arXiv 2610.24649 · cs.AI · 2026-10-17
Good Judgment, Bad Advice: Decision-Model Controllers Raise Agent Cost Without Improving Success
Naman Vats
机构待核实
一句话要点
在固定任务下,决策模型Jev虽未提升成功率,反而增加22%成本,且严重失准,导致无效验证行为。
Jev increases cost by 22% without improving success or calibration, leading to unnecessary verification.
要点
- 代理在决策点是否应引入低成本模型以提升效率,而非增加成本。
- 在Terminal-Bench 21个任务上对比无控制器、规则、小LLM和Jev决策模型的性能与成本。
- Jev使每尝试成本上升19%,通过率相同;其预测概率严重失准,误判完成状态,导致无效验证。
- 决策信号未带来实际收益,且所有控制器均未选择放弃任务,说明当前决策机制缺乏有效性。
English overview
The study evaluates whether a decision model (Jev) improves agent performance or reduces cost. Results show no improvement in pass rate, increased cost by 19%, and poor calibration. Jev's predictions misrepresent completion status, leading to unnecessary verification. No controller ever abandons a run, indicating limited utility of decision signals.
只根据 arXiv 官方摘要,由开源模型整理,经数字/专名自动核对(必须在摘要中出现),以论文原文为准。
阅读 arXiv 原文 PDF Hugging Face 讨论 代码仓库
主题与学科
同主题:智能体
- Raven: The Harness of Harnesses for Composable Agentic Intelligence ▲748
- False Frontiers: Diagnosing and Mitigating Co-Cheating in Self-Evolving Search Agents ▲642
- Omni-IO Skills: Harnessing Your Agent Omni-Native ▲485
- BizAgentBench: Unveiling Hidden Challenges in Deploying Long-Horizon Agents for Real-World Business ▲388
- Player-1: Learning to Play from Synthetic Experience ▲347
同主题:大语言模型
- Jeva: System One Models from First Principles 2026-10-13 · ▲371
- Post-Training Leaves Behavioral Shadows on Unrelated Decisions 2026-09-24 · ▲273
- WeVisOne-VL: Unified Visual Perception from Objects to Pixels 2026-10-22 · ▲247
- Thinking Inertia: LLMs Keep Thinking When Told Not To 2026-10-08 · ▲241
- OmniEdu: Open Foundation Models for Learning and Teaching 2026-09-19 · ▲240
数据说明
元数据来自 arXiv API(已核实);点赞来自 Hugging Face 每日论文;引用来自 Semantic Scholar。最近刷新 2026-10-17。
要点由本站根据摘要自写,不转载摘要原文,请阅读原文。