arXiv cs.CL· Weixiang Zhao, Yichen Zhang, Yanyan Zhao, Jinming Bai, Bin Liang, Ruifeng Xu, Bing Qin, Ting Liu·· 12 小时前AI 评分42
自进化智能体只能修复失败，难以从失败中泛化
Self-Evolving Agents Can Repair Failures but Struggle to Generalize from Them
AI 导读
研究通过 Cold-Start Failure-Only Evolution（CFOE）设定发现，自进化 LLM 智能体存在一致的“修复—泛化”差距：仅凭失败经验能修复此前失败的任务，却很少在未见任务上带来持续提升。该结论覆盖三种自进化范式、多个 LLM 主干和四个交互环境。放松 CFOE 后，验证器反馈与额外交互收益有限，而成功轨迹能显著改善迁移。
正文 · 原文
Abstract:Experience-driven self-evolution offers a promising path for large language model (LLM) agents to continually improve through interaction. Yet existing approaches often rely on successful trajectories or external evaluation signals, leaving unclear whether agents can evolve when such positive anchors are unavailable. We study this question through Cold-Start Failure-Only Evolution (CFOE), where agents start without accumulated experience and learn only from failed interactions without privileged guidance. Across three self-evolving paradigms, multiple LLM backbones, and four interactive environments, we uncover a consistent repair-generalization gap: failure-derived experience can help repair previously failed tasks, but rarely yields sustained improvements on unseen tasks. Further analysis shows that agents often misinterpret failures, while even verified-correct experience remains difficult to transfer. Relaxing CFOE reveals that verifier feedback and additional interaction provide limited gains, whereas successful trajectories substantially improve transfer. These findings suggest that current self-evolving agents are better at repairing observed failures than building generalizable capabilities from failure.
| Comments: | 28 pages, 9 figures, 6 tables. Weixiang Zhao and Yichen Zhang contributed equally |
| Subjects: | Computation and Language (cs.CL) |
| Cite as: | arXiv:2610.22984 [cs.CL] |
| (or arXiv:2610.22984v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.22984 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Yichen Zhang [view email]
[v1]
Fri, 16 Oct 2026 06:10:01 UTC (1,772 KB)
来源：arXiv cs.CL · arxiv.org