arXiv cs.AI· Jianghong Zhou, Hao Rong, Chen Lin, Daoan Zhang·· 3 天前AI 评分32
LLM 智能体的推理收缩与安全记忆复用边界
Many Steps into One: Reasoning Contraction and the Limits of Safe Memory Reuse for LLM Agents
AI 导读
研究提出"推理收缩"概念，指出思维模板、技能库、智能体记忆与缓存把多步推理压缩为一步，但事实变化后记忆会悄然给出错误答案。在五个 LLM 的多跳问答测试中，ExpeL 与 ACE 式计数器比经认证记忆多输出 7.0-13.5 倍过期答案，代码任务上多出 35-73 倍。作者提出按各记忆自身似然比加权失败、对照实时推理认证效用、事实变化时重启证书的方案。
正文 · 原文
Abstract:Thought templates, skill libraries, agent memories and caches turn many reasoning steps into one. We call this reasoning contraction. A contracted memory saves reasoning steps and errors until its facts change, then quietly answers wrong. Most systems trust memories by counting successes and failures. We prove that such counters are badly tuned change detectors: a symmetric counter notices a partial change only after a power of its false-alarm horizon, and an unclipped one may never notice it. On five LLMs answering multi-hop questions over edited facts, ExpeL- and ACE-style counters serve 7.0-13.5x more stale answers than certified memories given the same information, and on code tasks with an exact verifier 35-73x more. Safety also has a price. When facts change and each memory is judged only by its own verified outcomes, one number, the certification load, sets how much reasoning any safe memory can skip, and under Zipf demand it yields a scaling law. Our recipe weighs each memory's failures by its own likelihood ratio, certifies its utility against live reasoning, and restarts the certificate when facts change, with the likelihood-ratio detector as the safety net for unannounced changes; for large edits it also invalidates the memory.
| Comments: | 57 pages |
| Subjects: | Computation and Language (cs.CL); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2610.17544 [cs.CL] |
| (or arXiv:2610.17544v1 [cs.CL] for this version) | |
| https://doi.org/10.48550/arXiv.2610.17544 arXiv-issued DOI via DataCite (pending registration) |
Submission history
From: Jianghong Zhou [view email]
[v1]
Tue, 13 Oct 2026 01:16:37 UTC (444 KB)
来源：arXiv cs.AI · arxiv.org