Social sciences · OECD FoS 5.3 · China MoE 0401 教育学
Education
教育学
L2Local breakthroughs ↑ Rising Score 2.10 · Last reviewed 2026-10-21 · Next review 2027-01-19
BIGGEST BREAKTHROUGH 2025-06-03 AI · majorField · major
Harvard physics RCT: AI tutoring beats active learning (Scientific Reports) ↗
RCT: AI-tutor learning gains about double active learning
Why this level
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.
S = 0.30·method + 0.35·discovery + 0.25·adoption + 0.10·education (education scores 0 when there is no evidence; if empty, the other weights are renormalised)
Sources
Composite 1.5–2.49: at least one peer-reviewed/officially verified AI-key result, or one dimension at 3; routine workflow largely unchanged. All levels →
Research directions
AI tutorsAdaptive practiceAutomated grading & feedback
Papers
- Putting AI to the test: Evidence from a large-scale RCT in rural China
Large null result
2026-10-17 · Economics of Education Review
- Generative AI without guardrails can harm learning: Evidence from high school mathematics
Practice gains vanish and exam scores drop 17% after removal
2025-06-25 · Proceedings of the National Academy of Sciences
- AI tutoring outperforms in-class active learning: an RCT introducing a novel research-based design in an authentic educational setting
RCT: AI-tutor learning gains about double active learning
2025-06-03 · Scientific Reports
- A Stratified Multi-Rater Evaluation of LLM-Based Virtual Standardized Patients with a Deployed Data-Generation Platform
A stratified multi-rater evaluation of LLM-based virtual standardised patients for OSCE medical education.
2026-10-21 · arXiv · Yongjin Yi, Jinyong Park, Sejoong Kim
- Bias and Reliability in Learning-Style Inference from Student--AI Dialogues: An Evaluation of GPT-4o and Gemini 2.0 Flash
LLMs infer learning styles from dialogues mostly by predicting majority labels, under-identifying intuitive and global learners.
2026-10-20 · arXiv · Samaneh Zamanifard, Sajad Goudarzi, Andrew Robb
- ValiAssist: Scaling LLM-generated Programming Feedback Validation by Demonstration
Infers reusable validation logic from instructors’ edits to AI programming feedback; tested with 16 instructors.
2026-10-19 · arXiv · Xiaohang Tang, Rehema Abulikemu, Peng Gao, Yan Chen
- MITRA-lexicon: Lexica of attested translations from Sanskrit into English, Tibetan and Chinese
Corpus-grounded Sanskrit–English/Tibetan/Chinese lexica with 2.18M headwords and 10.4M attestations.
2026-10-19 · arXiv · Sebastian Nehrdich, Kengo Harimoto, Sven Sellmer, Oliver Hellwig
- It's Complicated: Factors Influencing the Effectiveness of LLM-Generated Feedback in a Live Programming Course
Three iterations of LLM feedback in a live course: mixed associations, with longer feedback linked to lower next-submission scores.
2026-10-18 · arXiv · Tsvetomila Mihaylova, Arto Hellas, Bita Akram, Narges Norouzi
- Acharya 2.0: An AI Tutor for Self-Learning Subjects in Hindi
An AI tutor for self-learning school subjects in Hindi.
2026-10-18 · arXiv · Shantipriya Parida, Debasish Dhal, Sonal Khosla, Sambit Sekhar
- EduWild: How Teacher-Configured Chatbots Fail in Real-World Use and How to Fix Them
1,726 conversations with 217 teacher-configured chatbots; most common failure is skipping required workflow steps (53.8%).
2026-10-16 · arXiv · Ziqi Liu, Ruiwei Xiao, Yuxuan Li, John Stamper
- Can AI interview? Comparing AI-led to Human-led Chat Interviews
First direct comparison of data quality between AI-led and human-led chat interviews.
2026-10-16 · arXiv · Semra Guven, Tobias Gårdhus, Andreas Bjerre-Nielsen, Anna Rogers
- From arXiv, bioRxiv, medRxiv and OpenAlex (last ~30 days); summaries written by AIMap from the abstract.
Milestones 6
Big glowing nodes are leaps: AI leap cool · Field leap warm · double ring = both
-
Paper
Large RCT of 8,647 rural Chinese pupils: AI-personalised learning no significant maths gainLarge null result
Economics of Education Review -
Paper
NBER: two-year Khanmigo school experimentTwo-year school trial, ~0.06–0.08 SD/yr
NBERTo be verified -
Paper
NBER: AI tutoring in mastery-based maths practice (6,000+ students)6,000+ student tutoring experiment
NBERTo be verified -
Policy
China moves to build an AI-literacy system; AI education in schools to be near-universal by 2030National curriculum policy
gov.cn -
Paper
Generative AI without guardrails can harm learning (PNAS)Practice gains vanish and exam scores drop 17% after removal
PNAS -
Paper
Harvard physics RCT: AI tutoring beats active learning (Scientific Reports)RCT: AI-tutor learning gains about double active learning
Scientific Reports
No milestones match this filter.
More papers & repos 3
Latest news 12
- SLA-LENS: From Classroom Transcripts to CSLA-Informed and Evidence-Linked L2 Interaction AnalyticsarXiv cs.CLenPaperPreprint
- Voice-Based AI Feedback in Higher Education: A Field Study on Quality and ActionabilityarXiv cs.CYenPaperPreprint
- Exploring Visual Access and Proactive AI Tutoring for Handwritten MathematicsarXiv cs.CYenPaperPreprint
- Extending synthetic medical education: generating OSCE cases with LLMsnpj Digital Medicineen