Apps","titleZh":"科技遇上神学：理解穆斯林祈祷应用中的 AI 健康功能","summaryZh":"通过深度访谈发现，AI 驱动的祈祷应用帮助穆斯林用户建立虚拟社群、跨越地理距离进行交流。","summaryEn":"Interviews show AI-driven Muslim prayer apps foster virtual community and religious communication across distances.","authors":["Ahmed Alrawi","Mustafa Öz"],"date":"2026-10-21","url":"https://doi.org/10.1163/21659214-bja10168","site":"OpenAlex","venue":"Journal of Religion Media and Digital Culture","category":null,"fetchedAt":"2026-10-22T10:10:40"},{"id":"arXiv:2610.28452","title":"MythoSemantic: An Interactive System for Discovering Semantic Parallels Across Mythological and Religious Traditions","titleZh":"MythoSemantic：发现不同神话与宗教传统语义平行的交互系统","summaryZh":"用神经语言模型把神话和宗教文本嵌入共享语义空间，嵌入片段的粗略几何结构与已知的传统间历史关系一致，并量化词汇重叠与专名的贡献。","summaryEn":"Embeds mythological and religious texts in a shared space; its geometry matches known historical relations between traditions.","authors":["Nikita Ganziuk","Innokentiy S. Humonen","Ilya Makarov","Pavel Lebedev"],"date":"2026-10-20","url":"https://arxiv.org/abs/2610.28452","site":"arXiv","venue":null,"category":"cs.CL","fetchedAt":"2026-10-22T10:10:40"},{"id":"doi:10.20871/tjsq.v9i1.726","title":"The Shifting Visibility of Tafsīr Authority in the AI Era","titleZh":"AI 时代经注（Tafsīr）权威可见性的转变","summaryZh":"比较算法解经与以传承链（sanad）为基础的古典经注方法，讨论 AI 对古兰经诠释权威与有效性的影响。","summaryEn":"Examines how AI reshapes the visibility and authority of Qur’anic tafsīr compared with sanad-based classical methods.","authors":["Rochmad Rochmad","Miftahul Huda","Ahmad Faris Zein","Mu’adz Ali Murtadho"],"date":"2026-10-20","url":"https://doi.org/10.20871/tjsq.v9i1.726","site":"OpenAlex","venue":"Tanzil Jurnal Studi Al-Quran","category":null,"fetchedAt":"2026-10-22T10:10:40"},{"id":"https://openalex.org/W7222862103","title":"Komunikasi Dakwah Berbasis Artificial Intelligence: Negosiasi Otoritas Keagamaan antara Dai dan Chatbot AI di Kalangan Muslim Indonesia","titleZh":"基于 AI 的宣教传播：印尼穆斯林中宣教士与 AI 聊天机器人之间的宗教权威协商","summaryZh":"质性研究发现 AI 聊天机器人日益成为印尼穆斯林获取宗教知识的第一参考，引发与宣教士之间的权威协商。","summaryEn":"Qualitative study of how AI chatbots negotiate religious authority with preachers among Indonesian Muslims.","authors":["Subairi Subairi","Asyaari Asyaari"],"date":"2026-10-16","url":"https://journal.stidkialhamidy.ac.id/index.php/dakwatulhaq/article/view/29","site":"OpenAlex","venue":"Dakwatul Haq: Journal of Da'wah and Islamic Communication","category":null,"fetchedAt":"2026-10-22T10:10:40"},{"id":"doi:10.1080/15295036.2026.2739381","title":"“Well done, My good and faithful servant”: AI-generated religious imagery, white evangelical U.S. American Christian nationalism, and political martyrdom","titleZh":"“良善又忠心的仆人”：AI 生成宗教图像、美国白人福音派基督教民族主义与政治殉道","summaryZh":"分析 2025 年查理·柯克遇刺后社交媒体上把他塑造成殉道者的 AI 生成宗教图像，以泰德·克鲁兹的帖子为例。","summaryEn":"Analyses AI-generated religious imagery framing Charlie Kirk as a martyr in Christian-nationalist discourse.","authors":["Miriam Wolff"],"date":"2026-10-14","url":"https://doi.org/10.1080/15295036.2026.2739381","site":"OpenAlex","venue":"Critical Studies in Media Communication","category":null,"fetchedAt":"2026-10-22T10:10:40"},{"id":"arXiv:2610.14229","title":"Who Guides Whom? Negotiating Care, Religious Authority, and Technology in Bangladeshi Muslim Families","titleZh":"谁引导谁？孟加拉穆斯林家庭中的照护、宗教权威与技术","summaryZh":"访谈 38 位孟加拉穆斯林家庭成员，发现老人仍是宗教指导来源，年轻人则在数字风险和 AI 方面提供专业意见。","summaryEn":"Interviews with 38 Bangladeshi Muslim family members on religious authority, digital risk and AI.","authors":["Susmita Biswas","Saiful Islam Tuhin","Faria Soroni","Sharifa Sultana"],"date":"2026-10-09","url":"https://arxiv.org/abs/2610.14229","site":"arXiv","venue":null,"category":"cs.HC","fetchedAt":"2026-10-22T10:10:40"},{"id":"arXiv:2610.12064","title":"ILM: An AI-Powered Storytelling Educational Tool","titleZh":"ILM：AI 驱动的讲故事教育工具","summaryZh":"面向《先知故事》的互动学习平台，结合阿拉伯语 NLP、知识图谱和检索式问题生成。","summaryEn":"An AI platform for learning Stories of the Prophets using Arabic NLP and knowledge graphs.","authors":["Suhaila Mohammed","Abdelaziz Serour","Allison Lahnala"],"date":"2026-10-08","url":"https://arxiv.org/abs/2610.12064","site":"arXiv","venue":null,"category":"cs.CL","fetchedAt":"2026-10-22T10:10:40"},{"id":"arXiv:2609.35201","title":"From Normative Frameworks to Alignment Data: Constructing and Evaluating SFT and Preference Data","titleZh":"从规范框架到对齐数据：构建与评估 SFT 和偏好数据","summaryZh":"7 位领域专家用约一年时间，基于伊斯兰伦理、神学和法学传统构建约 2800 条微调样本和 5400 对偏好数据。","summaryEn":"Experts built ~2.8K SFT examples and 5.4K preference pairs grounded in Islamic ethical traditions.","authors":["Husrev Taha Sencar","Rezart Beka","Danish Naeem","Seda Ozalkan"],"date":"2026-09-28","url":"https://arxiv.org/abs/2609.35201","site":"arXiv","venue":null,"category":"cs.CL","fetchedAt":"2026-10-22T10:10:40"}]},{"id":"literature","slug":"literature","category":"humanities","categoryName":{"zh":"人文","en":"Humanities"},"name":{"zh":"文学","en":"Literature"},"classification":{"oecdFos":"6.2","moe":"0501 中国语言文学 / 0502 外国语言文学"},"level":0,"levelKey":"not-started","levelLabel":{"zh":"尚未改造","en":"Not yet transformed"},"trend":"flat","reason":{"zh":"下调为 L0（尚未改造）：创作端有个案（九段理江 2024 芥川奖作品约 5% 文字来自 ChatGPT）和实验研究（AI 诗歌在盲测中难以区分），但文学研究本身没有以 AI 为关键的成果，从业者采用缺乏数据。综合得分低于 0.75。","en":"Lowered to L0: isolated cases in creative writing (Akutagawa Prize 2024) and experiments (AI poetry hard to tell apart in blind tests), but no AI-key results in literary studies and no adoption data. Composite below 0.75."},"reasonSources":[{"url":"https://www.japantimes.co.jp/culture/2024/01/19/books/akutagawa-prize-book-chatgpt/","linkCheck":"ok"},{"url":"https://doi.org/10.1038/s41598-024-76900-1","linkCheck":"ok-crossref"}],"directions":{"zh":["AI 辅助创作","计算文学批评","文学翻译"],"en":["AI-assisted writing","Computational criticism","Literary translation"]},"stats":{"milestones":8,"milestonesSince2025":4,"milestones2026":4,"milestonesLast6Months":4,"news":5,"latestProgress":"2026-10-21","latestMilestone":"2026-10-21","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":0,"leaps":{"ai":0,"field":0},"latestPapers":3,"keyPapers":2},"milestones":[{"date":"2023-09-20","title":{"zh":"美国作家协会与格里沙姆、乔治·R·R·马丁等 17 位作家对 OpenAI 提起集体诉讼","en":"Authors Guild and 17 authors incl. Grisham and G.R.R. Martin sue OpenAI"},"url":"https://authorsguild.org/news/ag-and-authors-file-class-action-suit-against-openai/","source":"Authors Guild","flags":["contested"],"flagLabels":[{"zh":"有争议","en":"Contested"}],"linkCheck":"ok","datePrecision":"day","type":"legal","typeLabel":{"zh":"诉讼/判决","en":"Legal"},"why":{"zh":"作家群体就训练数据维权的代表性诉讼，案件仍在审理","en":"Representative authors’ training-data suit; ongoing"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"作家群体就训练数据维权的代表性诉讼，案件仍在审理","en":"Representative authors’ training-data suit; ongoing"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"作家群体就训练数据维权的代表性诉讼，案件仍在审理","en":"Representative authors’ training-data suit; ongoing"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2024-01-19","title":{"zh":"九段理江获芥川奖，称约 5% 文本由 ChatGPT 生成","en":"Rie Kudan wins Akutagawa Prize, says ~5% of text came from ChatGPT"},"url":"https://www.japantimes.co.jp/culture/2024/01/19/books/akutagawa-prize-book-chatgpt/","source":"The Japan Times","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"prize","typeLabel":{"zh":"奖项","en":"Prize"},"why":{"zh":"日本最高纯文学奖作品作者公开承认使用 ChatGPT","en":"Akutagawa winner discloses ChatGPT use"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"日本最高纯文学奖作品作者公开承认使用 ChatGPT","en":"Akutagawa winner discloses ChatGPT use"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"日本最高纯文学奖作品作者公开承认使用 ChatGPT","en":"Akutagawa winner discloses ChatGPT use"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2024-07-12","title":{"zh":"生成式 AI 提升个人创造力但降低集体多样性：短篇故事实验（Science Advances）","en":"Generative AI enhances individual creativity but reduces the collective diversity of novel content (Science Advances)"},"url":"https://doi.org/10.1126/sciadv.adn5290","source":"Science Advances","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"首个关于 AI 辅助写作对作品多样性影响的大样本实验","en":"First large experiment on AI-assisted writing and diversity"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"首个关于 AI 辅助写作对作品多样性影响的大样本实验","en":"First large experiment on AI-assisted writing and diversity"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首个关于 AI 辅助写作对作品多样性影响的大样本实验","en":"First large experiment on AI-assisted writing and diversity"}},"impactNotes":[],"verified":{"id":"10.1126/sciadv.adn5290","title":"Generative AI enhances individual creativity but reduces the collective diversity of novel content","venue":"Science Advances","date":"2024-07-12"},"addedIn":"v2"},{"date":"2024-11-14","title":{"zh":"AI 生成的诗歌与人类诗歌难以区分且更受读者喜爱（Scientific Reports）","en":"AI-generated poetry is indistinguishable from human-written poetry and is rated more favourably (Sci. Rep.)"},"url":"https://doi.org/10.1038/s41598-024-76900-1","source":"Scientific Reports","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"非专业读者在盲测中无法区分，并更偏好 AI 诗歌","en":"Non-expert readers cannot tell and prefer AI poems"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"非专业读者在盲测中无法区分，并更偏好 AI 诗歌","en":"Non-expert readers cannot tell and prefer AI poems"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"非专业读者在盲测中无法区分，并更偏好 AI 诗歌","en":"Non-expert readers cannot tell and prefer AI poems"}},"impactNotes":[],"verified":{"id":"10.1038/s41598-024-76900-1","title":"AI-generated poetry is indistinguishable from human-written poetry and is rated more favorably","venue":"Scientific Reports","date":"2024-11-14"},"addedIn":"v2"},{"date":"2026-07-20","title":{"zh":"美国法院最终批准 Anthropic 与作者的 15 亿美元版权和解（Bartz 案，约 48 万部作品，每部约 3000 美元）","en":"US court gives final approval to Anthropic’s $1.5B copyright settlement with authors (Bartz)"},"url":"https://www.reuters.com/world/us-judge-approves-anthropics-15-billion-settlement-copyright-lawsuit-2026-07-20/","source":"Reuters","flags":[],"flagLabels":[],"linkCheck":"site-blocks-bots","datePrecision":"day","type":"legal","typeLabel":{"zh":"诉讼/判决","en":"Legal"},"why":{"zh":"迄今最大的 AI 训练数据版权和解，确立图书作者获偿先例","en":"Largest AI training-data copyright settlement; precedent for authors"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"迄今最大的 AI 训练数据版权和解，确立图书作者获偿先例","en":"Largest AI training-data copyright settlement; precedent for authors"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"迄今最大的 AI 训练数据版权和解，确立图书作者获偿先例","en":"Largest AI training-data copyright settlement; precedent for authors"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-09-25","title":{"zh":"龚古尔学院以“维护奖项完整性”为由，将被指“几乎全由 AI 写成”的 Orélien 小说《C’était ça ou mourir》移出长名单，作者否认","en":"Académie Goncourt removes Thélyson Orélien’s novel from its longlist over claims it was largely AI-written; author denies"},"url":"https://www.theguardian.com/books/2026/sep/25/thelyson-orelien-goncourt-prize-france","source":"The Guardian","flags":["contested"],"flagLabels":[{"zh":"有争议","en":"Contested"}],"linkCheck":"ok","datePrecision":"day","type":"industry","typeLabel":{"zh":"产业/采用","en":"Industry"},"why":{"zh":"AI：起因是匿名账号与 AI 检测结果，各检测工具结论不一；学科：法国最高文学奖首次因疑似 AI 写作撤下作品，随后出版社 Boréal 于 10-21 与作者解约，Renaudot 奖则保留该书（争议未决）","en":"AI: triggered by an anonymous account and conflicting AI-detector results. Field: France’s top prize drops a book over suspected AI authorship for the first time; publisher Boréal later cut ties, Renaudot kept it (unresolved)"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：起因是匿名账号与 AI 检测结果，各检测工具结论不一；学科：法国最高文学奖首次因疑似 AI 写作撤下作品，随后出版社 Boréal 于 10-21 与作者解约，Renaudot 奖则保留该书（争议未决）","en":"AI: triggered by an anonymous account and conflicting AI-detector results. Field: France’s top prize drops a book over suspected AI authorship for the first time; publisher Boréal later cut ties, Renaudot kept it (unresolved)"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：起因是匿名账号与 AI 检测结果，各检测工具结论不一；学科：法国最高文学奖首次因疑似 AI 写作撤下作品，随后出版社 Boréal 于 10-21 与作者解约，Renaudot 奖则保留该书（争议未决）","en":"AI: triggered by an anonymous account and conflicting AI-detector results. Field: France’s top prize drops a book over suspected AI authorship for the first time; publisher Boréal later cut ties, Renaudot kept it (unresolved)"}},"impactNotes":[],"verified":null,"addedIn":"v2-20261022pm"},{"date":"2026-10-14","title":{"zh":"凯恩非洲写作奖获奖短篇被 AI 检测工具判定为几乎全由 AI 生成，作者否认，奖项方启动人工审查","en":"Caine Prize-winning story flagged as AI-generated; author denies, prize opens a human-led review"},"url":"https://www.theguardian.com/technology/2026/oct/14/zimbabwean-author-simbarashe-kundizeza-denies-ai-story","source":"The Guardian","flags":["contested"],"flagLabels":[{"zh":"有争议","en":"Contested"}],"linkCheck":"ok","datePrecision":"day","type":"industry","typeLabel":{"zh":"产业/采用","en":"Industry"},"why":{"zh":"AI：检测工具 Pangram 的判断成为争议焦点；学科：主要文学奖首次因 AI 疑云进入正式审查，作者署名规范受到考验","en":"AI: a detector’s verdict is the crux; Field: a major literary prize opens a formal review over AI authorship"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：检测工具 Pangram 的判断成为争议焦点；学科：主要文学奖首次因 AI 疑云进入正式审查，作者署名规范受到考验","en":"AI: a detector’s verdict is the crux; Field: a major literary prize opens a formal review over AI authorship"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：检测工具 Pangram 的判断成为争议焦点；学科：主要文学奖首次因 AI 疑云进入正式审查，作者署名规范受到考验","en":"AI: a detector’s verdict is the crux; Field: a major literary prize opens a formal review over AI authorship"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-10-21","title":{"zh":"法国设立 Prix Pascal 人工智能文学奖：征集与大模型合写的法语小说稿，须申报所用模型与提示词，获奖作品出版","en":"Prix Pascal de Littérature IA launches for French-language fiction co-written with AI"},"url":"https://www.theguardian.com/books/2026/oct/21/new-prize-launches-books-co-written-with-ai-france","source":"The Guardian","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"prize","typeLabel":{"zh":"奖项","en":"Prize"},"why":{"zh":"AI：明确允许 ChatGPT、Claude、Gemini、LLaMA 等模型参与创作并要求公开提示词；学科：在龚古尔、凯恩、英联邦短篇奖接连出现 AI 争议之际，出现专门面向人机合写的文学奖（由科技创业者发起，评委待公布）","en":"AI: explicitly allows LLM co-writing and requires disclosure of model and prompts. Field: a prize dedicated to human–AI co-writing amid AI controversies at major prizes (founded by tech entrepreneurs)"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：明确允许 ChatGPT、Claude、Gemini、LLaMA 等模型参与创作并要求公开提示词；学科：在龚古尔、凯恩、英联邦短篇奖接连出现 AI 争议之际，出现专门面向人机合写的文学奖（由科技创业者发起，评委待公布）","en":"AI: explicitly allows LLM co-writing and requires disclosure of model and prompts. Field: a prize dedicated to human–AI co-writing amid AI controversies at major prizes (founded by tech entrepreneurs)"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：明确允许 ChatGPT、Claude、Gemini、LLaMA 等模型参与创作并要求公开提示词；学科：在龚古尔、凯恩、英联邦短篇奖接连出现 AI 争议之际，出现专门面向人机合写的文学奖（由科技创业者发起，评委待公布）","en":"AI: explicitly allows LLM co-writing and requires disclosure of model and prompts. Field: a prize dedicated to human–AI co-writing amid AI controversies at major prizes (founded by tech entrepreneurs)"}},"impactNotes":[],"verified":null,"addedIn":"v2-20261022pm"}],"papersRepos":[{"title":"Generative AI enhances individual creativity but reduces the collective diversity of novel content","url":"https://doi.org/10.1126/sciadv.adn5290","date":"2024-07-12","kind":"paper","linkCheck":"ok-crossref"},{"title":"AI-generated poetry is indistinguishable from human-written poetry and is rated more favorably","url":"https://doi.org/10.1038/s41598-024-76900-1","date":"2024-11-14","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[],"news":[{"id":"d2d11dc4f544","date":"2026-10-21","title":"Canadian publisher drops writer Thelyson Orelien over AI use in novel","url":"https://www.aljazeera.com/news/2026/10/21/canadian-publisher-cuts-ties-with-best-selling-writer-orelien-over-ai-use","source":"Al Jazeera","lang":"en","kind":"news","summaryZh":"加拿大出版社 Boréal 称调查证实海地裔作家 Thelyson Orelien 的成名小说使用了 AI，宣布与其解约。","via":"pool-v2-20261022","linkCheck":"ok"},{"id":"94a27ef2597c","date":"2026-10-20","title":"Language Models are Lazy Poets: Creative Mode Collapse at Scale","url":"https://arxiv.org/abs/2610.24223","source":"arXiv cs.CL","lang":"en","kind":"paper","summaryZh":"arXiv cs.CL于2026-10-20发布的论文，涉及AI 创意媒体，原题《Language Models are Lazy Poets: Creative Mode Collapse at Scale》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"bcaad4abff6c","date":"2026-10-20","title":"番茄小说决定杀死AI","url":"https://36kr.com/p/4034128921827207","source":"36氪","lang":"zh","kind":"news","summaryZh":null,"via":"pool-v2-20261021","linkCheck":"from-feed"},{"id":"48d4adca3a2b","date":"2026-10-19","title":"Who is Elias the lighthouse keeper? AI fiction tells the same old story","url":"https://techxplore.com/news/2026-10-elias-lighthouse-keeper-ai-fiction.html","source":"Phys.org 机器学习与AI","lang":"en","kind":"news","summaryZh":"Phys.org 机器学习与AI于2026-10-20发布的文章，涉及AI 科研与调研，原题《Who is Elias the lighthouse keeper? AI fiction tells the same old story》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"3a86afd21c3a","date":"2026-10-14","title":"Yet Another Literary Prizewinner Seems to Be AI","url":"https://www.theatlantic.com/books/2026/10/caine-prize-winning-short-story-seems-be-ai/688967/","source":"The Atlantic（经 AI Incident Database 收录）","lang":"en","kind":"news","summaryZh":"《大西洋月刊》报道：凯恩非洲写作奖的一篇获奖短篇疑似由 AI 生成。","via":"pool-v2-20261021","linkCheck":"site-blocks-bots"}],"review":{"lastReviewed":"2026-10-22","cycleDays":180,"nextReview":"2027-04-19"},"urls":{"zh":"/zh-hans/disciplines/literature/","en":"/en/disciplines/literature/"},"dimensions":{"method":{"score":1,"label":{"zh":"研究方法","en":"Research method"},"evidence":"数字人文文本分析辅助","source":null,"weight":0.3},"discovery":{"score":0,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"无","source":null,"weight":0.35},"adoption":{"score":1,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"部分作家使用（芥川奖个案），无普及数据","source":"https://www.japantimes.co.jp/culture/2024/01/19/books/akutagawa-prize-book-chatgpt/","weight":0.25},"education":{"score":0,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到证据","source":null,"weight":0.1}},"compositeScore":0.55,"levelCapped":false,"levelChange":{"from":1,"to":0,"direction":"down","reason":{"zh":"下调为 L0（尚未改造）：创作端有个案（九段理江 2024 芥川奖作品约 5% 文字来自 ChatGPT）和实验研究（AI 诗歌在盲测中难以区分），但文学研究本身没有以 AI 为关键的成果，从业者采用缺乏数据。综合得分低于 0.75。","en":"Lowered to L0: isolated cases in creative writing (Akutagawa Prize 2024) and experiments (AI poetry hard to tell apart in blind tests), but no AI-key results in literary studies and no adoption data. Composite below 0.75."}},"biggestBreakthrough":{"date":"2026-09-25","title":{"zh":"龚古尔学院以“维护奖项完整性”为由，将被指“几乎全由 AI 写成”的 Orélien 小说《C’était ça ou mourir》移出长名单，作者否认","en":"Académie Goncourt removes Thélyson Orélien’s novel from its longlist over claims it was largely AI-written; author denies"},"url":"https://www.theguardian.com/books/2026/sep/25/thelyson-orelien-goncourt-prize-france","aiImpact":"incremental","fieldImpact":"major","why":{"zh":"AI：起因是匿名账号与 AI 检测结果，各检测工具结论不一；学科：法国最高文学奖首次因疑似 AI 写作撤下作品，随后出版社 Boréal 于 10-21 与作者解约，Renaudot 奖则保留该书（争议未决）","en":"AI: triggered by an anonymous account and conflicting AI-detector results. Field: France’s top prize drops a book over suspected AI authorship for the first time; publisher Boréal later cut ties, Renaudot kept it (unresolved)"}},"keyPapers":[{"title":"Generative AI enhances individual creativity but reduces the collective diversity of novel content","titleZh":"生成式 AI 提升个人创造力但降低集体多样性：短篇故事实验（Science Advances）","id":"10.1126/sciadv.adn5290","url":"https://doi.org/10.1126/sciadv.adn5290","venue":"Science Advances","date":"2024-07-12","why":{"zh":"首个关于 AI 辅助写作对作品多样性影响的大样本实验","en":"First large experiment on AI-assisted writing and diversity"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"AI-generated poetry is indistinguishable from human-written poetry and is rated more favorably","titleZh":"AI 生成的诗歌与人类诗歌难以区分且更受读者喜爱（Scientific Reports）","id":"10.1038/s41598-024-76900-1","url":"https://doi.org/10.1038/s41598-024-76900-1","venue":"Scientific Reports","date":"2024-11-14","why":{"zh":"非专业读者在盲测中无法区分，并更偏好 AI 诗歌","en":"Non-expert readers cannot tell and prefer AI poems"},"fieldImpact":"incremental","aiImpact":"major"}],"latest_papers":[{"id":"arXiv:2610.27894","title":"Testing Associative Creativity in Humans and Multimodal LLMs with Difficult-to-Name Visual Stimuli","titleZh":"用难以命名的视觉刺激测试人类与多模态大模型的联想创造力","summaryZh":"人类对抽象图形的描述比多模态大模型更多样、语义距离更远、表达更新颖；难度增加时人类转向描述低层视觉特征。","summaryEn":"Humans produce more diverse and novel associative descriptions of abstract stimuli than MLLMs.","authors":["Manar Ali","Lisa Gottschalk","Simeon Junker","Lisa Hudalla"],"date":"2026-10-20","url":"https://arxiv.org/abs/2610.27894","site":"arXiv","venue":null,"category":"cs.CL","fetchedAt":"2026-10-22T10:10:44"},{"id":"arXiv:2610.24223","title":"Language Models are Lazy Poets: Creative Mode Collapse at Scale","titleZh":"语言模型是懒惰的诗人：大规模创作模式坍缩","summaryZh":"用 16 个开源模型生成 110 多万首诗，发现模型词汇和押韵能力丰富却严重未被利用，过度依赖少量词语、开头和韵脚。","summaryEn":"1.1M+ poems from 16 models show LLMs over-rely on few words, openings and rhymes despite large inventories.","authors":["Jacob A. Matthews","Sil Hamilton"],"date":"2026-10-16","url":"https://arxiv.org/abs/2610.24223","site":"arXiv","venue":null,"category":"cs.CL","fetchedAt":"2026-10-22T10:10:44"},{"id":"doi:10.24919/2308-4863/103-37","title":"ІНТЕРАКТИВНЕ ПОСТРЕДАГУВАННЯ КИТАЙСЬКО-УКРАЇНСЬКИХ ХУДОЖНІХ І ФІЛОСОФСЬКИХ ТЕКСТІВ: МОДЕЛЬ AIPE ТА ТИПОЛОГІЯ ПРОМПТ-РОЛЕЙ","titleZh":"汉乌文学与哲学文本的交互式译后编辑：AIPE 模型与提示角色类型","summaryZh":"提出 AI 辅助交互式译后编辑模型，归纳大模型在文言及文白文本翻译中的 14 类陷阱，并建立提示角色类型。","summaryEn":"An AI-assisted interactive post-editing model for Chinese–Ukrainian literary/philosophical translation, cataloguing 14 LLM pitfalls.","authors":["Maksym Kryvonis"],"date":"2026-10-16","url":"https://doi.org/10.24919/2308-4863/103-37","site":"OpenAlex","venue":"Humanities science current issues","category":null,"fetchedAt":"2026-10-22T10:10:44"}]},{"id":"music","slug":"music","category":"arts","categoryName":{"zh":"艺术","en":"Arts"},"name":{"zh":"音乐","en":"Music"},"classification":{"oecdFos":"6.4","moe":"1352 音乐"},"level":2,"levelKey":"breakthrough","levelLabel":{"zh":"局部突破","en":"Local breakthroughs"},"trend":"up","reason":{"zh":"AI 音乐从诉讼走向授权：华纳与 Suno、Udio 和解，环球与 Udio 和解，Suno 推出基于授权曲库的新模型（2026）；IFPI 发布 AI 录音进入官方榜单的全球原则。","en":"AI music moved from lawsuits to licensing: Warner settled with Suno and Udio, Universal with Udio, and Suno launched label-licensed models (2026); IFPI issued global principles for AI recordings on official charts."},"reasonSources":[{"url":"https://www.latimes.com/entertainment-arts/business/story/2026-09-21/ai-music-licensing-deals-labels-artists-who-benefits","linkCheck":"ok"},{"url":"https://www.ifpi.org/ifpi-rolls-out-global-principles-for-the-eligibility-of-recordings-developed-using-ai-in-official-music-charts-worldwide/","linkCheck":"ok"}],"directions":{"zh":["文本生成音乐","授权训练与版权","AI 辅助编曲"],"en":["Text-to-music","Licensed training & copyright","AI-assisted production"]},"stats":{"milestones":9,"milestonesSince2025":6,"milestones2026":5,"milestonesLast6Months":5,"news":11,"latestProgress":"2026-10-21","latestMilestone":"2026-10-20","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":0,"leaps":{"ai":0,"field":0},"latestPapers":8,"keyPapers":3},"milestones":[{"date":"2018-09-12","title":{"zh":"Music Transformer：生成长时程结构的音乐（arXiv 1809.04281）","en":"Music Transformer: generating music with long-term structure"},"url":"https://arxiv.org/abs/1809.04281","source":"arXiv","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok-crossref","datePrecision":"day","type":"system","typeLabel":{"zh":"系统发布","en":"System"},"why":{"zh":"Transformer 首次用于长篇音乐生成","en":"Transformers for long-form music"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"Transformer 首次用于长篇音乐生成","en":"Transformers for long-form music"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"Transformer 首次用于长篇音乐生成","en":"Transformers for long-form music"}},"impactNotes":[],"verified":{"id":"arXiv:1809.04281","title":"Music Transformer","venue":"arXiv","date":"2018-09-12"},"addedIn":"v2"},{"date":"2020-04-30","title":{"zh":"Jukebox：OpenAI 生成带人声的原始音频音乐（arXiv 2005.00341）","en":"Jukebox: generating raw-audio music with singing (OpenAI)"},"url":"https://arxiv.org/abs/2005.00341","source":"arXiv","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok-crossref","datePrecision":"day","type":"system","typeLabel":{"zh":"系统发布","en":"System"},"why":{"zh":"首个能生成带歌声的完整歌曲的模型","en":"First model generating songs with vocals"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首个能生成带歌声的完整歌曲的模型","en":"First model generating songs with vocals"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"首个能生成带歌声的完整歌曲的模型","en":"First model generating songs with vocals"}},"impactNotes":[],"verified":{"id":"arXiv:2005.00341","title":"Jukebox: A Generative Model for Music","venue":"arXiv","date":"2020-04-30"},"addedIn":"v2"},{"date":"2023-01-26","title":{"zh":"MusicLM：从文本生成高保真音乐（Google，arXiv 2301.11325）","en":"MusicLM: generating music from text (Google)"},"url":"https://arxiv.org/abs/2301.11325","source":"arXiv","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok-crossref","datePrecision":"day","type":"system","typeLabel":{"zh":"系统发布","en":"System"},"why":{"zh":"文生音乐进入可用水平","en":"Text-to-music reaches usable quality"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"文生音乐进入可用水平","en":"Text-to-music reaches usable quality"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"文生音乐进入可用水平","en":"Text-to-music reaches usable quality"}},"impactNotes":[],"verified":{"id":"arXiv:2301.11325","title":"MusicLM: Generating Music From Text","venue":"arXiv","date":"2023-01-26"},"addedIn":"v2"},{"date":"2025-11-25","title":{"zh":"华纳音乐与 Suno 和解版权诉讼并建立授权合作","en":"Warner Music Group settles with Suno and forms a licensing partnership"},"url":"https://www.reuters.com/legal/litigation/warner-music-group-settles-copyright-case-with-suno-licensed-ai-music-2025-11-25/","source":"Reuters","flags":[],"flagLabels":[],"linkCheck":"site-blocks-bots","datePrecision":"day","type":"industry","typeLabel":{"zh":"产业/采用","en":"Industry"},"why":{"zh":"唱片业从诉讼转向授权，AI 音乐获得合法化路径","en":"Industry moves from lawsuits to licensing"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"唱片业从诉讼转向授权，AI 音乐获得合法化路径","en":"Industry moves from lawsuits to licensing"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"唱片业从诉讼转向授权，AI 音乐获得合法化路径","en":"Industry moves from lawsuits to licensing"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-07-30","title":{"zh":"IFPI 发布 AI 录音进入官方榜单的全球原则：须授权、须有实质性人类创作","en":"IFPI global principles for AI-developed recordings in official charts"},"url":"https://www.ifpi.org/ifpi-rolls-out-global-principles-for-the-eligibility-of-recordings-developed-using-ai-in-official-music-charts-worldwide/","source":"IFPI","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"policy","typeLabel":{"zh":"政策","en":"Policy"},"why":{"zh":"全球唱片业首次统一 AI 作品上榜规则（原“待核实”已确认 2026-07-30）","en":"First global chart rules for AI recordings"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"全球唱片业首次统一 AI 作品上榜规则（原“待核实”已确认 2026-07-30）","en":"First global chart rules for AI recordings"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"全球唱片业首次统一 AI 作品上榜规则（原“待核实”已确认 2026-07-30）","en":"First global chart rules for AI recordings"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-07-31","title":{"zh":"慕尼黑地区法院判决 Suno 侵犯 GEMA 管理的德国歌曲版权：训练、模型记忆和可辨识输出均构成侵权","en":"Munich Regional Court rules Suno infringed GEMA-administered German song copyrights"},"url":"https://www.billboard.com/pro/suno-liable-gema-german-copyright-lawsuit/","source":"Billboard","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"legal","typeLabel":{"zh":"诉讼/判决","en":"Legal"},"why":{"zh":"AI：法院认定模型对受保护歌曲的记忆与可辨识输出属于复制；学科：AI 音乐公司在欧洲的首批重大败诉之一，Suno 须披露收入并赔偿（可上诉）","en":"AI: memorisation and recognisable outputs held to be reproduction. Field: one of the first major European losses for an AI music firm (appealable)"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：法院认定模型对受保护歌曲的记忆与可辨识输出属于复制；学科：AI 音乐公司在欧洲的首批重大败诉之一，Suno 须披露收入并赔偿（可上诉）","en":"AI: memorisation and recognisable outputs held to be reproduction. Field: one of the first major European losses for an AI music firm (appealable)"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：法院认定模型对受保护歌曲的记忆与可辨识输出属于复制；学科：AI 音乐公司在欧洲的首批重大败诉之一，Suno 须披露收入并赔偿（可上诉）","en":"AI: memorisation and recognisable outputs held to be reproduction. Field: one of the first major European losses for an AI music firm (appealable)"}},"impactNotes":[],"verified":null,"addedIn":"v2-20261022pm"},{"date":"2026-09-09","title":{"zh":"Suno 联合华纳、BMG 发布基于授权曲库的新模型","en":"Suno releases new models built with Warner Music and BMG licences"},"url":"https://www.reuters.com/legal/litigation/suno-releases-new-ai-music-models-partnership-with-warner-music-bmg-2026-09-09/","source":"Reuters","flags":[],"flagLabels":[],"linkCheck":"site-blocks-bots","datePrecision":"day","type":"system","typeLabel":{"zh":"系统发布","en":"System"},"why":{"zh":"首批唱片公司授权训练的商用 AI 音乐模型","en":"First label-licensed commercial music models"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"首批唱片公司授权训练的商用 AI 音乐模型","en":"First label-licensed commercial music models"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首批唱片公司授权训练的商用 AI 音乐模型","en":"First label-licensed commercial music models"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-09-21","title":{"zh":"洛杉矶时报：Suno 与唱片公司合作推出授权模型","en":"LA Times: Suno builds a new model with major labels"},"url":"https://www.latimes.com/entertainment-arts/business/story/2026-09-21/ai-music-licensing-deals-labels-artists-who-benefits","source":"Los Angeles Times","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"industry","typeLabel":{"zh":"产业/采用","en":"Industry"},"why":{"zh":"媒体报道授权合作","en":"Coverage of licensing deals"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"媒体报道授权合作","en":"Coverage of licensing deals"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"媒体报道授权合作","en":"Coverage of licensing deals"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-10-20","title":{"zh":"Spotify：今年已从推荐中剔除 5000 多万首 AI 垃圾/刷量曲目、另下架 7500 多万首，AI 人设歌手默认不进推荐，AI 创作署名每日新增数万条","en":"Spotify: 50M+ AI slop/spam tracks removed from recommendations this year, 75M+ removed outright; AI Personas excluded from recommendations by default"},"url":"https://newsroom.spotify.com/2026-10-20/prioritizing-human-artistry/","source":"Spotify Newsroom","flags":["vendor"],"flagLabels":[{"zh":"厂商自述","en":"Vendor claim"}],"linkCheck":"ok","datePrecision":"day","type":"industry","typeLabel":{"zh":"产业/采用","en":"Industry"},"why":{"zh":"AI：平台综合听众行为、上传模式和多家第三方 AI 检测工具识别 AI 垃圾内容，并承认检测会越来越难；学科：最大流媒体平台把“人类创作优先”写进推荐与标注规则，影响音乐分发与收益分配（数字为 Spotify 自述）","en":"AI: combines engagement, upload patterns and third-party AI detectors (acknowledged to be getting harder). Field: the largest streamer codifies human-first recommendation and labelling (Spotify’s figures)"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：平台综合听众行为、上传模式和多家第三方 AI 检测工具识别 AI 垃圾内容，并承认检测会越来越难；学科：最大流媒体平台把“人类创作优先”写进推荐与标注规则，影响音乐分发与收益分配（数字为 Spotify 自述）","en":"AI: combines engagement, upload patterns and third-party AI detectors (acknowledged to be getting harder). Field: the largest streamer codifies human-first recommendation and labelling (Spotify’s figures)"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：平台综合听众行为、上传模式和多家第三方 AI 检测工具识别 AI 垃圾内容，并承认检测会越来越难；学科：最大流媒体平台把“人类创作优先”写进推荐与标注规则，影响音乐分发与收益分配（数字为 Spotify 自述）","en":"AI: combines engagement, upload patterns and third-party AI detectors (acknowledged to be getting harder). Field: the largest streamer codifies human-first recommendation and labelling (Spotify’s figures)"}},"impactNotes":[],"verified":null,"addedIn":"v2-20261022pm"}],"papersRepos":[{"title":"Music Transformer","url":"https://arxiv.org/abs/1809.04281","date":"2018-09-12","kind":"paper","linkCheck":"ok-crossref"},{"title":"Jukebox: A Generative Model for Music","url":"https://arxiv.org/abs/2005.00341","date":"2020-04-30","kind":"paper","linkCheck":"ok-crossref"},{"title":"MusicLM: Generating Music From Text","url":"https://arxiv.org/abs/2301.11325","date":"2023-01-26","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"Suno","url":"https://suno.com"},{"name":"Udio","url":"https://www.udio.com"},{"name":"IFPI","url":"https://www.ifpi.org"}],"news":[{"id":"3831e49cc943","date":"2026-10-21","title":"Tod Machover’s new symphony plays with AI while reflecting on democracy","url":"https://news.mit.edu/2026/tod-machover-new-symphony-plays-ai-while-reflecting-democracy-1021","source":"MIT News","lang":"en","kind":"news","summaryZh":"MIT News：作曲家 Tod Machover 的新交响曲把 AI 融入创作与演出，同时探讨民主议题。","via":"pool-v2-20261022","linkCheck":"ok"},{"id":"49643288fe24","date":"2026-10-20","title":"ICMP says AI firms are trying to ‘strong-arm’ Australia on copyright – and that Kylie Minogue’s songs have already been used to train AI (report)","url":"https://www.musicbusinessworldwide.com/icmp-says-ai-firms-are-trying-to-strong-arm-australia-on-copyright-and-that-kylie-minogues-songs-have-already-been-used-to-train-ai-report/","source":"Music Business Worldwide","lang":"en","kind":"news","summaryZh":"Music Business Worldwide于2026-10-20发布的文章，涉及AI 创意媒体、AI 法律政策，原题《ICMP says AI firms are trying to ‘strong-arm’ Australia on copyright – and that Kylie Mino》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"d92393c9d260","date":"2026-10-20","title":"Show HN: Shazam but for AI Slop Music","url":"https://apps.apple.com/de/app/daft-slop-ai-music-detector/id6811013099","source":"hn-show-ai","lang":"en","kind":"news","summaryZh":"hn-show-ai于2026-10-20发布的文章，涉及AI 软件开发、AI 创意媒体，原题《Show HN: Shazam but for AI Slop Music》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"1801c4040f2f","date":"2026-10-19","title":"‘The only way to properly create value in music is to make lasting copyrights.’","url":"https://www.musicbusinessworldwide.com/aaron-bay-schuck-the-only-way-to-properly-create-value-in-music-is-to-make-lasting-copyrights/","source":"Music Business Worldwide","lang":"en","kind":"news","summaryZh":"Music Business Worldwide于2026-10-20发布的文章，涉及AI 创意媒体、AI 法律政策，原题《‘The only way to properly create value in music is to make lasting copyrights.’》，详见原文链接。","via":"collector-run-20261020","linkCheck":"from-feed"},{"id":"a76e2a2a3d83","date":"2026-10-17","title":"The moral case for human music","url":"https://www.ft.com/content/2c21c250-d186-44e1-822c-f65436d213f4","source":"Financial Times AI 栏目","lang":"en","kind":"news","summaryZh":"Financial Times AI 栏目于2026-10-17发布的文章，涉及AI 金融、AI 创意媒体，原题《The moral case for human music》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"191dd3fdcd2b","date":"2026-10-17","title":"Olivia Rodrigo’s expectations music video hit with AI allegations","url":"https://www.creativebloq.com/news/olivia-rodrigos-expectations-music-video-hit-with-ai-allegations","source":"Creative Bloq","lang":"en","kind":"news","summaryZh":"Creative Bloq于2026-10-17发布的文章，涉及AI 创意媒体，原题《Olivia Rodrigo’s expectations music video hit with AI allegations》，详见原文链接。","via":"collector-run-20261020","linkCheck":"from-feed"},{"id":"6388b3d480b1","date":"2026-10-15","title":"Singer or clanker? YouTube Music will tell you with new gen AI labeling.","url":"https://arstechnica.com/ai/2026/10/youtube-music-begins-labeling-ai-songs/","source":"Ars Technica AI","lang":"en","kind":"news","summaryZh":"Ars Technica AI于2026-10-15发布的文章，涉及AI 创意媒体，原题《Singer or clanker? YouTube Music will tell you with new gen AI labeling.》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"41c85884e6bd","date":"2026-10-08","title":"American jailed for commanding 10,000 bots to stream his own AI-generated songs","url":"https://www.tomshardware.com/service-providers/streaming/american-jailed-for-commanding-10-000-bots-to-stream-his-own-ai-generated-songs-and-earn-millions-in-fraudulent-royalty-payments-beating-taylor-swift-is-the-first-person-to-end-up-in-prison-for-ai-assisted-music-streaming-crime","source":"Tom's Hardware AI","lang":"en","kind":"news","summaryZh":"Tom's Hardware AI于2026-10-08发布的文章，涉及AI 芯片硬件、AI 创意媒体，原题《American jailed for commanding 10,000 bots to stream his own AI-generated songs》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"1bb24fc99b27","date":"2026-09-21","title":"Suno made a new AI music model with major labels. Here's what it means","url":"https://www.latimes.com/entertainment-arts/business/story/2026-09-21/ai-music-licensing-deals-labels-artists-who-benefits","source":"Los Angeles Times","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"ok"},{"id":"8dda94cb5c9e","date":"2026-09-09","title":"Suno releases new AI music models in partnership with Warner Music, BMG","url":"https://www.reuters.com/legal/litigation/suno-releases-new-ai-music-models-partnership-with-warner-music-bmg-2026-09-09/","source":"Reuters","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"site-blocks-bots"},{"id":"42b77c184ced","date":"2026-07-29","title":"We’re launching Lyria 3.5 in Google Flow Music, with advances across musicality, lyrics, vocals, and creative control","url":"https://deepmind.google/blog/were-launching-lyria-35-in-google-flow-music-with-advances-across-musicality-lyrics-vocals-and-creative-control/","source":"Google DeepMind Blog","lang":"en","kind":"news","summaryZh":"Google DeepMind Blog于2026-07-30发布的文章，涉及AI 创意媒体，原题《We’re launching Lyria 3.5 in Google Flow Music, with advances across musicality, lyrics, v》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"}],"review":{"lastReviewed":"2026-10-22","cycleDays":90,"nextReview":"2027-01-19"},"urls":{"zh":"/zh-hans/disciplines/music/","en":"/en/disciplines/music/"},"dimensions":{"method":{"score":1,"label":{"zh":"研究方法","en":"Research method"},"evidence":"音乐学研究中 AI 主要用于分析与检索","source":null,"weight":0.3},"discovery":{"score":2,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"AI 歌曲进入 Billboard 部分榜单；Suno 授权模型发布","source":"https://www.reuters.com/legal/litigation/suno-releases-new-ai-music-models-partnership-with-warner-music-bmg-2026-09-09/","weight":0.35},"adoption":{"score":2,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"唱片业从诉讼转向授权合作","source":"https://www.reuters.com/legal/litigation/warner-music-group-settles-copyright-case-with-suno-licensed-ai-music-2025-11-25/","weight":0.25},"education":{"score":1,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到课程证据","source":null,"weight":0.1}},"compositeScore":1.6,"levelCapped":false,"levelChange":{"from":2,"to":2,"direction":"same","reason":null},"biggestBreakthrough":{"date":"2026-10-20","title":{"zh":"Spotify：今年已从推荐中剔除 5000 多万首 AI 垃圾/刷量曲目、另下架 7500 多万首，AI 人设歌手默认不进推荐，AI 创作署名每日新增数万条","en":"Spotify: 50M+ AI slop/spam tracks removed from recommendations this year, 75M+ removed outright; AI Personas excluded from recommendations by default"},"url":"https://newsroom.spotify.com/2026-10-20/prioritizing-human-artistry/","aiImpact":"incremental","fieldImpact":"major","why":{"zh":"AI：平台综合听众行为、上传模式和多家第三方 AI 检测工具识别 AI 垃圾内容，并承认检测会越来越难；学科：最大流媒体平台把“人类创作优先”写进推荐与标注规则，影响音乐分发与收益分配（数字为 Spotify 自述）","en":"AI: combines engagement, upload patterns and third-party AI detectors (acknowledged to be getting harder). Field: the largest streamer codifies human-first recommendation and labelling (Spotify’s figures)"}},"keyPapers":[{"title":"Music Transformer","titleZh":"Music Transformer：生成长时程结构的音乐（arXiv 1809.04281）","id":"arXiv:1809.04281","url":"https://arxiv.org/abs/1809.04281","venue":"arXiv","date":"2018-09-12","why":{"zh":"Transformer 首次用于长篇音乐生成","en":"Transformers for long-form music"},"fieldImpact":"incremental","aiImpact":"major"},{"title":"Jukebox: A Generative Model for Music","titleZh":"Jukebox：OpenAI 生成带人声的原始音频音乐（arXiv 2005.00341）","id":"arXiv:2005.00341","url":"https://arxiv.org/abs/2005.00341","venue":"arXiv","date":"2020-04-30","why":{"zh":"首个能生成带歌声的完整歌曲的模型","en":"First model generating songs with vocals"},"fieldImpact":"incremental","aiImpact":"major"},{"title":"MusicLM: Generating Music From Text","titleZh":"MusicLM：从文本生成高保真音乐（Google，arXiv 2301.11325）","id":"arXiv:2301.11325","url":"https://arxiv.org/abs/2301.11325","venue":"arXiv","date":"2023-01-26","why":{"zh":"文生音乐进入可用水平","en":"Text-to-music reaches usable quality"},"fieldImpact":"incremental","aiImpact":"major"}],"latest_papers":[{"id":"arXiv:2610.29256","title":"Heard But Not Said: Prompting Understates Temporal Information of Musical Moments in Frozen Large Audio-Language Models","titleZh":"“听到却说不出”：提示低估了冻结音频大模型中的音乐时间信息","summaryZh":"用配对查询探针发现，大型音频–语言模型内部其实编码了音乐片段的时间信息，问题在于提示方式难以提取。","summaryEn":"Probes show frozen audio-language models encode when musical events occur, but prompting fails to extract it.","authors":["Eunsik Shin"],"date":"2026-10-21","url":"https://arxiv.org/abs/2610.29256","site":"arXiv","venue":null,"category":"cs.SD","fetchedAt":"2026-10-22T10:10:48"},{"id":"arXiv:2610.29060","title":"Derivation Depth Exposes Reasoning Failures in Tool-Augmented Music Analysis","titleZh":"推导深度暴露工具增强音乐分析中的推理失败","summaryZh":"固定音乐信息、改变从工具输出到答案所需的推理步数，发现模型随推导深度增加明显出错。","summaryEn":"Varying derivation depth from MIR tool outputs reveals reasoning failures in tool-augmented music analysis.","authors":["Rupali Rajesh","Sethuraman T","Savya Khosla","Srinidhi Sunkara"],"date":"2026-10-21","url":"https://arxiv.org/abs/2610.29060","site":"arXiv","venue":null,"category":"cs.SD","fetchedAt":"2026-10-22T10:10:48"},{"id":"arXiv:2610.26684","title":"MuLaCover: Flexible Cover Song Generation via Symbolic Tonal Control","titleZh":"MuLaCover：通过符号化调性控制灵活生成翻唱歌曲","summaryZh":"用预训练音乐语言模型显式控制旋律与和弦，用文本指定歌词和风格，支持变速和结构重排。","summaryEn":"Cover-song generation with symbolic tonal control and text-specified lyrics and style.","authors":["Lynn Yi","Zihan Xiong","Junjie Cao","Xiaolong Weng"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.26684","site":"arXiv","venue":null,"category":"cs.SD","fetchedAt":"2026-10-22T10:10:48"},{"id":"arXiv:2610.22989","title":"MUB: An Identity-Masked Benchmark for Music Understanding in Multimodal Language Models","titleZh":"MUB：面向多模态语言模型音乐理解的身份屏蔽基准","summaryZh":"500 段去除曲名和艺人信息的录音，带速度、调性、拍号和段落标注，检验模型是否真正“听懂”音乐。","summaryEn":"A 500-recording identity-masked benchmark of tempo, key, metre and structure.","authors":["Kun Liu","Lyra Tao"],"date":"2026-10-16","url":"https://arxiv.org/abs/2610.22989","site":"arXiv","venue":null,"category":"cs.SD","fetchedAt":"2026-10-22T10:10:48"},{"id":"arXiv:2610.21467","title":"DuET: Note-Frame Dual Encoding for Piano Performance-to-Score Transcription","titleZh":"DuET：音符-帧双编码的钢琴演奏转谱","summaryZh":"双编码器 Transformer 联合学习音符级演奏 MIDI 转乐谱和帧级节拍追踪。","summaryEn":"A dual-encoder transformer for piano performance-to-score transcription with beat tracking.","authors":["Wei Zeng","Ye Wang"],"date":"2026-10-15","url":"https://arxiv.org/abs/2610.21467","site":"arXiv","venue":null,"category":"cs.SD","fetchedAt":"2026-10-22T10:10:48"},{"id":"arXiv:2610.20310","title":"MUSEBENCH: A Benchmark for AI-Generated Music Detection and Artifact Analysis","titleZh":"MUSEBENCH：AI 生成音乐检测与伪迹分析基准","summaryZh":"含 36.1 万多首曲目，包括人类作品、Suno、Udio 等 10 个模型生成的音乐、人机共创及对抗处理版本。","summaryEn":"361K+ tracks for detecting AI-generated music from 10 models including Suno and Udio.","authors":["Yi Chen Liu","Syed Irfan Ali Meerza","Jiawei Yu","Xi Guo"],"date":"2026-10-14","url":"https://arxiv.org/abs/2610.20310","site":"arXiv","venue":null,"category":"cs.SD","fetchedAt":"2026-10-22T10:10:48"},{"id":"arXiv:2610.17036","title":"PitchIn: Natural-Language Vocal Coaching from Unstructured Community Feedback","titleZh":"PitchIn：从社区非结构化反馈中学习自然语言声乐指导","summaryZh":"利用在线歌唱社区散落的评论作为监督，训练音频-语言模型给出声乐问题诊断与建议。","summaryEn":"Audio-language models learn vocal coaching from unstructured singing-community comments.","authors":["Hayeon Bang","Hounsu Kim","Joonyong Park","Jung-Jae Yu"],"date":"2026-10-12","url":"https://arxiv.org/abs/2610.17036","site":"arXiv","venue":null,"category":"cs.SD","fetchedAt":"2026-10-22T10:10:48"},{"id":"arXiv:2610.15957","title":"Harmonic Representations in Symbolic Music Transformers are Localized and Controllable","titleZh":"符号音乐 Transformer 中的和声表示是局部且可控的","summaryZh":"以和声分析为例，研究符号音乐 Transformer 内部如何表示音乐结构，并利用这些表示控制生成。","summaryEn":"Harmonic representations in symbolic music transformers are localised and can steer generation.","authors":["Adi Silberschein","Megan Wei","Tamar Rott Shaham"],"date":"2026-10-11","url":"https://arxiv.org/abs/2610.15957","site":"arXiv","venue":null,"category":"cs.SD","fetchedAt":"2026-10-22T10:10:48"}]},{"id":"visual-arts-design","slug":"visual-arts-design","category":"arts","categoryName":{"zh":"艺术","en":"Arts"},"name":{"zh":"美术与设计","en":"Visual arts & design"},"classification":{"oecdFos":"6.4","moe":"1356 美术与书法 / 1357 设计"},"level":2,"levelKey":"breakthrough","levelLabel":{"zh":"局部突破","en":"Local breakthroughs"},"trend":"flat","reason":{"zh":"AI 作品已进入艺术市场与奖项：2018 年佳士得拍出 AI 肖像《Edmond de Belamy》，2022 年 Midjourney 作品获科罗拉多州博览会数字艺术一等奖并引发争议；设计行业广泛使用生成式工具。","en":"AI works have entered markets and prizes: Christie's sold “Portrait of Edmond de Belamy” in 2018, and a Midjourney image won the Colorado State Fair digital-art prize in 2022; design widely uses generative tools."},"reasonSources":[{"url":"https://www.christies.com/features/A-collaboration-between-two-artists-one-human-one-a-machine-9332-1.aspx","linkCheck":"ok"},{"url":"https://www.nytimes.com/2022/09/02/technology/ai-artificial-intelligence-artists.html","linkCheck":"site-blocks-bots"}],"directions":{"zh":["图像生成","设计工具集成","版权与署名"],"en":["Image generation","Design-tool integration","Copyright & attribution"]},"stats":{"milestones":14,"milestonesSince2025":9,"milestones2026":8,"milestonesLast6Months":7,"news":6,"latestProgress":"2026-10-20","latestMilestone":"2026-10-20","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":0,"leaps":{"ai":1,"field":0},"latestPapers":8,"keyPapers":4},"milestones":[{"date":"2018-10-25","title":{"zh":"佳士得拍出 AI 生成肖像《Edmond de Belamy》","en":"Christie's sells AI “Portrait of Edmond de Belamy”"},"url":"https://www.christies.com/features/A-collaboration-between-two-artists-one-human-one-a-machine-9332-1.aspx","source":"Christie's","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"industry","typeLabel":{"zh":"产业/采用","en":"Industry"},"why":{"zh":"首件在大型拍卖行拍卖的 AI 作品，成交 43.25 万美元","en":"First AI artwork sold at a major auction house ($432,500)"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首件在大型拍卖行拍卖的 AI 作品，成交 43.25 万美元","en":"First AI artwork sold at a major auction house ($432,500)"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首件在大型拍卖行拍卖的 AI 作品，成交 43.25 万美元","en":"First AI artwork sold at a major auction house ($432,500)"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2021-12-20","title":{"zh":"潜在扩散模型（Stable Diffusion 基础，arXiv 2112.10752）","en":"Latent diffusion models (basis of Stable Diffusion)"},"url":"https://arxiv.org/abs/2112.10752","source":"arXiv","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"system","typeLabel":{"zh":"系统发布","en":"System"},"why":{"zh":"让高质量图像生成可在消费级显卡运行，催生开源图像生成生态","en":"Made high-quality image generation run on consumer GPUs"},"aiImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"让高质量图像生成可在消费级显卡运行，催生开源图像生成生态","en":"Made high-quality image generation run on consumer GPUs"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"让高质量图像生成可在消费级显卡运行，催生开源图像生成生态","en":"Made high-quality image generation run on consumer GPUs"}},"impactNotes":[],"verified":{"id":"arXiv:2112.10752","title":"High-Resolution Image Synthesis with Latent Diffusion Models","venue":"arXiv","date":"2021-12-20"},"addedIn":"v2"},{"date":"2022-09-02","title":{"zh":"Midjourney 作品获科罗拉多州博览会数字艺术一等奖","en":"Midjourney image wins Colorado State Fair digital-art prize"},"url":"https://www.nytimes.com/2022/09/02/technology/ai-artificial-intelligence-artists.html","source":"The New York Times","flags":[],"flagLabels":[],"linkCheck":"site-blocks-bots","datePrecision":"day","type":"prize","typeLabel":{"zh":"奖项","en":"Prize"},"why":{"zh":"AI 作品获州博览会数字艺术一等奖并引发争议","en":"AI work wins a state fair prize, sparking controversy"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI 作品获州博览会数字艺术一等奖并引发争议","en":"AI work wins a state fair prize, sparking controversy"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI 作品获州博览会数字艺术一等奖并引发争议","en":"AI work wins a state fair prize, sparking controversy"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2023-06-16","title":{"zh":"艺术与生成式 AI 的科学（Science 政策论坛）","en":"Art and the science of generative AI (Science)"},"url":"https://doi.org/10.1126/science.adh4451","source":"Science","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"系统讨论生成式 AI 对艺术创作、版权和审美的影响","en":"Systematic discussion of impacts on art"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"系统讨论生成式 AI 对艺术创作、版权和审美的影响","en":"Systematic discussion of impacts on art"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"系统讨论生成式 AI 对艺术创作、版权和审美的影响","en":"Systematic discussion of impacts on art"}},"impactNotes":[],"verified":{"id":"10.1126/science.adh4451","title":"Art and the science of generative AI","venue":"Science","date":"2023-06-16"},"addedIn":"v2"},{"date":"2024-03-05","title":{"zh":"生成式 AI 让艺术家产出提升 25%、作品更受欢迎，但新颖性随时间下降（PNAS Nexus）","en":"Generative AI raises artists’ productivity 25% and favourability, novelty declines (PNAS Nexus)"},"url":"https://doi.org/10.1093/pnasnexus/pgae052","source":"PNAS Nexus","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"基于 5 万多名艺术家数据的实证研究","en":"Empirical study of 50k+ artists"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"基于 5 万多名艺术家数据的实证研究","en":"Empirical study of 50k+ artists"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"基于 5 万多名艺术家数据的实证研究","en":"Empirical study of 50k+ artists"}},"impactNotes":[],"verified":{"id":"10.1093/pnasnexus/pgae052","title":"Generative artificial intelligence, human creativity, and art","venue":"PNAS Nexus","date":"2024-03-05"},"addedIn":"v2"},{"date":"2025-03-05","title":{"zh":"佳士得首场 AI 艺术专场“Augmented Intelligence”成交 72.9 万美元，34 件拍品成交 28 件","en":"Christie’s first AI art sale “Augmented Intelligence” totals $728,784"},"url":"https://press.christies.com/augmented-intelligence-totals-728784/","source":"Christie’s","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"industry","typeLabel":{"zh":"产业/采用","en":"Industry"},"why":{"zh":"顶级拍卖行为 AI 艺术设立专场，伴随艺术家联名抗议","en":"Major auction house hosts dedicated AI sale amid artist protest"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"顶级拍卖行为 AI 艺术设立专场，伴随艺术家联名抗议","en":"Major auction house hosts dedicated AI sale amid artist protest"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"顶级拍卖行为 AI 艺术设立专场，伴随艺术家联名抗议","en":"Major auction house hosts dedicated AI sale amid artist protest"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-03-02","title":{"zh":"美国最高法院拒绝受理 Thaler v. Perlmutter，维持“无人类作者的 AI 生成艺术不受版权保护”","en":"US Supreme Court declines Thaler v. Perlmutter, leaving AI-only art without copyright"},"url":"https://www.reuters.com/legal/government/us-supreme-court-declines-hear-dispute-over-copyrights-ai-generated-material-2026-03-02/","source":"Reuters","flags":[],"flagLabels":[],"linkCheck":"site-blocks-bots","datePrecision":"day","type":"legal","typeLabel":{"zh":"诉讼/判决","en":"Legal"},"why":{"zh":"确定了 AI 生成艺术的版权边界","en":"Settles the copyright boundary for AI-only art"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"确定了 AI 生成艺术的版权边界","en":"Settles the copyright boundary for AI-only art"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"确定了 AI 生成艺术的版权边界","en":"Settles the copyright boundary for AI-only art"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-06-30","title":{"zh":"纽约现代艺术博物馆（MoMA）收藏 Domestic Data Streamers 用 AI 重建个人记忆的作品 9 件","en":"MoMA acquires nine AI “Synthetic Memories” works by Domestic Data Streamers"},"url":"https://www.elisava.net/en/moma-acquires-nine-works-by-domestic-data-streamers-a-collective-formed-by-elisava-alumni/","source":"Elisava","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"industry","typeLabel":{"zh":"产业/采用","en":"Industry"},"why":{"zh":"顶级美术馆持续收藏 AI 艺术（来源为艺术家母校，待更多来源印证）","en":"Leading museum keeps collecting AI art (single school source)"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"顶级美术馆持续收藏 AI 艺术（来源为艺术家母校，待更多来源印证）","en":"Leading museum keeps collecting AI art (single school source)"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"顶级美术馆持续收藏 AI 艺术（来源为艺术家母校，待更多来源印证）","en":"Leading museum keeps collecting AI art (single school source)"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-07-24","title":{"zh":"PNAS：上下文感知多模态 AI 分析五个世纪 3.5 万幅西方绘画，揭示艺术演化的隐藏路径","en":"Context-aware multimodal AI navigates hidden pathways in five centuries of art evolution (PNAS)"},"url":"https://www.pnas.org/doi/10.1073/pnas.2517969123","source":"PNAS","flags":[],"flagLabels":[],"linkCheck":"ok-crossref","datePrecision":"day","type":"paper","typeLabel":{"zh":"论文","en":"Paper"},"why":{"zh":"AI：用 Stable Diffusion/CLIP 表征区分时期、风格与画家；学科：语义上下文特征比形式特征更能解释艺术史分期，为定量艺术史提供新方法","en":"AI: Stable Diffusion/CLIP representations separate periods and artists; Field: semantic context explains art-historical periods better than form"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：用 Stable Diffusion/CLIP 表征区分时期、风格与画家；学科：语义上下文特征比形式特征更能解释艺术史分期，为定量艺术史提供新方法","en":"AI: Stable Diffusion/CLIP representations separate periods and artists; Field: semantic context explains art-historical periods better than form"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：用 Stable Diffusion/CLIP 表征区分时期、风格与画家；学科：语义上下文特征比形式特征更能解释艺术史分期，为定量艺术史提供新方法","en":"AI: Stable Diffusion/CLIP representations separate periods and artists; Field: semantic context explains art-historical periods better than form"}},"impactNotes":[],"verified":{"id":"10.1073/pnas.2517969123","title":"Context-aware multimodal AI navigates hidden pathways in five centuries of art evolution","venue":"Proceedings of the National Academy of Sciences","date":"2026-07-24"},"addedIn":"v2"},{"date":"2026-09-02","title":{"zh":"印度版权局在 DABUS 美术作品案中拒绝将 AI 登记为作者：作品可具独创性，但作者须是促成创作的人","en":"India’s Copyright Office rejects AI system DABUS as author of an artwork"},"url":"https://timesofindia.indiatimes.com/city/hyderabad/prompter-or-creator-indias-copyright-office-rejects-ai-dabus-as-author-in-artwork-case/articleshow/133699120.cms","source":"The Times of India","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"legal","typeLabel":{"zh":"诉讼/判决","en":"Legal"},"why":{"zh":"AI：AI 不能被承认为作者；学科：继美国最高法院拒审 Thaler 案后，印度在美术作品上采取相同立场，同时承认 AI 参与的作品可以有人类作者","en":"AI: an AI cannot be an author. Field: India aligns with the US position while allowing human authorship of AI-assisted works"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：AI 不能被承认为作者；学科：继美国最高法院拒审 Thaler 案后，印度在美术作品上采取相同立场，同时承认 AI 参与的作品可以有人类作者","en":"AI: an AI cannot be an author. Field: India aligns with the US position while allowing human authorship of AI-assisted works"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：AI 不能被承认为作者；学科：继美国最高法院拒审 Thaler 案后，印度在美术作品上采取相同立场，同时承认 AI 参与的作品可以有人类作者","en":"AI: an AI cannot be an author. Field: India aligns with the US position while allowing human authorship of AI-assisted works"}},"impactNotes":[],"verified":null,"addedIn":"v2-20261022pm"},{"date":"2026-09-19","title":{"zh":"麦迪逊当代艺术博物馆（MMoCA）展出 Beeple 生成式 AI 装置《Diffuse Control》，观众可实时改写馆藏作品图像；900 多人联署公开信要求撤展，指其未经艺术家同意使用作品","en":"Madison Museum of Contemporary Art opens Beeple’s generative-AI installation “Diffuse Control”; an open letter with 900+ signatures demands its closure over artist consent"},"url":"https://www.wpr.org/news/madison-art-museum-new-exhibition-generative-ai-pushback","source":"WPR","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"industry","typeLabel":{"zh":"产业/采用","en":"Industry"},"why":{"zh":"AI：12 块同步屏幕，访客用手机和控制台改写由馆藏作品生成的基础图像；学科：美术馆把馆藏作品交给生成式 AI 处理，引发“公有领域作品能否喂给 AI”的同意与伦理争论，馆方称依法取得了在世版权人同意","en":"AI: 12 synced screens let visitors manipulate images derived from the collection. Field: sparks a consent debate over feeding (even public-domain) collection works to generative AI; the museum says legally required consents were obtained"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：12 块同步屏幕，访客用手机和控制台改写由馆藏作品生成的基础图像；学科：美术馆把馆藏作品交给生成式 AI 处理，引发“公有领域作品能否喂给 AI”的同意与伦理争论，馆方称依法取得了在世版权人同意","en":"AI: 12 synced screens let visitors manipulate images derived from the collection. Field: sparks a consent debate over feeding (even public-domain) collection works to generative AI; the museum says legally required consents were obtained"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：12 块同步屏幕，访客用手机和控制台改写由馆藏作品生成的基础图像；学科：美术馆把馆藏作品交给生成式 AI 处理，引发“公有领域作品能否喂给 AI”的同意与伦理争论，馆方称依法取得了在世版权人同意","en":"AI: 12 synced screens let visitors manipulate images derived from the collection. Field: sparks a consent debate over feeding (even public-domain) collection works to generative AI; the museum says legally required consents were obtained"}},"impactNotes":[],"verified":null,"addedIn":"v2-20261022audit"},{"date":"2026-09-23","title":{"zh":"长沙开福区法院：仅有概括性提示词和瑕疵修正的 AI 生成图不具独创性，不构成美术作品，驳回侵权诉请（判决已生效）","en":"Changsha Kaifu court: an AI-generated image made with generic prompts and minor fixes lacks originality and is not a protected work"},"url":"https://finance.sina.com.cn/jjxw/2026-09-23/doc-inisvqcz1105942.shtml","source":"华声在线 / 新浪财经","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"legal","typeLabel":{"zh":"诉讼/判决","en":"Legal"},"why":{"zh":"AI：法院把生成过程拆为前端构思、生成控制、后端加工三层逐层审查独创性；学科：为 AI 图像能否受著作权保护给出可操作的审查标准，并提醒创作者保留提示词与修改记录","en":"AI: court reviews originality across three layers (concept, generation control, post-editing). Field: operational test for copyright in AI images"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：法院把生成过程拆为前端构思、生成控制、后端加工三层逐层审查独创性；学科：为 AI 图像能否受著作权保护给出可操作的审查标准，并提醒创作者保留提示词与修改记录","en":"AI: court reviews originality across three layers (concept, generation control, post-editing). Field: operational test for copyright in AI images"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：法院把生成过程拆为前端构思、生成控制、后端加工三层逐层审查独创性；学科：为 AI 图像能否受著作权保护给出可操作的审查标准，并提醒创作者保留提示词与修改记录","en":"AI: court reviews originality across three layers (concept, generation control, post-editing). Field: operational test for copyright in AI images"}},"impactNotes":[],"verified":null,"addedIn":"v2-20261022pm"},{"date":"2026-10-10","title":{"zh":"日本百合漫画展“百合展”因宣传图和装饰使用生成式 AI 遭漫画家批评，取消 10 周年池袋活动，主办方承诺今后完全不用生成式 AI","en":"Japan’s Yuriten exhibition cancels its 10th-anniversary Ikebukuro event after manga artists criticise AI-generated promotional imagery, pledging no generative AI in future"},"url":"https://www.animenewsnetwork.com/interest/2026-10-10/yuriten-exhibit-cancels-10th-anniversary-event-after-artists-criticize-its-ai-use/.242698","source":"Anime News Network","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"industry","typeLabel":{"zh":"产业/采用","en":"Industry"},"why":{"zh":"AI：外包方在海报与会场装饰中使用 AI 图像；学科：插画与漫画界对生成式 AI 的抵制已能直接取消商业活动，“不用 AI”正成为展会和品牌的审查条款","en":"AI: a contractor used AI imagery for posters and decorations. Field: illustrator backlash can now cancel commercial events; “no generative AI” is becoming a vetting clause"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：外包方在海报与会场装饰中使用 AI 图像；学科：插画与漫画界对生成式 AI 的抵制已能直接取消商业活动，“不用 AI”正成为展会和品牌的审查条款","en":"AI: a contractor used AI imagery for posters and decorations. Field: illustrator backlash can now cancel commercial events; “no generative AI” is becoming a vetting clause"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：外包方在海报与会场装饰中使用 AI 图像；学科：插画与漫画界对生成式 AI 的抵制已能直接取消商业活动，“不用 AI”正成为展会和品牌的审查条款","en":"AI: a contractor used AI imagery for posters and decorations. Field: illustrator backlash can now cancel commercial events; “no generative AI” is becoming a vetting clause"}},"impactNotes":[],"verified":null,"addedIn":"v2-20261022audit"},{"date":"2026-10-20","title":{"zh":"巴黎小皇宫美术馆开幕“Visionaries”展：谷歌艺术与文化用 Gemini 等生成式模型，依据档案照片和书信重建 19 世纪女性雕塑家（含卡米耶·克洛岱尔）已消失的工作室，展期至 2027 年 1 月","en":"Petit Palais opens “Visionaries”: Google Arts & Culture uses Gemini-based tools to reconstruct the lost studios of 19th-century women sculptors from archival photos and letters"},"url":"https://blog.google/company-news/outreach-and-initiatives/arts-culture/visionaries-petit-palais-art-basel-paris-prune-nourry/","source":"Google Blog","flags":["vendor"],"flagLabels":[{"zh":"厂商自述","en":"Vendor claim"}],"linkCheck":"ok","datePrecision":"day","type":"industry","typeLabel":{"zh":"产业/采用","en":"Industry"},"why":{"zh":"AI：用定制 Gemini 工具把档案照片转为 3D 模型并生成镜头路径；学科：生成式 AI 被国家级美术馆用于艺术史“复原”叙事，但复原画面属推测性重建，不是史料","en":"AI: a custom Gemini tool turns archival photos into 3D models. Field: a major museum uses generative AI for art-historical reconstruction — speculative imagery, not historical evidence"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：用定制 Gemini 工具把档案照片转为 3D 模型并生成镜头路径；学科：生成式 AI 被国家级美术馆用于艺术史“复原”叙事，但复原画面属推测性重建，不是史料","en":"AI: a custom Gemini tool turns archival photos into 3D models. Field: a major museum uses generative AI for art-historical reconstruction — speculative imagery, not historical evidence"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：用定制 Gemini 工具把档案照片转为 3D 模型并生成镜头路径；学科：生成式 AI 被国家级美术馆用于艺术史“复原”叙事，但复原画面属推测性重建，不是史料","en":"AI: a custom Gemini tool turns archival photos into 3D models. Field: a major museum uses generative AI for art-historical reconstruction — speculative imagery, not historical evidence"}},"impactNotes":[],"verified":null,"addedIn":"v2-20261022audit"}],"papersRepos":[{"title":"High-Resolution Image Synthesis with Latent Diffusion Models","url":"https://arxiv.org/abs/2112.10752","date":"2021-12-20","kind":"paper","linkCheck":"ok-crossref"},{"title":"Art and the science of generative AI","url":"https://doi.org/10.1126/science.adh4451","date":"2023-06-16","kind":"paper","linkCheck":"ok-crossref"},{"title":"Generative artificial intelligence, human creativity, and art","url":"https://doi.org/10.1093/pnasnexus/pgae052","date":"2024-03-05","kind":"paper","linkCheck":"ok-crossref"},{"title":"Context-aware multimodal AI navigates hidden pathways in five centuries of art evolution","url":"https://doi.org/10.1073/pnas.2517969123","date":"2026-07-24","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"Adobe Firefly","url":"https://www.adobe.com/products/firefly.html"},{"name":"Midjourney","url":"https://www.midjourney.com"}],"news":[{"id":"cf5b9e5a110d","date":"2026-10-20","title":"AI creates art in seconds, but creative growth still matters, paper argues","url":"https://techxplore.com/news/2026-10-ai-art-seconds-creative-growth.html","source":"Phys.org 机器学习与AI","lang":"en","kind":"news","summaryZh":"Phys.org 机器学习与AI于2026-10-20发布的文章，涉及AI 科研与调研、AI 创意媒体，原题《AI creates art in seconds, but creative growth still matters, paper argues》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"0a752e7d1bb6","date":"2026-10-14","title":"Can AI help ‘humanise’ museums? New report reminds us why art matters in a changing world","url":"https://theconversation.com/can-ai-help-humanise-museums-new-report-reminds-us-why-art-matters-in-a-changing-world-294227","source":"The Conversation · AI","lang":"en","kind":"news","summaryZh":"The Conversation · AI于2026-10-14发布的文章，涉及AI 科研与调研、AI 创意媒体，原题《Can AI help ‘humanise’ museums? New report reminds us why art matters in a changing world》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"3ded07aae24f","date":"2026-09-17","title":"Scaling Intelligence (Gray Area exhibition, Oct 28–Nov 1)","url":"https://grayarea.org/exhibitions/scaling-intelligence-modal-commissions/","source":"Gray Area","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"unchecked"},{"id":"4ab1629cc2f6","date":"2026-08-27","title":"Student Artists Wrestle with AI’s Promise and Peril","url":"https://edsurge.com/news/student-artists-wrestle-with-ais-promise-and-peril","source":"EdSurge","lang":"en","kind":"news","summaryZh":"EdSurge于2026-08-27发布的文章，涉及AI 教育、AI 创意媒体，原题《Student Artists Wrestle with AI’s Promise and Peril》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"b48c2a5a69f4","date":"2026-08-18","title":"When AI art has no author: Study finds generated images often can’t be traced to training data","url":"https://news.mit.edu/2026/when-ai-art-has-no-author-generated-images-often-cant-be-traced-to-training-data-0818","source":"MIT News：Artificial intelligence","lang":"en","kind":"news","summaryZh":"MIT News：Artificial intelligence于2026-08-19发布的文章，涉及AI 科研与调研、AI 创意媒体，原题《When AI art has no author: Study finds generated images often can’t be traced to training 》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"0a9179664e16","date":"2026-07-07","title":"Inside the Emerging AI Art Market","url":"https://spectrum.ieee.org/ai-art-market","source":"IEEE Spectrum","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"unchecked"}],"review":{"lastReviewed":"2026-10-22","cycleDays":90,"nextReview":"2027-01-19"},"urls":{"zh":"/zh-hans/disciplines/visual-arts-design/","en":"/en/disciplines/visual-arts-design/"},"dimensions":{"method":{"score":1,"label":{"zh":"研究方法","en":"Research method"},"evidence":"艺术史研究中 AI 辅助有限","source":null,"weight":0.3},"discovery":{"score":2,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"AI 作品进入拍卖与奖项，引发作者身份争议","source":"https://doi.org/10.1126/science.adh4451","weight":0.35},"adoption":{"score":2,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"设计行业广泛使用生成式工具，缺权威调查","source":null,"weight":0.25},"education":{"score":1,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到课程证据","source":null,"weight":0.1}},"compositeScore":1.6,"levelCapped":false,"levelChange":{"from":2,"to":2,"direction":"same","reason":null},"biggestBreakthrough":{"date":"2021-12-20","title":{"zh":"潜在扩散模型（Stable Diffusion 基础，arXiv 2112.10752）","en":"Latent diffusion models (basis of Stable Diffusion)"},"url":"https://arxiv.org/abs/2112.10752","aiImpact":"leap","fieldImpact":"major","why":{"zh":"让高质量图像生成可在消费级显卡运行，催生开源图像生成生态","en":"Made high-quality image generation run on consumer GPUs"}},"keyPapers":[{"title":"High-Resolution Image Synthesis with Latent Diffusion Models","titleZh":"潜在扩散模型（Stable Diffusion 基础，arXiv 2112.10752）","id":"arXiv:2112.10752","url":"https://arxiv.org/abs/2112.10752","venue":"arXiv","date":"2021-12-20","why":{"zh":"让高质量图像生成可在消费级显卡运行，催生开源图像生成生态","en":"Made high-quality image generation run on consumer GPUs"},"fieldImpact":"major","aiImpact":"leap"},{"title":"Art and the science of generative AI","titleZh":"艺术与生成式 AI 的科学（Science 政策论坛）","id":"10.1126/science.adh4451","url":"https://doi.org/10.1126/science.adh4451","venue":"Science","date":"2023-06-16","why":{"zh":"系统讨论生成式 AI 对艺术创作、版权和审美的影响","en":"Systematic discussion of impacts on art"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Generative artificial intelligence, human creativity, and art","titleZh":"生成式 AI 让艺术家产出提升 25%、作品更受欢迎，但新颖性随时间下降（PNAS Nexus）","id":"10.1093/pnasnexus/pgae052","url":"https://doi.org/10.1093/pnasnexus/pgae052","venue":"PNAS Nexus","date":"2024-03-05","why":{"zh":"基于 5 万多名艺术家数据的实证研究","en":"Empirical study of 50k+ artists"},"fieldImpact":"major","aiImpact":"incremental"},{"title":"Context-aware multimodal AI navigates hidden pathways in five centuries of art evolution","titleZh":"PNAS：上下文感知多模态 AI 分析五个世纪 3.5 万幅西方绘画，揭示艺术演化的隐藏路径","id":"10.1073/pnas.2517969123","url":"https://doi.org/10.1073/pnas.2517969123","venue":"Proceedings of the National Academy of Sciences","date":"2026-07-24","why":{"zh":"AI：用 Stable Diffusion/CLIP 表征区分时期、风格与画家；学科：语义上下文特征比形式特征更能解释艺术史分期，为定量艺术史提供新方法","en":"AI: Stable Diffusion/CLIP representations separate periods and artists; Field: semantic context explains art-historical periods better than form"},"fieldImpact":"major","aiImpact":"incremental"}],"latest_papers":[{"id":"arXiv:2610.30033","title":"RLMuscle: Muscle-Driven Character Animation via Reinforcement Learning","titleZh":"RLMuscle：用强化学习驱动肌肉的角色动画","summaryZh":"强化学习控制器自动找出跟踪参考动作所需的每块肌肉激活，减少动画师手工设定并提升形变真实感。","summaryEn":"An RL controller discovers per-muscle activations to track reference motion for realistic character animation.","authors":["Chunlei Li","Siyuan Yu","Yang Gao","Shuai Li"],"date":"2026-10-21","url":"https://arxiv.org/abs/2610.30033","site":"arXiv","venue":null,"category":"cs.GR","fetchedAt":"2026-10-22T10:10:51"},{"id":"arXiv:2610.29003","title":"T-DIG: Text-Conditioned Diversity for Image Generation","titleZh":"T-DIG：文本条件控制的图像生成多样性","summaryZh":"让用户用文字指定希望变化的维度（背景、风格、服装等），提升文生图多图结果的有用多样性。","summaryEn":"Lets users specify via text which aspects should vary, improving contextual diversity in text-to-image outputs.","authors":["Simran Khanuja","Rohit Gandikota","Gaurav Parmar","Graham Neubig"],"date":"2026-10-21","url":"https://arxiv.org/abs/2610.29003","site":"arXiv","venue":null,"category":"cs.CV","fetchedAt":"2026-10-22T10:10:51"},{"id":"arXiv:2610.28416","title":"Code2Video-Compass: A Comprehensive Benchmark for General-Purpose Code-Based Video Generation across T2V, I2V, and V2V","titleZh":"Code2Video-Compass：通用的基于代码的视频生成基准","summaryZh":"包含 2104 个任务、10 个类别，覆盖文生视频、图生视频和视频生视频，评估多模态大模型把视觉与时间需求写成可执行程序来生成完整视频的能力。","summaryEn":"A 2,104-task benchmark for MLLMs generating complete videos via code across T2V, I2V and V2V.","authors":["Zhenghan Chen","Zekai Shao","Xin Lin","Leting Cheng"],"date":"2026-10-20","url":"https://arxiv.org/abs/2610.28416","site":"arXiv","venue":null,"category":"cs.CV","fetchedAt":"2026-10-22T10:10:51"},{"id":"arXiv:2610.22744","title":"OmniXRef: A Unified Framework for Multimodal Multi-Reference Video Editing","titleZh":"OmniXRef：多模态多参考视频编辑统一框架","summaryZh":"允许组合文本、图像和视频参考，控制源视频的多种属性。","summaryEn":"Unified multi-reference video editing combining text, image and video references.","authors":["Xiaoyan Cong","Tianyu Wang","Hyeonho Jeong","Zhifei Zhang"],"date":"2026-10-15","url":"https://arxiv.org/abs/2610.22744","site":"arXiv","venue":null,"category":"cs.CV","fetchedAt":"2026-10-22T10:10:51"},{"id":"arXiv:2610.24010","title":"Beyond Explicit Intent: Discovering and Shaping Artistic Intent through Human-AI Co-Creation","titleZh":"超越明确意图：在人机共创中发现并塑造艺术意图","summaryZh":"让艺术家通过观察、比较和选择多样的生成结果，发现原本说不清的偏好，而不是事先写好提示。","summaryEn":"Human–AI co-creation helps artists discover latent intentions by exploring diverse visual options.","authors":["Renye Yan","Jialun Zhong","Jikang Cheng","Yi Gao"],"date":"2026-10-15","url":"https://arxiv.org/abs/2610.24010","site":"arXiv","venue":null,"category":"cs.AI","fetchedAt":"2026-10-22T10:10:51"},{"id":"arXiv:2610.20283","title":"From Semantic Editing Intentions to Persistent Parametric Controls for AI-Assisted Image Editing","titleZh":"从语义编辑意图到持久的参数化控制：AI 辅助图像编辑","summaryZh":"Atelier 工具根据图像语义和用户意图生成可直接调整的编辑控件，而非直接给出成品。","summaryEn":"Atelier surfaces editable parametric controls from semantic editing intent.","authors":["Rifat Mehreen Amin","Michelle Dutoit","Anton Rockenstein","Baptiste Caramiaux"],"date":"2026-10-14","url":"https://arxiv.org/abs/2610.20283","site":"arXiv","venue":null,"category":"cs.HC","fetchedAt":"2026-10-22T10:10:51"},{"id":"arXiv:2610.20161","title":"The Body-Algorithm-Emotion Loop: Making Emotion AI Misrecognition Visible through Artistic Practice","titleZh":"身体-算法-情绪回路：用艺术实践呈现情绪 AI 的误识别","summaryZh":"互动装置模拟商业情绪 AI 如何把面部动作归类，让观众亲身体验算法误识别。","summaryEn":"An interactive installation making Emotion-AI misrecognition visible through art.","authors":["Chunyi Shen"],"date":"2026-10-14","url":"https://arxiv.org/abs/2610.20161","site":"arXiv","venue":null,"category":"cs.HC","fetchedAt":"2026-10-22T10:10:51"},{"id":"arXiv:2610.19025","title":"How Far Can Coding Agents Go in Visual Generation?","titleZh":"编码智能体在视觉生成上能走多远？","summaryZh":"在 200 个提示的 AgentDrawBench 上，通用编码智能体通过写程序作图，擅长结构和精确约束，但感知丰富度有限。","summaryEn":"Coding agents drawing via programs excel at structure and constraints but lack perceptual richness.","authors":["Hengguang Zhou","Yuanhao Ban","Jiaqi Feng","Xiaohuan Pei"],"date":"2026-10-13","url":"https://arxiv.org/abs/2610.19025","site":"arXiv","venue":null,"category":"cs.CV","fetchedAt":"2026-10-22T10:10:51"}]},{"id":"film-animation","slug":"film-animation","category":"arts","categoryName":{"zh":"艺术","en":"Arts"},"name":{"zh":"影视与动画","en":"Film & animation"},"classification":{"oecdFos":"6.4","moe":"1354 戏剧与影视"},"level":2,"levelKey":"breakthrough","levelLabel":{"zh":"局部突破","en":"Local breakthroughs"},"trend":"up","reason":{"zh":"文生视频（Sora 2024 起）进入制作流程；第 99 届奥斯卡规则明确使用生成式 AI 不影响参评，但表演和编剧奖须为人类作品（2026-05）。","en":"Text-to-video (Sora from 2024) has entered production; the 99th Oscars rules (May 2026) say generative AI use does not disqualify a film, but acting and writing awards require human work."},"reasonSources":[{"url":"https://openai.com/index/sora/","linkCheck":"site-blocks-bots"},{"url":"https://www.reuters.com/lifestyle/ai-actors-writers-will-be-ineligible-oscars-2026-05-01/","linkCheck":"site-blocks-bots"}],"directions":{"zh":["文生视频","虚拟制片与特效","配音与数字人"],"en":["Text-to-video","Virtual production & VFX","Dubbing & digital humans"]},"stats":{"milestones":7,"milestonesSince2025":4,"milestones2026":3,"milestonesLast6Months":3,"news":12,"latestProgress":"2026-10-20","latestMilestone":"2026-10-10","monthsSinceLatestMilestone":0,"staleLabel":null,"pendingVerification":0,"leaps":{"ai":1,"field":0},"latestPapers":8,"keyPapers":2},"milestones":[{"date":"2023-11-25","title":{"zh":"Stable Video Diffusion：开源视频生成模型（arXiv 2311.15127）","en":"Stable Video Diffusion: open video generation model"},"url":"https://arxiv.org/abs/2311.15127","source":"arXiv","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok-crossref","datePrecision":"day","type":"system","typeLabel":{"zh":"系统发布","en":"System"},"why":{"zh":"开源视频生成进入可用阶段","en":"Open video generation becomes usable"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"开源视频生成进入可用阶段","en":"Open video generation becomes usable"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"开源视频生成进入可用阶段","en":"Open video generation becomes usable"}},"impactNotes":[],"verified":{"id":"arXiv:2311.15127","title":"Stable Video Diffusion: Scaling Latent Video Diffusion Models to Large Datasets","venue":"arXiv","date":"2023-11-25"},"addedIn":"v2"},{"date":"2024-02-15","title":{"zh":"OpenAI 发布文生视频模型 Sora","en":"OpenAI unveils text-to-video model Sora"},"url":"https://openai.com/index/sora/","source":"OpenAI","flags":[],"flagLabels":[],"linkCheck":"site-blocks-bots","datePrecision":"day","type":"system","typeLabel":{"zh":"系统发布","en":"System"},"why":{"zh":"首个能生成一分钟高保真视频的模型，引发影视业广泛反应","en":"First minute-long high-fidelity video model"},"aiImpact":{"level":"leap","zh":"飞跃性突破","en":"Leap","reason":{"zh":"首个能生成一分钟高保真视频的模型，引发影视业广泛反应","en":"First minute-long high-fidelity video model"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"首个能生成一分钟高保真视频的模型，引发影视业广泛反应","en":"First minute-long high-fidelity video model"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2024-10-17","title":{"zh":"Movie Gen：Meta 媒体基础模型，生成带同步音频的高清视频（arXiv 2410.13720）","en":"Movie Gen: Meta’s media foundation models with synced audio"},"url":"https://arxiv.org/abs/2410.13720","source":"arXiv","flags":["preprint"],"flagLabels":[{"zh":"预印本，未同行评审","en":"Preprint, not peer-reviewed"}],"linkCheck":"ok-crossref","datePrecision":"day","type":"system","typeLabel":{"zh":"系统发布","en":"System"},"why":{"zh":"视频与音效一体生成","en":"Joint video and audio generation"},"aiImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"视频与音效一体生成","en":"Joint video and audio generation"}},"fieldImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"视频与音效一体生成","en":"Joint video and audio generation"}},"impactNotes":[],"verified":{"id":"arXiv:2410.13720","title":"Movie Gen: A Cast of Media Foundation Models","venue":"arXiv","date":"2024-10-17"},"addedIn":"v2"},{"date":"2025-07-18","title":{"zh":"Netflix 首次在原创剧《永恒者》(El Eternauta) 中使用生成式 AI 制作最终画面的建筑坍塌特效，速度快 10 倍","en":"Netflix uses generative AI for final on-screen VFX for the first time (El Eternauta)"},"url":"https://www.reuters.com/business/media-telecom/netflix-says-it-used-genai-argentine-tv-series-2025-07-18/","source":"Reuters","flags":[],"flagLabels":[],"linkCheck":"site-blocks-bots","datePrecision":"day","type":"deployment","typeLabel":{"zh":"落地部署","en":"Deployment"},"why":{"zh":"主流流媒体首次把生成式 AI 镜头用于成片","en":"First generative-AI final footage in a major streamer original"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"主流流媒体首次把生成式 AI 镜头用于成片","en":"First generative-AI final footage in a major streamer original"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"主流流媒体首次把生成式 AI 镜头用于成片","en":"First generative-AI final footage in a major streamer original"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-05-01","title":{"zh":"奥斯卡：AI 演员与 AI 编剧不具参评资格","en":"Oscars: AI actors and writers ineligible"},"url":"https://www.reuters.com/lifestyle/ai-actors-writers-will-be-ineligible-oscars-2026-05-01/","source":"Reuters","flags":[],"flagLabels":[],"linkCheck":"site-blocks-bots","datePrecision":"day","type":"policy","typeLabel":{"zh":"政策","en":"Policy"},"why":{"zh":"奥斯卡明确 AI 演员和 AI 编剧不具参评资格","en":"Oscars bar AI actors and writers"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"奥斯卡明确 AI 演员和 AI 编剧不具参评资格","en":"Oscars bar AI actors and writers"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"奥斯卡明确 AI 演员和 AI 编剧不具参评资格","en":"Oscars bar AI actors and writers"}},"impactNotes":[],"verified":null,"addedIn":"v1"},{"date":"2026-10-06","title":{"zh":"首部全 AI 生成长片《Gods Don’t Give Gifts》发布预告，计划 12 月院线上映并冲击奥斯卡","en":"Fully AI-generated feature “Gods Don’t Give Gifts” releases trailer; theatrical release and Oscar bid planned"},"url":"https://variety.com/2026/film/news/ai-film-gods-dont-give-gifts-sets-trailer-plans-oscar-bid-1236897843/","source":"Variety","flags":[],"flagLabels":[],"linkCheck":"ok","datePrecision":"day","type":"industry","typeLabel":{"zh":"产业/采用","en":"Industry"},"why":{"zh":"AI 长片进入院线发行渠道（尚未上映）","en":"AI feature enters theatrical distribution (not yet released)"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI 长片进入院线发行渠道（尚未上映）","en":"AI feature enters theatrical distribution (not yet released)"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI 长片进入院线发行渠道（尚未上映）","en":"AI feature enters theatrical distribution (not yet released)"}},"impactNotes":[],"verified":null,"addedIn":"v2"},{"date":"2026-10-10","title":{"zh":"Netflix 亚太制作负责人在釜山电影节透露：约 300 部作品已在预演、美术、剪辑或视效中使用生成式 AI；导演姜允成称 AI 流程可省 30%–40% 预算","en":"Netflix says generative AI has been used on about 300 titles; director Kang Yoon-sung cites 30–40% budget savings"},"url":"https://variety.com/2026/tv/festivals/netflix-ai-300-titles-kang-yoon-sung-40-percent-budget-busan-1236907359/","source":"Variety","flags":["vendor"],"flagLabels":[{"zh":"厂商自述","en":"Vendor claim"}],"linkCheck":"ok","datePrecision":"day","type":"industry","typeLabel":{"zh":"产业/采用","en":"Industry"},"why":{"zh":"AI：生成式 AI 已进入流媒体主要制作环节；学科：首次由 Netflix 给出采用规模（约 300 部），省钱比例为导演个人估计，未经独立核实","en":"AI: generative AI now in mainstream streaming production. Field: Netflix’s first adoption figure (~300 titles); savings are a director’s estimate"},"aiImpact":{"level":"incremental","zh":"增量进展","en":"Incremental","reason":{"zh":"AI：生成式 AI 已进入流媒体主要制作环节；学科：首次由 Netflix 给出采用规模（约 300 部），省钱比例为导演个人估计，未经独立核实","en":"AI: generative AI now in mainstream streaming production. Field: Netflix’s first adoption figure (~300 titles); savings are a director’s estimate"}},"fieldImpact":{"level":"major","zh":"重要进展","en":"Major","reason":{"zh":"AI：生成式 AI 已进入流媒体主要制作环节；学科：首次由 Netflix 给出采用规模（约 300 部），省钱比例为导演个人估计，未经独立核实","en":"AI: generative AI now in mainstream streaming production. Field: Netflix’s first adoption figure (~300 titles); savings are a director’s estimate"}},"impactNotes":[],"verified":null,"addedIn":"v2-20261022pm"}],"papersRepos":[{"title":"Stable Video Diffusion: Scaling Latent Video Diffusion Models to Large Datasets","url":"https://arxiv.org/abs/2311.15127","date":"2023-11-25","kind":"paper","linkCheck":"ok-crossref"},{"title":"Movie Gen: A Cast of Media Foundation Models","url":"https://arxiv.org/abs/2410.13720","date":"2024-10-17","kind":"paper","linkCheck":"ok-crossref"}],"orgsTools":[{"name":"OpenAI Sora","url":"https://openai.com/sora"},{"name":"Runway","url":"https://runwayml.com"},{"name":"Academy of Motion Picture Arts and Sciences","url":"https://www.oscars.org"}],"news":[{"id":"83360bdcbabd","date":"2026-10-20","title":"奥斯卡大师加持，「电影精神」怎么造就AI剧场？","url":"https://www.leiphone.com/category/industrynews/6Vycsmo3EhndKZSe.html","source":"雷峰网","lang":"zh","kind":"news","summaryZh":"雷峰网于2026-10-20发布的文章，涉及AI 创意媒体，原题《奥斯卡大师加持，「电影精神」怎么造就AI剧场？》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"e4563b326740","date":"2026-10-20","title":"做了10年影视制片人，现在我每天陪老板搓“AI玩具”","url":"https://www.tmtpost.com/8174207.html","source":"钛媒体","lang":"zh","kind":"news","summaryZh":"钛媒体于2026-10-20发布的文章，涉及AI 创意媒体，原题《做了10年影视制片人，现在我每天陪老板搓“AI玩具”》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"05912fe97c3c","date":"2026-10-20","title":"AI短剧里的“彩票工厂”不再赚钱","url":"https://www.tmtpost.com/8174771.html","source":"钛媒体","lang":"zh","kind":"news","summaryZh":"钛媒体于2026-10-20发布的文章，涉及AI 综合，原题《AI短剧里的“彩票工厂”不再赚钱》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"1923d0a0e40d","date":"2026-10-19","title":"Why Film Studios Are Trading Sound Stages for Data Centers","url":"https://nofilmschool.com/sound-stages-data-centers","source":"No Film School","lang":"en","kind":"news","summaryZh":"No Film School于2026-10-20发布的文章，涉及AI 创意媒体、AI 金融、AI 芯片硬件，原题《Why Film Studios Are Trading Sound Stages for Data Centers》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"12c1bf2f2e63","date":"2026-10-13","title":"‘Artificial’ is not the AI film we deserve","url":"https://www.transformernews.ai/p/artificial-review-not-the-ai-film-we-deserve","source":"Transformer (AI 政策)","lang":"en","kind":"news","summaryZh":"Transformer (AI 政策)于2026-10-13发布的文章，涉及AI 法律政策、AI 创意媒体，原题《‘Artificial’ is not the AI film we deserve》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"d7ded6c33ed8","date":"2026-10-12","title":"An overlooked sci-fi movie shows how disruption by AI could be positive for society","url":"https://theconversation.com/an-overlooked-sci-fi-movie-shows-how-disruption-by-ai-could-be-positive-for-society-293636","source":"The Conversation · AI","lang":"en","kind":"news","summaryZh":"The Conversation · AI于2026-10-12发布的文章，涉及AI 科研与调研、AI 创意媒体，原题《An overlooked sci-fi movie shows how disruption by AI could be positive for society》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"8bfbfc9f8068","date":"2026-10-10","title":"Prime Video Takes Global on Turkish AI-Generated Movie ‘Çukur:Vartolu,’ a Spin-Off From the Hit Series Known Internationally as ‘The Pit’ (EXCLUSIVE)","url":"https://variety.com/2026/film/global/prime-video-turkish-ai-generated-movie-cukurvartolu-the-pit-1236907346/","source":"Variety · AI","lang":"en","kind":"news","summaryZh":"Variety · AI于2026-10-10发布的文章，涉及AI 创意媒体，原题《Prime Video Takes Global on Turkish AI-Generated Movie ‘Çukur:Vartolu,’ a Spin-Off From th》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"55ef3851482a","date":"2026-10-09","title":"全球最大独立 AI 原生影视公司，要打造「AI 版 Disney」","url":"https://www.ifanr.com/1683309?utm_source=rss&utm_medium=rss&utm_campaign=","source":"爱范儿","lang":"zh","kind":"news","summaryZh":"爱范儿报道一家 AI 原生影视公司，计划 2027 年暑期档把 AI 长片送进院线。","via":"research-samples-20261010","linkCheck":"from-feed"},{"id":"2f163e720de9","date":"2026-10-06","title":"AI film Gods Don’t Give Gifts sets trailer, plans Oscar bid","url":"https://variety.com/2026/film/news/ai-film-gods-dont-give-gifts-sets-trailer-plans-oscar-bid-1236897843/","source":"Variety","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"ok"},{"id":"c29d1dc3deae","date":"2026-10-05","title":"Iberseries Takeaways: Woody Allen on His New Film, Los Javis Pact With HBO Max, the Tales of Cachou the Brown Bear and More","url":"https://variety.com/2026/tv/global/iberseries-takes-woody-allen-los-javis-and-hbo-max-and-more-1236899264/","source":"Variety · AI","lang":"en","kind":"news","summaryZh":"Variety · AI于2026-10-06发布的文章，涉及AI 创意媒体，原题《Iberseries Takeaways: Woody Allen on His New Film, Los Javis Pact With HBO Max, the Tales 》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"08a791fe4148","date":"2026-10-02","title":"What Mattered to Hollywood at This Year’s Iberseries & Platino Industria","url":"https://www.hollywoodreporter.com/business/business-news/woody-allen-madrid-film-iberseries-platino-industria-wrap-1236714367/","source":"Hollywood Reporter · AI","lang":"en","kind":"news","summaryZh":"Hollywood Reporter · AI于2026-10-02发布的文章，涉及AI 创意媒体，原题《What Mattered to Hollywood at This Year’s Iberseries & Platino Industria》，详见原文链接。","via":"collector-run-20261021","linkCheck":"from-feed"},{"id":"a58e3717fc7f","date":"2026-05-01","title":"AI actors and writers will be ineligible for Oscars","url":"https://www.reuters.com/lifestyle/ai-actors-writers-will-be-ineligible-oscars-2026-05-01/","source":"Reuters","lang":"en","kind":"news","summaryZh":null,"via":"manual-websearch-2026102x","linkCheck":"site-blocks-bots"}],"review":{"lastReviewed":"2026-10-22","cycleDays":90,"nextReview":"2027-01-19"},"urls":{"zh":"/zh-hans/disciplines/film-animation/","en":"/en/disciplines/film-animation/"},"dimensions":{"method":{"score":1,"label":{"zh":"研究方法","en":"Research method"},"evidence":"电影研究中 AI 辅助有限","source":null,"weight":0.3},"discovery":{"score":2,"label":{"zh":"科学发现","en":"Scientific discovery"},"evidence":"文生视频进入制作流程；首部 AI 生成长片计划院线上映","source":"https://variety.com/2026/film/news/ai-film-gods-dont-give-gifts-sets-trailer-plans-oscar-bid-1236897843/","weight":0.35},"adoption":{"score":2,"label":{"zh":"从业者采用","en":"Practitioner adoption"},"evidence":"制作流程广泛试用，奥斯卡规则允许使用 AI 工具","source":"https://www.reuters.com/lifestyle/ai-actors-writers-will-be-ineligible-oscars-2026-05-01/","weight":0.25},"education":{"score":1,"label":{"zh":"教育/课程","en":"Education & curriculum"},"evidence":"未找到课程证据","source":null,"weight":0.1}},"compositeScore":1.6,"levelCapped":false,"levelChange":{"from":2,"to":2,"direction":"same","reason":null},"biggestBreakthrough":{"date":"2024-02-15","title":{"zh":"OpenAI 发布文生视频模型 Sora","en":"OpenAI unveils text-to-video model Sora"},"url":"https://openai.com/index/sora/","aiImpact":"leap","fieldImpact":"major","why":{"zh":"首个能生成一分钟高保真视频的模型，引发影视业广泛反应","en":"First minute-long high-fidelity video model"}},"keyPapers":[{"title":"Stable Video Diffusion: Scaling Latent Video Diffusion Models to Large Datasets","titleZh":"Stable Video Diffusion：开源视频生成模型（arXiv 2311.15127）","id":"arXiv:2311.15127","url":"https://arxiv.org/abs/2311.15127","venue":"arXiv","date":"2023-11-25","why":{"zh":"开源视频生成进入可用阶段","en":"Open video generation becomes usable"},"fieldImpact":"incremental","aiImpact":"major"},{"title":"Movie Gen: A Cast of Media Foundation Models","titleZh":"Movie Gen：Meta 媒体基础模型，生成带同步音频的高清视频（arXiv 2410.13720）","id":"arXiv:2410.13720","url":"https://arxiv.org/abs/2410.13720","venue":"arXiv","date":"2024-10-17","why":{"zh":"视频与音效一体生成","en":"Joint video and audio generation"},"fieldImpact":"incremental","aiImpact":"major"}],"latest_papers":[{"id":"arXiv:2610.30129","title":"Rethinking Video Autoencoders for Interactive World Models","titleZh":"为交互式世界模型重新设计视频自编码器","summaryZh":"指出世界模型沿用文生视频的自编码器并不合适，重新考虑压缩率与动作响应延迟的权衡。","summaryEn":"Rethinks video autoencoders for interactive world models, trading compression against action latency.","authors":["Adrien Ramanana Rahary","Yutong Bai","Nicolas Dufour","Patrick Pérez"],"date":"2026-10-21","url":"https://arxiv.org/abs/2610.30129","site":"arXiv","venue":null,"category":"cs.CV","fetchedAt":"2026-10-22T10:10:55"},{"id":"arXiv:2610.29915","title":"CausalAlign: Causal Alignment of Video Generation Models for Consistent Ego-Motion","titleZh":"CausalAlign：对齐视频生成模型以保持相机自运动一致","summaryZh":"诊断视频生成模型内部如何表示和保留相机运动，并提出因果对齐方法减少长时段镜头漂移。","summaryEn":"A causal alignment framework that reduces long-horizon camera ego-motion drift in video generation models.","authors":["Kai Zhu","Yintianrun Zhang","Shuwei Jiang","Ruolin Zuo"],"date":"2026-10-21","url":"https://arxiv.org/abs/2610.29915","site":"arXiv","venue":null,"category":"cs.CV","fetchedAt":"2026-10-22T10:10:55"},{"id":"arXiv:2610.28632","title":"Joints, Camera, Action! Co-Generating 3D Motion and Video for Arbitrary Rigs","titleZh":"关节、镜头、开拍！为任意骨骼同时生成三维动作与视频","summaryZh":"在预训练视频模型旁接一个支持任意骨骼的三维动作模型，联合生成视频、可回放的三维动作和对齐场景的相机。","summaryEn":"Co-generates video, replayable 3D motion for arbitrary rigs, and a registered camera.","authors":["Noam Rotstein","Yoad Tewel","Chen Tessler","Haggai Maron"],"date":"2026-10-20","url":"https://arxiv.org/abs/2610.28632","site":"arXiv","venue":null,"category":"cs.CV","fetchedAt":"2026-10-22T10:10:55"},{"id":"arXiv:2610.28631","title":"AEON: Unifying Video and 3D World Models","titleZh":"AEON：统一视频与三维世界模型","summaryZh":"在视频潜空间中生成可渲染的四维高斯（4DGS）作为显式世界表示，兼顾视频模型的真实感与三维模型的几何一致性。","summaryEn":"Generates renderable 4D Gaussians from video latents, unifying video and 3D world models.","authors":["Weiqi Zhang","Wenyuan Zhang","Yu-Shen Liu","Junsheng Zhou"],"date":"2026-10-20","url":"https://arxiv.org/abs/2610.28631","site":"arXiv","venue":null,"category":"cs.CV","fetchedAt":"2026-10-22T10:10:55"},{"id":"arXiv:2610.26853","title":"InfiniDream: Open-Ended Photorealistic 3D Worlds Through Procedural Generation","titleZh":"InfiniDream：通过程序化生成构建开放式逼真三维世界","summaryZh":"用程序化生成创建可控、可编辑、多样的逼真三维世界，用于合成数据和游戏开发。","summaryEn":"Procedural generation of open-ended photorealistic 3D worlds.","authors":["Yi-Chuan Huang","Hao-Jen Chien","Zheng-Hui Huang","Yu-Lun Liu"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.26853","site":"arXiv","venue":null,"category":"cs.CV","fetchedAt":"2026-10-22T10:10:55"},{"id":"arXiv:2610.26851","title":"Manga2Animate: An Agentic Framework for Faithful Manga Animation with Script and Storyboard Planning","titleZh":"Manga2Animate：通过剧本与分镜规划忠实改编漫画动画的智能体框架","summaryZh":"多智能体把漫画先转为动画剧本和视觉分镜，再生成视频，以保持角色、画风和故事线。","summaryEn":"Multi-agent manga-to-animation via script and storyboard planning.","authors":["Sheng-Wei Zhang","Hsi-An Chen","Yan-Ting Liou","Chih-Hsuan Chen"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.26851","site":"arXiv","venue":null,"category":"cs.CV","fetchedAt":"2026-10-22T10:10:55"},{"id":"arXiv:2610.26801","title":"Mise-en-Video: Fine-Grained 3D Control for Multi-Shot Video Generation","titleZh":"Mise-en-Video：多镜头视频生成的细粒度三维控制","summaryZh":"在多镜头叙事视频生成中同时实现镜头间一致性和细粒度三维控制。","summaryEn":"Fine-grained 3D control for consistent multi-shot video generation.","authors":["Yingshuang Zou","Xiaoyu Shi","Yawen Luo","Ziang Li"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.26801","site":"arXiv","venue":null,"category":"cs.CV","fetchedAt":"2026-10-22T10:10:55"},{"id":"arXiv:2610.26711","title":"PVBench: Quantifying Physical Plausibility in Video Generation with Object-Level Physical Measurements","titleZh":"PVBench：用物体级物理测量量化视频生成的物理合理性","summaryZh":"从 7 个领域 46 个维度定量评估生成视频是否符合几何、运动学、动力学和守恒定律。","summaryEn":"Measures physical plausibility of generated videos across 46 dimensions.","authors":["Kaiwei Chen","Moran Li","Zhenye Gan","Yabiao Wang"],"date":"2026-10-19","url":"https://arxiv.org/abs/2610.26711","site":"arXiv","venue":null,"category":"cs.CV","fetchedAt":"2026-10-22T10:10:55"}]}],"feed":[{"type":"milestone","discipline":"life-sciences","date":"2026-10-21","title":{"zh":"西湖大学等发布 Pinal：160 亿参数模型直接从自然语言功能描述设计蛋白质，荧光蛋白、PET 水解酶、醇脱氢酶和 H 蛋白四类设计均经实验验证（Nature）","en":"Pinal: a 16B-parameter foundation model that designs functional proteins from natural-language descriptions, validated in four experimental tasks (Nature)"},"url":"https://www.nature.com/articles/s41586-026-11205-z","source":"Nature","flags":[]},{"type":"milestone","discipline":"genomics","date":"2026-10-21","title":{"zh":"hEVO：在 6.38 万人、47.8 亿条单倍型序列上训练的人类基因组基础模型，19 项主要对比中 18 项排名第一（bioRxiv）","en":"hEVO: a human genomic foundation model trained on 4.784B haplotype sequences from 63,809 genomes, ranking first in 18 of 19 primary comparisons (bioRxiv)"},"url":"https://www.biorxiv.org/content/10.64898/2026.10.19.760276v1","source":"bioRxiv","flags":["preprint"]},{"type":"milestone","discipline":"literature","date":"2026-10-21","title":{"zh":"法国设立 Prix Pascal 人工智能文学奖：征集与大模型合写的法语小说稿，须申报所用模型与提示词，获奖作品出版","en":"Prix Pascal de Littérature IA launches for French-language fiction co-written with AI"},"url":"https://www.theguardian.com/books/2026/oct/21/new-prize-launches-books-co-written-with-ai-france","source":"The Guardian","flags":[]},{"type":"milestone","discipline":"mathematics","date":"2026-10-20","title":{"zh":"人类数学家借助 AI 证明“双帽猜想”（球面上无正交点对的集合测度上限），主结论另有几乎完全由 AI 完成形式化的 Lean 4 证明","en":"AI-assisted proof of the double-cap conjecture, with a separate Lean 4 proof formalised almost entirely using AI tools (arXiv 2610.27523)"},"url":"https://arxiv.org/abs/2610.27523","source":"arXiv","flags":["preprint"]},{"type":"news","discipline":"mathematics","date":"2026-10-20","title":{"zh":"arXiv math.HO 数学史与综述于2026-10-20发布的论文，涉及AI 数学，原题《The crisis of AI-generated mathematics》，详见原文链接。","en":"The crisis of AI-generated mathematics"},"url":"https://arxiv.org/abs/2608.02859","source":"arXiv math.HO 数学史与综述","flags":[]},{"type":"news","discipline":"mathematics","date":"2026-10-20","title":{"zh":"arXiv cs.LO 逻辑与形式化于2026-10-20发布的论文，涉及AI 数学，原题《Studying proof data with proof theory: the case of AI-generated mathematics》，详见原文链接。","en":"Studying proof data with proof theory: the case of AI-generated mathematics"},"url":"https://arxiv.org/abs/2610.25357","source":"arXiv cs.LO 逻辑与形式化","flags":[]},{"type":"news","discipline":"mathematics","date":"2026-10-20","title":{"zh":"arXiv cs.LO 逻辑与形式化于2026-10-20发布的论文，涉及AI 数学、AI 科研与调研、AI 商业融资，原题《Formalization of R\\'{e}nyi Differential Privacy in Isabelle/HOL (Draft version)》，详见原文链接。","en":"Formalization of R\\'{e}nyi Differential Privacy in Isabelle/HOL (Draft version)"},"url":"https://arxiv.org/abs/2610.25636","source":"arXiv cs.LO 逻辑与形式化","flags":[]},{"type":"news","discipline":"statistics","date":"2026-10-20","title":{"zh":"Nature 学科·数学于2026-10-20发布的文章，涉及AI 数学、AI 气候能源、AI 软件开发，原题《A robust deep learning framework for smart grid stability classification using statistical》，详见原文链接。","en":"A robust deep learning framework for smart grid stability classification using statistical validation"},"url":"https://www.nature.com/articles/s41598-026-74799-4","source":"Nature 学科·数学","flags":[]},{"type":"news","discipline":"physics","date":"2026-10-20","title":{"zh":"HPCwire于2026-10-20发布的文章，涉及AI 芯片硬件，原题《NVIDIA Runs Record Fault-Tolerant Quantum Simulation at Jülich Supercomputing Centre》，详见原文链接。","en":"NVIDIA Runs Record Fault-Tolerant Quantum Simulation at Jülich Supercomputing Centre"},"url":"https://www.hpcwire.com/off-the-wire/nvidia-runs-record-fault-tolerant-quantum-simulation-at-julich-supercomputing-centre/","source":"HPCwire","flags":[]},{"type":"news","discipline":"physics","date":"2026-10-20","title":{"zh":"arXiv physics.chem-ph于2026-10-20发布的论文，涉及AI 化学材料、AI 气候能源、AI 软件开发，原题《Measurement-efficient energy estimator for quantum chemistry of photodynamic materials》，详见原文链接。","en":"Measurement-efficient energy estimator for quantum chemistry of photodynamic materials"},"url":"https://arxiv.org/abs/2610.26689","source":"arXiv physics.chem-ph","flags":[]},{"type":"news","discipline":"chemistry","date":"2026-10-20","title":{"zh":"TechCrunch AI于2026-10-20发布的文章，涉及AI 化学材料、AI 商业融资，原题《Otter.ai raises $320 million from General Catalyst’s growth fund for its meeting tool》，详见原文链接。","en":"Otter.ai raises $320 million from General Catalyst’s growth fund for its meeting tool"},"url":"https://techcrunch.com/2026/10/20/otter-ai-raises-320-million-from-general-catalysts-growth-fund-for-its-meeting-tool/","source":"TechCrunch AI","flags":[]},{"type":"news","discipline":"chemistry","date":"2026-10-20","title":{"zh":"arXiv q-bio.BM（生物分子）于2026-10-20发布的论文，涉及AI 生物/医药，原题《Aptamers as a sandbox for lab-in-the-loop agentic molecular discovery》，详见原文链接。","en":"Aptamers as a sandbox for lab-in-the-loop agentic molecular discovery"},"url":"https://arxiv.org/abs/2610.24806","source":"arXiv q-bio.BM（生物分子）","flags":[]},{"type":"news","discipline":"chemistry","date":"2026-10-20","title":{"zh":"arXiv cs.AI于2026-10-20发布的论文，涉及AI 科研与调研、AI 教育、AI 化学材料，原题《Distillation, Self-Training, and the Fragility of Learned Chemical Reasoning》，详见原文链接。","en":"Distillation, Self-Training, and the Fragility of Learned Chemical Reasoning"},"url":"https://arxiv.org/abs/2610.24199","source":"arXiv cs.AI","flags":[]},{"type":"news","discipline":"materials","date":"2026-10-20","title":{"zh":"36氪于2026-10-20发布的文章，涉及AI 化学材料、AI 商业融资、AI 科研与调研，原题《硬氪首发｜用AI智能体辅助新材料研发，睿质科技完成数千万元天使轮融资》，详见原文链接。","en":"硬氪首发｜用AI智能体辅助新材料研发，睿质科技完成数千万元天使轮融资"},"url":"https://36kr.com/p/4033523733042053?f=rss","source":"36氪","flags":[]},{"type":"news","discipline":"materials","date":"2026-10-20","title":{"zh":"arXiv cond-mat.mtrl-sci于2026-10-20发布的论文，涉及AI 化学材料，原题《Multi-property Feynman-Kac Steering of Crystal Diffusion Models》，详见原文链接。","en":"Multi-property Feynman-Kac Steering of Crystal Diffusion Models"},"url":"https://arxiv.org/abs/2610.24828","source":"arXiv cond-mat.mtrl-sci","flags":[]},{"type":"news","discipline":"materials","date":"2026-10-20","title":{"zh":"arXiv cond-mat.mtrl-sci于2026-10-20发布的论文，涉及AI 化学材料，原题《Crystal Structure Based Graph Neural Network And LLM Multi-Modal Approach for Material Mag》，详见原文链接。","en":"Crystal Structure Based Graph Neural Network And LLM Multi-Modal Approach for Material Magnetic Ordering Prediction"},"url":"https://arxiv.org/abs/2610.25775","source":"arXiv cond-mat.mtrl-sci","flags":[]},{"type":"news","discipline":"life-sciences","date":"2026-10-20","title":{"zh":"Phys.org 化学·材料(纳米)于2026-10-20发布的文章，涉及AI 化学材料、AI 生物/医药、AI 教育，原题《Nanoscale tech tracks proteins inside living neurons to guide Alzheimer's and Parkinson's 》，详见原文链接。","en":"Nanoscale tech tracks proteins inside living neurons to guide Alzheimer's and Parkinson's drug discovery"},"url":"https://phys.org/news/2026-10-nanoscale-tech-tracks-proteins-neurons.html","source":"Phys.org 化学·材料(纳米)","flags":[]},{"type":"news","discipline":"life-sciences","date":"2026-10-20","title":{"zh":"arXiv q-bio于2026-10-20发布的论文，涉及AI 生物/医药、AI 商业融资、AI 软件开发，原题《CADS: Measuring the Incremental Contribution of LLM Proposers in Sequential Biological Scr》，详见原文链接。","en":"CADS: Measuring the Incremental Contribution of LLM Proposers in Sequential Biological Screens"},"url":"https://arxiv.org/abs/2610.24597","source":"arXiv q-bio","flags":[]},{"type":"news","discipline":"life-sciences","date":"2026-10-20","title":{"zh":"arXiv q-bio于2026-10-20发布的论文，涉及AI 生物/医药、AI 气候能源，原题《Disentanglement with Biological Constraints: A Theory of Functional Cell Types》，详见原文链接。","en":"Disentanglement with Biological Constraints: A Theory of Functional Cell Types"},"url":"https://arxiv.org/abs/2210.01768","source":"arXiv q-bio","flags":[]},{"type":"news","discipline":"genomics","date":"2026-10-20","title":{"zh":"Nature 学科·机器学习于2026-10-20发布的文章，涉及AI 科研与调研、AI 生物/医药、AI 安全，原题《High-risk antimicrobial resistance genomic traits in multidrug-resistant Salmonella across》，详见原文链接。","en":"High-risk antimicrobial resistance genomic traits in multidrug-resistant Salmonella across China’s food chain: projections to 2050"},"url":"https://www.nature.com/articles/s41467-026-78346-7","source":"Nature 学科·机器学习","flags":[]},{"type":"news","discipline":"genomics","date":"2026-10-20","title":{"zh":"arXiv q-bio.BM（生物分子）于2026-10-20发布的论文，涉及AI 生物/医药、AI 安全、AI 软件开发，原题《CPTCs Drive Somatic-Visceral Communication via the Wnt Axis in Somatic Mechanotherapy: A S》，详见原文链接。","en":"CPTCs Drive Somatic-Visceral Communication via the Wnt Axis in Somatic Mechanotherapy: A Single-Cell Deep Learning Study"},"url":"https://arxiv.org/abs/2602.05451","source":"arXiv q-bio.BM（生物分子）","flags":[]},{"type":"news","discipline":"genomics","date":"2026-10-20","title":{"zh":"arXiv q-bio.BM（生物分子）于2026-10-20发布的论文，涉及AI 生物/医药、AI 软件开发，原题《RNAGenScape: Property-Guided, Optimized Generation of mRNA Sequences with Manifold Langevi》，详见原文链接。","en":"RNAGenScape: Property-Guided, Optimized Generation of mRNA Sequences with Manifold Langevin Dynamics"},"url":"https://arxiv.org/abs/2510.24736","source":"arXiv q-bio.BM（生物分子）","flags":[]},{"type":"news","discipline":"neuroscience","date":"2026-10-20","title":{"zh":"Nature 学科·计算生物于2026-10-20发布的文章，涉及AI 生物/医药，原题《EEG-based Alzheimer’s disease detection using two-branch multi-scale deep learning with in》，详见原文链接。","en":"EEG-based Alzheimer’s disease detection using two-branch multi-scale deep learning with integrated gradients-guided channel selection"},"url":"https://www.nature.com/articles/s41598-026-76382-3","source":"Nature 学科·计算生物","flags":[]},{"type":"news","discipline":"neuroscience","date":"2026-10-20","title":{"zh":"Phys.org 化学·材料(纳米)于2026-10-20发布的文章，涉及AI 化学材料、AI 生物/医药、AI 教育，原题《Nanoscale tech tracks proteins inside living neurons to guide Alzheimer's and Parkinson's 》，详见原文链接。","en":"Nanoscale tech tracks proteins inside living neurons to guide Alzheimer's and Parkinson's drug discovery"},"url":"https://phys.org/news/2026-10-nanoscale-tech-tracks-proteins-neurons.html","source":"Phys.org 化学·材料(纳米)","flags":[]},{"type":"news","discipline":"neuroscience","date":"2026-10-20","title":{"zh":"36氪于2026-10-20发布的文章，涉及AI 商业融资、AI 生物/医药、AI 科研与调研，原题《顺为投了一位北大PI，想用AI“解密”人脑》，详见原文链接。","en":"顺为投了一位北大PI，想用AI“解密”人脑"},"url":"https://36kr.com/p/4032248929686151?f=rss","source":"36氪","flags":[]},{"type":"milestone","discipline":"pharmacy","date":"2026-10-20","title":{"zh":"Radical Numerics：Omnii-1 设计的噬菌体在同情用药途径下首次用于耐药大肠杆菌感染患者","en":"Radical Numerics: Omnii-1-designed phages administered to a patient under compassionate use"},"url":"https://www.radicalnumerics.ai/blog/omnii-phage","source":"Radical Numerics","flags":["vendor"]},{"type":"news","discipline":"pharmacy","date":"2026-10-20","title":{"zh":"pharmaphorum于2026-10-20发布的文章，涉及AI 生物/医药、AI 医疗，原题《AZ and Oracle team up on AI platform for clinical trials》，详见原文链接。","en":"AZ and Oracle team up on AI platform for clinical trials"},"url":"https://pharmaphorum.com/news/az-and-oracle-team-ai-platform-clinical-trials","source":"pharmaphorum","flags":[]},{"type":"news","discipline":"pharmacy","date":"2026-10-20","title":{"zh":"STAT News于2026-10-20发布的文章，涉及AI 生物/医药、AI 医疗、AI 商业融资，原题《STAT+: Pharmalittle: We’re reading about drug launch prices, an AstraZeneca AI deal, and m》，详见原文链接。","en":"STAT+: Pharmalittle: We’re reading about drug launch prices, an AstraZeneca AI deal, and more"},"url":"https://www.statnews.com/pharmalot/2026/10/20/drug-launch-prices-rose-astrazeneca-ai-deal/?utm_campaign=rss","source":"STAT News","flags":[]},{"type":"news","discipline":"pharmacy","date":"2026-10-20","title":{"zh":"Bloomberg Technology于2026-10-20发布的文章，涉及AI 金融、AI 生物/医药，原题《Citrini Bets on Life Sciences After Claude AI Drug Study》，详见原文链接。","en":"Citrini Bets on Life Sciences After Claude AI Drug Study"},"url":"https://www.bloomberg.com/news/articles/2026-10-20/citrini-bets-on-life-sciences-stocks-after-claude-ai-drug-study","source":"Bloomberg Technology","flags":[]},{"type":"news","discipline":"clinical-medicine","date":"2026-10-20","title":{"zh":"Nature 学科·数学于2026-10-20发布的文章，涉及AI 数学、AI 医疗，原题《Liquid neural network-driven intelligent topology perception and path anomaly diagnosis fo》，详见原文链接。","en":"Liquid neural network-driven intelligent topology perception and path anomaly diagnosis for Ipv6 smart grids"},"url":"https://www.nature.com/articles/s41598-026-76378-z","source":"Nature 学科·数学","flags":[]},{"type":"news","discipline":"clinical-medicine","date":"2026-10-20","title":{"zh":"Nature 学科·计算生物于2026-10-20发布的文章，涉及AI 生物/医药、AI 医疗，原题《Expertise modulates automation bias and sentinel behavior in human–AI collaborative diagno》，详见原文链接。","en":"Expertise modulates automation bias and sentinel behavior in human–AI collaborative diagnosis of neonatal pneumoperitoneum"},"url":"https://www.nature.com/articles/s41746-026-03399-3","source":"Nature 学科·计算生物","flags":[]},{"type":"news","discipline":"clinical-medicine","date":"2026-10-20","title":{"zh":"pharmaphorum于2026-10-20发布的文章，涉及AI 生物/医药、AI 医疗，原题《AZ and Oracle team up on AI platform for clinical trials》，详见原文链接。","en":"AZ and Oracle team up on AI platform for clinical trials"},"url":"https://pharmaphorum.com/news/az-and-oracle-team-ai-platform-clinical-trials","source":"pharmaphorum","flags":[]},{"type":"milestone","discipline":"medical-imaging","date":"2026-10-20","title":{"zh":"Sophont 发布 Holon-S1：从常规 H&E 切片预测空间基因表达与通路活性的泛癌病理基础模型（仅限科研）","en":"Sophont launches Holon-S1, a pan-cancer pathology model predicting spatial gene expression from H&E (research use)"},"url":"https://www.prnewswire.com/news-releases/sophont-launches-holon-s1-to-reveal-tumor-biology-from-routine-pathology-slides-302914452.html","source":"PR Newswire","flags":["vendor"]},{"type":"news","discipline":"medical-imaging","date":"2026-10-20","title":{"zh":"arXiv q-bio于2026-10-20发布的论文，涉及AI 生物/医药、AI 医疗、AI 软件开发，原题《GliomaINK: A Tumor and Anatomy Conditioned Flow Matching Framework for 3D MRI Sequence Syn》，详见原文链接。","en":"GliomaINK: A Tumor and Anatomy Conditioned Flow Matching Framework for 3D MRI Sequence Synthesis in Glioma Patients"},"url":"https://arxiv.org/abs/2610.24044","source":"arXiv q-bio","flags":[]},{"type":"news","discipline":"medical-imaging","date":"2026-10-20","title":{"zh":"arXiv q-bio于2026-10-20发布的论文，涉及AI 生物/医药、AI 医疗、AI 软件开发，原题《What Language Adds, and at What Cost: Four Controlled Probes of Pathology Vision-Language 》，详见原文链接。","en":"What Language Adds, and at What Cost: Four Controlled Probes of Pathology Vision-Language Models"},"url":"https://arxiv.org/abs/2610.26805","source":"arXiv q-bio","flags":[]},{"type":"news","discipline":"medical-imaging","date":"2026-10-20","title":{"zh":"arXiv q-bio于2026-10-20发布的论文，涉及AI 生物/医药、AI 医疗，原题《WILSON - Multi-magnification composites as a foundation for patient-level pathology analys》，详见原文链接。","en":"WILSON - Multi-magnification composites as a foundation for patient-level pathology analysis and diagnostic text generation"},"url":"https://arxiv.org/abs/2609.25123","source":"arXiv q-bio","flags":[]},{"type":"news","discipline":"public-health","date":"2026-10-20","title":{"zh":"Nature 学科·机器学习于2026-10-20发布的文章，涉及AI 科研与调研、AI 生物/医药、AI 安全，原题《High-risk antimicrobial resistance genomic traits in multidrug-resistant Salmonella across》，详见原文链接。","en":"High-risk antimicrobial resistance genomic traits in multidrug-resistant Salmonella across China’s food chain: projections to 2050"},"url":"https://www.nature.com/articles/s41467-026-78346-7","source":"Nature 学科·机器学习","flags":[]},{"type":"milestone","discipline":"ecology","date":"2026-10-20","title":{"zh":"生物多样性基础模型 BioCLIP 2.5 迁移到 2727 种蛾类野外监测（Scientific Reports）","en":"Biodiversity foundation model transfers to field moth monitoring of 2,727 species (Sci. Rep.)"},"url":"https://doi.org/10.1038/s41598-026-75760-1","source":"Scientific Reports","flags":[]},{"type":"news","discipline":"ecology","date":"2026-10-20","title":{"zh":"Nature 学科·计算生物于2026-10-20发布的文章，涉及AI 生物/医药，原题《Bridging the deployment gap in automated moth monitoring through biodiversity foundation m》，详见原文链接。","en":"Bridging the deployment gap in automated moth monitoring through biodiversity foundation models"},"url":"https://www.nature.com/articles/s41598-026-75760-1","source":"Nature 学科·计算生物","flags":[]},{"type":"news","discipline":"computer-science","date":"2026-10-20","title":{"zh":"arXiv q-fin于2026-10-20发布的论文，涉及AI 金融、AI 安全，原题《TRACE: Governing AI-Assisted Research with Principles from Computer Science》，详见原文链接。","en":"TRACE: Governing AI-Assisted Research with Principles from Computer Science"},"url":"https://arxiv.org/abs/2610.24035","source":"arXiv q-fin","flags":[]},{"type":"news","discipline":"computer-science","date":"2026-10-20","title":{"zh":"arXiv cs.CL于2026-10-20发布的论文，涉及AI 综合、AI 科研与调研、AI 软件开发，原题《Discuss More, Comment Less: Boosting Automated Research via Multi-Perspective Collaborativ》，详见原文链接。","en":"Discuss More, Comment Less: Boosting Automated Research via Multi-Perspective Collaborative Review"},"url":"https://arxiv.org/abs/2610.24050","source":"arXiv cs.CL","flags":[]},{"type":"news","discipline":"software-engineering","date":"2026-10-20","title":{"zh":"EE Times于2026-10-20发布的文章，涉及AI 芯片硬件、AI 软件开发，原题《Chris Lattner: AI Code Generation Doesn’t Solve Software Engineering》，详见原文链接。","en":"Chris Lattner: AI Code Generation Doesn’t Solve Software Engineering"},"url":"https://www.eetimes.com/chris-lattner-ai-code-generation-doesnt-solve-software-engineering/","source":"EE Times","flags":[]},{"type":"news","discipline":"software-engineering","date":"2026-10-20","title":{"zh":"hn-show-ai于2026-10-20发布的文章，涉及AI 软件开发，原题《Show HN: Code-mode – keep MCP tool results out of Claude Code's context》，详见原文链接。","en":"Show HN: Code-mode – keep MCP tool results out of Claude Code's context"},"url":"https://github.com/gabe4coding/claude-code-mode","source":"hn-show-ai","flags":[]},{"type":"news","discipline":"software-engineering","date":"2026-10-20","title":{"zh":"hn-show-ai于2026-10-20发布的文章，涉及AI 软件开发，原题《Show HN: Agent Archive – Session Handoffs and Backups for Claude Code and Codex》，详见原文链接。","en":"Show HN: Agent Archive – Session Handoffs and Backups for Claude Code and Codex"},"url":"https://github.com/wangjohn/agent-archive","source":"hn-show-ai","flags":[]},{"type":"news","discipline":"cybersecurity","date":"2026-10-20","title":{"zh":"Insurance Journal于2026-10-20发布的文章，涉及AI 金融、AI 安全，原题《AI Accelerates Cyber Threat Landscape, Poses New Risks for Insurers and Businesses》，详见原文链接。","en":"AI Accelerates Cyber Threat Landscape, Poses New Risks for Insurers and Businesses"},"url":"https://www.insurancejournal.com/news/national/2026/10/20/889665.htm","source":"Insurance Journal","flags":[]},{"type":"news","discipline":"cybersecurity","date":"2026-10-20","title":{"zh":"The Decoder于2026-10-20发布的文章，涉及AI 安全、AI 商业融资，原题《Army Cyber Command's ECHOTRIBBLE would turn AI agents into self-spreading cyberweapons》，详见原文链接。","en":"Army Cyber Command's ECHOTRIBBLE would turn AI agents into self-spreading cyberweapons"},"url":"https://the-decoder.com/army-cyber-commands-echotribble-would-turn-ai-agents-into-self-spreading-cyberweapons/","source":"The Decoder","flags":[]},{"type":"news","discipline":"cybersecurity","date":"2026-10-20","title":{"zh":"The Hacker News于2026-10-20发布的文章，涉及AI 安全，原题《Hackers Use Autonomous AI Agents to Breach Company Websites, Target Over 1,200 Sites》，详见原文链接。","en":"Hackers Use Autonomous AI Agents to Breach Company Websites, Target Over 1,200 Sites"},"url":"https://thehackernews.com/2026/10/hackers-use-autonomous-ai-agents-to.html","source":"The Hacker News","flags":[]},{"type":"news","discipline":"robotics","date":"2026-10-20","title":{"zh":"HIT Consultant于2026-10-20发布的文章，涉及AI 医疗、AI 教育、AI 机器人，原题《How Peristaltic Pumps Are Enabling the Next Generation of Robotic Surgery》，详见原文链接。","en":"How Peristaltic Pumps Are Enabling the Next Generation of Robotic Surgery"},"url":"https://hitconsultant.net/2026/10/20/how-peristaltic-pumps-enable-next-generation-robotic-surgery-watson-marlow-rodd-turnquist/","source":"HIT Consultant","flags":[]},{"type":"news","discipline":"robotics","date":"2026-10-20","title":{"zh":"TechXplore 机器人于2026-10-20发布的文章，涉及AI 机器人，原题《From wombat burrows to Noah's Ark: Robot joins search》，详见原文链接。","en":"From wombat burrows to Noah's Ark: Robot joins search"},"url":"https://techxplore.com/news/2026-10-wombat-burrows-noah-ark-robot.html","source":"TechXplore 机器人","flags":[]},{"type":"news","discipline":"robotics","date":"2026-10-20","title":{"zh":"雷峰网于2026-10-20发布的文章，涉及AI 机器人、AI 科研与调研、AI 化学材料，原题《启元机器人新进展：北大董豪团队推出 HIL-UMI，让后训练摆脱「真机依赖」》，详见原文链接。","en":"启元机器人新进展：北大董豪团队推出 HIL-UMI，让后训练摆脱「真机依赖」"},"url":"https://www.leiphone.com/category/ai/XgCxObsqu8pc44px.html","source":"雷峰网","flags":[]},{"type":"news","discipline":"energy-fusion","date":"2026-10-20","title":{"zh":"Tom's Hardware AI于2026-10-20发布的文章，涉及AI 芯片硬件、AI 气候能源，原题《US restarts three shut-down nuclear reactors to feed AI data centers》，详见原文链接。","en":"US restarts three shut-down nuclear reactors to feed AI data centers"},"url":"https://www.tomshardware.com/tech-industry/big-tech/us-takes-unprecedented-step-of-recommissioning-nuclear-plants-due-to-ai-power-demand-2-25-gw-currently-on-tap-from-refurbished-reactors","source":"Tom's Hardware AI","flags":[]},{"type":"news","discipline":"weather-climate","date":"2026-10-20","title":{"zh":"The Conversation · AI于2026-10-20发布的文章，涉及AI 科研与调研、AI 气候能源，原题《AI reveals how a warming climate will intensify NZ’s most extreme downpours》，详见原文链接。","en":"AI reveals how a warming climate will intensify NZ’s most extreme downpours"},"url":"https://theconversation.com/ai-reveals-how-a-warming-climate-will-intensify-nzs-most-extreme-downpours-294147","source":"The Conversation · AI","flags":[]},{"type":"news","discipline":"weather-climate","date":"2026-10-20","title":{"zh":"arXiv physics.ao-ph于2026-10-20发布的论文，涉及AI 气候能源、AI 安全、AI 软件开发，原题《Interpretability of AI weather models via intermediate decoding》，详见原文链接。","en":"Interpretability of AI weather models via intermediate decoding"},"url":"https://arxiv.org/abs/2610.24110","source":"arXiv physics.ao-ph","flags":[]},{"type":"news","discipline":"earth-sciences","date":"2026-10-20","title":{"zh":"Nature 学科·数学于2026-10-20发布的文章，涉及AI 数学，原题《A hybrid data-centric machine learning for lost circulation prediction using imbalanced dr》，详见原文链接。","en":"A hybrid data-centric machine learning for lost circulation prediction using imbalanced drilling and geological datasets"},"url":"https://www.nature.com/articles/s41598-026-75043-9","source":"Nature 学科·数学","flags":[]},{"type":"news","discipline":"earth-sciences","date":"2026-10-20","title":{"zh":"arXiv physics.ao-ph于2026-10-20发布的论文，涉及AI 气候能源、AI 安全、AI 软件开发，原题《BridgeCast: Bridging Ocean Wave Forecasts to Reanalysis via Flow Matching with Exogenous V》，详见原文链接。","en":"BridgeCast: Bridging Ocean Wave Forecasts to Reanalysis via Flow Matching with Exogenous Variables"},"url":"https://arxiv.org/abs/2610.03759","source":"arXiv physics.ao-ph","flags":[]},{"type":"milestone","discipline":"geography-remote-sensing","date":"2026-10-20","title":{"zh":"KAIST 发布 GeoCore-9B：90 亿参数、位置与尺度感知的卫星影像生成基础模型（NeurIPS 2026 录用）","en":"KAIST GeoCore-9B: 9B-parameter geo-aware generative EO foundation model (NeurIPS 2026)"},"url":"https://www.eurekalert.org/news-releases/1148233","source":"KAIST / EurekAlert!","flags":["preprint"]},{"type":"news","discipline":"economics","date":"2026-10-20","title":{"zh":"The GitHub Blog于2026-10-21发布的文章，涉及AI 软件开发，原题《How 3 Java Champions are using GitHub Copilot for increased productivity》，详见原文链接。","en":"How 3 Java Champions are using GitHub Copilot for increased productivity"},"url":"https://github.blog/ai-and-ml/github-copilot/how-3-java-champions-are-using-github-copilot-for-increased-productivity/","source":"The GitHub Blog","flags":[]},{"type":"news","discipline":"economics","date":"2026-10-20","title":{"zh":"MIT Technology Review - AI于2026-10-20发布的文章，涉及AI 综合、AI 气候能源，原题《The new economics of enterprise AI》，详见原文链接。","en":"The new economics of enterprise AI"},"url":"https://www.technologyreview.com/2026/10/20/1146937/the-new-economics-of-enterprise-ai/","source":"MIT Technology Review - AI","flags":[]},{"type":"news","discipline":"finance","date":"2026-10-20","title":{"zh":"Financial Times AI 栏目于2026-10-20发布的文章，涉及AI 金融、AI 商业融资，原题《AI debt worries spark surge in default insurance trading》，详见原文链接。","en":"AI debt worries spark surge in default insurance trading"},"url":"https://www.ft.com/content/6dd31584-064c-4015-bf5d-e5747799ccd5?syn-25a6b1a6=1","source":"Financial Times AI 栏目","flags":[]},{"type":"news","discipline":"finance","date":"2026-10-20","title":{"zh":"华尔街见闻于2026-10-20发布的文章，涉及AI 金融、AI 法律政策、AI 商业融资，原题《卢比年内跌逾5%，印度央行向10亿美元级海外并购下发核查清单：不设新限额，但银行须独立核价》，详见原文链接。","en":"卢比年内跌逾5%，印度央行向10亿美元级海外并购下发核查清单：不设新限额，但银行须独立核价"},"url":"https://wallstreetcn.com/articles/3784016","source":"华尔街见闻","flags":[]},{"type":"news","discipline":"finance","date":"2026-10-20","title":{"zh":"PANews于2026-10-20发布的文章，涉及AI 金融、AI 法律政策、AI 安全，原题《周小川：应开展研究并做好预案 防范AI领域投资失败引发的金融风险》，详见原文链接。","en":"周小川：应开展研究并做好预案 防范AI领域投资失败引发的金融风险"},"url":"https://www.panewslab.com/zh/articles/01a1585e-d140-76cf-86ff-f44337274ec4","source":"PANews","flags":[]},{"type":"news","discipline":"law","date":"2026-10-20","title":{"zh":"36氪于2026-10-20发布的文章，涉及AI 法律政策、AI 商业融资，原题《在人人都能做漫剧之后，阅文打开了自己的版权库》，详见原文链接。","en":"在人人都能做漫剧之后，阅文打开了自己的版权库"},"url":"https://36kr.com/p/4032661941883525?f=rss","source":"36氪","flags":[]},{"type":"news","discipline":"law","date":"2026-10-20","title":{"zh":"Artificial Lawyer于2026-10-20发布的文章，涉及AI 法律政策，原题《What Happens When Lawyers Can Afford to Look Further?》，详见原文链接。","en":"What Happens When Lawyers Can Afford to Look Further?"},"url":"https://www.artificiallawyer.com/2026/10/20/what-happens-when-lawyers-can-afford-to-look-further/","source":"Artificial Lawyer","flags":[]},{"type":"news","discipline":"law","date":"2026-10-20","title":{"zh":"Artificial Lawyer于2026-10-20发布的文章，涉及AI 法律政策、AI 创意媒体，原题《More Legal Research Moves as Moonlit Launches Free Commentaries》，详见原文链接。","en":"More Legal Research Moves as Moonlit Launches Free Commentaries"},"url":"https://www.artificiallawyer.com/2026/10/20/more-legal-research-moves-as-moonlit-launches-free-commentaries/","source":"Artificial Lawyer","flags":[]},{"type":"news","discipline":"education","date":"2026-10-20","title":{"zh":"HN · AI 数学/定理证明于2026-10-20发布的文章，涉及AI 数学、AI 教育、AI 软件开发，原题《Show HN: Penmentor – Claude marks up your handwritten math like a teacher》，详见原文链接。","en":"Show HN: Penmentor – Claude marks up your handwritten math like a teacher"},"url":"https://github.com/AbdullahAlhariri/penmentor","source":"HN · AI 数学/定理证明","flags":[]},{"type":"news","discipline":"education","date":"2026-10-20","title":{"zh":"IT之家于2026-10-20发布的文章，涉及AI 教育、AI 安全、AI 商业融资，原题《儿童手表品牌成小学生社交准入门槛，“小天才封闭社交倒逼家长购买”冲上热搜》，详见原文链接。","en":"儿童手表品牌成小学生社交准入门槛，“小天才封闭社交倒逼家长购买”冲上热搜"},"url":"https://www.ithome.com/1/015/567.htm","source":"IT之家","flags":[]},{"type":"news","discipline":"education","date":"2026-10-20","title":{"zh":"arXiv cs.CL于2026-10-20发布的论文，涉及AI 教育、AI 商业融资、AI 软件开发，原题《SLA-LENS: From Classroom Transcripts to CSLA-Informed and Evidence-Linked L2 Interaction A》，详见原文链接。","en":"SLA-LENS: From Classroom Transcripts to CSLA-Informed and Evidence-Linked L2 Interaction Analytics"},"url":"https://arxiv.org/abs/2610.24259","source":"arXiv cs.CL","flags":[]},{"type":"news","discipline":"linguistics","date":"2026-10-20","title":{"zh":"arXiv cs.CL于2026-10-20发布的论文，涉及AI 综合、AI 法律政策、AI 安全，原题《Translating in style: An open-weight alternative to GPT-4 for culturally aligned machine t》，详见原文链接。","en":"Translating in style: An open-weight alternative to GPT-4 for culturally aligned machine translation"},"url":"https://arxiv.org/abs/2610.23868","source":"arXiv cs.CL","flags":[]},{"type":"news","discipline":"history","date":"2026-10-20","title":{"zh":"hn-show-ai于2026-10-20发布的文章，涉及AI 软件开发，原题《Show HN: Agent Archive – Session Handoffs and Backups for Claude Code and Codex》，详见原文链接。","en":"Show HN: Agent Archive – Session Handoffs and Backups for Claude Code and Codex"},"url":"https://github.com/wangjohn/agent-archive","source":"hn-show-ai","flags":[]},{"type":"news","discipline":"literature","date":"2026-10-20","title":{"zh":"arXiv cs.CL于2026-10-20发布的论文，涉及AI 创意媒体，原题《Language Models are Lazy Poets: Creative Mode Collapse at Scale》，详见原文链接。","en":"Language Models are Lazy Poets: Creative Mode Collapse at Scale"},"url":"https://arxiv.org/abs/2610.24223","source":"arXiv cs.CL","flags":[]},{"type":"news","discipline":"music","date":"2026-10-20","title":{"zh":"Music Business Worldwide于2026-10-20发布的文章，涉及AI 创意媒体、AI 法律政策，原题《ICMP says AI firms are trying to ‘strong-arm’ Australia on copyright – and that Kylie Mino》，详见原文链接。","en":"ICMP says AI firms are trying to ‘strong-arm’ Australia on copyright – and that Kylie Minogue’s songs have already been used to train AI (report)"},"url":"https://www.musicbusinessworldwide.com/icmp-says-ai-firms-are-trying-to-strong-arm-australia-on-copyright-and-that-kylie-minogues-songs-have-already-been-used-to-train-ai-report/","source":"Music Business Worldwide","flags":[]},{"type":"news","discipline":"music","date":"2026-10-20","title":{"zh":"hn-show-ai于2026-10-20发布的文章，涉及AI 软件开发、AI 创意媒体，原题《Show HN: Shazam but for AI Slop Music》，详见原文链接。","en":"Show HN: Shazam but for AI Slop Music"},"url":"https://apps.apple.com/de/app/daft-slop-ai-music-detector/id6811013099","source":"hn-show-ai","flags":[]},{"type":"news","discipline":"visual-arts-design","date":"2026-10-20","title":{"zh":"Phys.org 机器学习与AI于2026-10-20发布的文章，涉及AI 科研与调研、AI 创意媒体，原题《AI creates art in seconds, but creative growth still matters, paper argues》，详见原文链接。","en":"AI creates art in seconds, but creative growth still matters, paper argues"},"url":"https://techxplore.com/news/2026-10-ai-art-seconds-creative-growth.html","source":"Phys.org 机器学习与AI","flags":[]},{"type":"news","discipline":"film-animation","date":"2026-10-20","title":{"zh":"雷峰网于2026-10-20发布的文章，涉及AI 创意媒体，原题《奥斯卡大师加持，「电影精神」怎么造就AI剧场？》，详见原文链接。","en":"奥斯卡大师加持，「电影精神」怎么造就AI剧场？"},"url":"https://www.leiphone.com/category/industrynews/6Vycsmo3EhndKZSe.html","source":"雷峰网","flags":[]},{"type":"news","discipline":"film-animation","date":"2026-10-20","title":{"zh":"钛媒体于2026-10-20发布的文章，涉及AI 创意媒体，原题《做了10年影视制片人，现在我每天陪老板搓“AI玩具”》，详见原文链接。","en":"做了10年影视制片人，现在我每天陪老板搓“AI玩具”"},"url":"https://www.tmtpost.com/8174207.html","source":"钛媒体","flags":[]},{"type":"news","discipline":"film-animation","date":"2026-10-20","title":{"zh":"钛媒体于2026-10-20发布的文章，涉及AI 综合，原题《AI短剧里的“彩票工厂”不再赚钱》，详见原文链接。","en":"AI短剧里的“彩票工厂”不再赚钱"},"url":"https://www.tmtpost.com/8174771.html","source":"钛媒体","flags":[]},{"type":"milestone","discipline":"music","date":"2026-10-20","title":{"zh":"Spotify：今年已从推荐中剔除 5000 多万首 AI 垃圾/刷量曲目、另下架 7500 多万首，AI 人设歌手默认不进推荐，AI 创作署名每日新增数万条","en":"Spotify: 50M+ AI slop/spam tracks removed from recommendations this year, 75M+ removed outright; AI Personas excluded from recommendations by default"},"url":"https://newsroom.spotify.com/2026-10-20/prioritizing-human-artistry/","source":"Spotify Newsroom","flags":["vendor"]},{"type":"milestone","discipline":"religious-studies","date":"2026-10-20","title":{"zh":"OpenAI 分析：ChatGPT 每周约 7000 万次宗教相关请求，45% 关于伊斯兰教、34% 关于基督教，随斋月、复活节等节期波动","en":"OpenAI analysis: ChatGPT receives ~70M religion-related requests a week, 45% about Islam and 34% about Christianity"},"url":"https://www.christianpost.com/news/chatgpt-gets-70m-requests-about-religion-weekly-most-about-islam.html","source":"The Christian Post","flags":["vendor"]},{"type":"milestone","discipline":"statistics","date":"2026-10-20","title":{"zh":"MR-MAS（Bioinformatics）：大模型多智能体端到端自动完成孟德尔随机化因果分析——取数、效应估计、结果解读到成稿，跨模型后端重复运行结果一致","en":"MR-MAS (Bioinformatics): an LLM multi-agent system that runs end-to-end Mendelian randomization — data retrieval, causal estimation, interpretation and write-up"},"url":"https://doi.org/10.1093/bioinformatics/btag722","source":"Bioinformatics","flags":[]},{"type":"milestone","discipline":"visual-arts-design","date":"2026-10-20","title":{"zh":"巴黎小皇宫美术馆开幕“Visionaries”展：谷歌艺术与文化用 Gemini 等生成式模型，依据档案照片和书信重建 19 世纪女性雕塑家（含卡米耶·克洛岱尔）已消失的工作室，展期至 2027 年 1 月","en":"Petit Palais opens “Visionaries”: Google Arts & Culture uses Gemini-based tools to reconstruct the lost studios of 19th-century women sculptors from archival photos and letters"},"url":"https://blog.google/company-news/outreach-and-initiatives/arts-culture/visionaries-petit-palais-art-basel-paris-prune-nourry/","source":"Google Blog","flags":["vendor"]},{"type":"milestone","discipline":"mathematics","date":"2026-10-19","title":{"zh":"Quanta：数论专家基本认可 OpenAI 的“准黎曼猜想”证明（零点实部在 1/12 与 11/12 之间），Sarnak 改课专门研读","en":"Quanta: number theorists largely accept OpenAI’s proof of a quasi-Riemann hypothesis (zeros between 1/12 and 11/12)"},"url":"https://www.quantamagazine.org/ai-proved-the-quasi-riemann-hypothesis-what-happens-now-20261019/","source":"Quanta Magazine","flags":["preprint"]},{"type":"news","discipline":"physics","date":"2026-10-19","title":{"zh":"量子位于2026-10-19发布的文章，涉及AI 芯片硬件，原题《从百比特芯片到整机交付，这家量子计算公司正将量子算力推向可调用服务》，详见原文链接。","en":"从百比特芯片到整机交付，这家量子计算公司正将量子算力推向可调用服务"},"url":"https://www.qbitai.com/2026/10/504277.html","source":"量子位","flags":[]},{"type":"news","discipline":"agriculture","date":"2026-10-19","title":{"zh":"Financial Times AI 栏目于2026-10-19发布的文章，涉及AI 金融，原题《Is smart farming really smart business?》，详见原文链接。","en":"Is smart farming really smart business?"},"url":"https://www.ft.com/content/aeb432cd-5d95-4062-b7a8-f494592adb34","source":"Financial Times AI 栏目","flags":[]},{"type":"news","discipline":"chip-design","date":"2026-10-19","title":{"zh":"SemiWiki于2026-10-20发布的文章，涉及AI 芯片硬件，原题《ChipAgents Expands AI Across Chip Design. The Test Is Faster, Verified Results》，详见原文链接。","en":"ChipAgents Expands AI Across Chip Design. The Test Is Faster, Verified Results"},"url":"https://semiwiki.com/eda/chipagents-ai/375026-chipagents-expands-ai-across-chip-design-the-test-is-faster-verified-results/","source":"SemiWiki","flags":[]},{"type":"news","discipline":"aerospace","date":"2026-10-19","title":{"zh":"Phys.org 机器学习与AI于2026-10-19发布的文章，涉及AI 科研与调研，原题《AI that reconstructs cloud-covered ground in satellite images with photo-like clarity》，详见原文链接。","en":"AI that reconstructs cloud-covered ground in satellite images with photo-like clarity"},"url":"https://techxplore.com/news/2026-10-ai-reconstructs-cloud-ground-satellite.html","source":"Phys.org 机器学习与AI","flags":[]},{"type":"news","discipline":"energy-fusion","date":"2026-10-19","title":{"zh":"HPCwire于2026-10-20发布的文章，涉及AI 芯片硬件，原题《Great Sky, National Labs Explore Superconducting AI for Plasma Control》，详见原文链接。","en":"Great Sky, National Labs Explore Superconducting AI for Plasma Control"},"url":"https://www.hpcwire.com/off-the-wire/great-sky-national-labs-explore-superconducting-ai-for-plasma-control/","source":"HPCwire","flags":[]},{"type":"news","discipline":"energy-fusion","date":"2026-10-19","title":{"zh":"HPCwire于2026-10-20发布的文章，涉及AI 芯片硬件，原题《Superconducting Neuromorphic Chip Passes Fusion Test》，详见原文链接。","en":"Superconducting Neuromorphic Chip Passes Fusion Test"},"url":"https://www.hpcwire.com/2026/10/19/superconducting-neuromorphic-chip-passes-fusion-test/","source":"HPCwire","flags":[]},{"type":"milestone","discipline":"weather-climate","date":"2026-10-19","title":{"zh":"中国气象局“风清”AI 全球短中期集合预报系统 V1.0 通过业务化评审并投入运行，有效预报时效 11 天","en":"CMA’s Fengqing AI global ensemble forecast system V1.0 passes operational review and goes live (11-day skill)"},"url":"https://www.chinanews.com/sh/2026/10-19/10717398.shtml","source":"中国新闻网","flags":[]},{"type":"news","discipline":"geography-remote-sensing","date":"2026-10-19","title":{"zh":"Phys.org 机器学习与AI于2026-10-19发布的文章，涉及AI 科研与调研，原题《AI that reconstructs cloud-covered ground in satellite images with photo-like clarity》，详见原文链接。","en":"AI that reconstructs cloud-covered ground in satellite images with photo-like clarity"},"url":"https://techxplore.com/news/2026-10-ai-reconstructs-cloud-ground-satellite.html","source":"Phys.org 机器学习与AI","flags":[]},{"type":"news","discipline":"psychology","date":"2026-10-19","title":{"zh":"Phys.org 机器学习与AI于2026-10-19发布的文章，涉及AI 科研与调研、AI 教育，原题《AI chatbots could help mental health but also reinforce harmful thinking》，详见原文链接。","en":"AI chatbots could help mental health but also reinforce harmful thinking"},"url":"https://techxplore.com/news/2026-10-ai-chatbots-mental-health.html","source":"Phys.org 机器学习与AI","flags":[]},{"type":"news","discipline":"sociology-politics","date":"2026-10-19","title":{"zh":"The Verge AI于2026-10-19发布的文章，涉及AI 综合、AI 法律政策，原题《Democrats think cracking down on AI could give them an edge》，详见原文链接。","en":"Democrats think cracking down on AI could give them an edge"},"url":"https://www.theverge.com/policy/1013113/democrats-ai-midterms","source":"The Verge AI","flags":[]},{"type":"milestone","discipline":"philosophy","date":"2026-10-19","title":{"zh":"Seth、Block、Tononi 在 Nature 撰文：AI 没有意识、也许永远不会有的三个理由，Chalmers 等公开回应","en":"Seth, Block & Tononi in Nature: three reasons why AI is not conscious and perhaps never will be"},"url":"https://doi.org/10.1038/d41586-026-03313-7","source":"Nature","flags":[]},{"type":"news","discipline":"religious-studies","date":"2026-10-19","title":{"zh":"COMECE Assembly: EU bishops call for AI to remain at the service of the human person","en":"COMECE Assembly: EU bishops call for AI to remain at the service of the human person"},"url":"https://www.comece.eu/declaration-comece-assembly-eu-bishops-call-for-ai-to-remain-at-the-service-of-the-human-person/","source":"COMECE","flags":[]},{"type":"news","discipline":"literature","date":"2026-10-19","title":{"zh":"Phys.org 机器学习与AI于2026-10-20发布的文章，涉及AI 科研与调研，原题《Who is Elias the lighthouse keeper? AI fiction tells the same old story》，详见原文链接。","en":"Who is Elias the lighthouse keeper? AI fiction tells the same old story"},"url":"https://techxplore.com/news/2026-10-elias-lighthouse-keeper-ai-fiction.html","source":"Phys.org 机器学习与AI","flags":[]},{"type":"news","discipline":"music","date":"2026-10-19","title":{"zh":"Music Business Worldwide于2026-10-20发布的文章，涉及AI 创意媒体、AI 法律政策，原题《‘The only way to properly create value in music is to make lasting copyrights.’》，详见原文链接。","en":"‘The only way to properly create value in music is to make lasting copyrights.’"},"url":"https://www.musicbusinessworldwide.com/aaron-bay-schuck-the-only-way-to-properly-create-value-in-music-is-to-make-lasting-copyrights/","source":"Music Business Worldwide","flags":[]},{"type":"news","discipline":"chip-design","date":"2026-10-18","title":{"zh":"Semiconductor Engineering于2026-10-19发布的文章，涉及AI 芯片硬件、AI 教育，原题《A Composable Thermal Solver For Heterogeneous 2.5D/3D-IC Packages (Cadence, Notre Dame)》，详见原文链接。","en":"A Composable Thermal Solver For Heterogeneous 2.5D/3D-IC Packages (Cadence, Notre Dame)"},"url":"https://semiengineering.com/a-composable-thermal-solver-for-heterogeneous-2-5d-3d-ic-packages-cadence-notre-dame/","source":"Semiconductor Engineering","flags":[]},{"type":"news","discipline":"economics","date":"2026-10-18","title":{"zh":"Financial Times AI 栏目于2026-10-18发布的文章，涉及AI 金融，原题《Global economy set for AI-driven surge, FT survey suggests》，详见原文链接。","en":"Global economy set for AI-driven surge, FT survey suggests"},"url":"https://www.ft.com/content/c6aeb3a1-1874-4947-8ffc-4d993301244a?syn-25a6b1a6=1","source":"Financial Times AI 栏目","flags":[]},{"type":"milestone","discipline":"law","date":"2026-10-18","title":{"zh":"美国联邦上诉法院维持 Thomson Reuters 诉 Ross Intelligence 案判决：用 Westlaw 内容训练法律 AI 不构成合理使用（判决书仍密封）","en":"Appeals court upholds Thomson Reuters’ win over Ross Intelligence on AI-training fair use (opinion sealed)"},"url":"https://www.theverge.com/ai-artificial-intelligence/1013500/appeals-court-upholds-thomson-reuters-win-in-ai-training-lawsuit","source":"The Verge","flags":["tbv"]},{"type":"news","discipline":"philosophy","date":"2026-10-18","title":{"zh":"Alignment Forum于2026-10-18发布的文章，涉及AI 安全、AI 数学，原题《The Alignment Problem: Computer Science Meets Analytic Philosophy》，详见原文链接。","en":"The Alignment Problem: Computer Science Meets Analytic Philosophy"},"url":"https://www.alignmentforum.org/posts/Emp7awtxsYn8MusNq/the-alignment-problem-computer-science-meets-analytic","source":"Alignment Forum","flags":[]},{"type":"milestone","discipline":"education","date":"2026-10-17","title":{"zh":"中国农村 8647 名小学生大规模随机试验：AI 个性化学习未显著提高数学成绩（Economics of Education Review）","en":"Large RCT of 8,647 rural Chinese pupils: AI-personalised learning no significant maths gain"},"url":"https://doi.org/10.1016/j.econedurev.2026.102850","source":"Economics of Education Review","flags":[]}]}