Lessons from Luck: Do LLM-Written Lessons Carry Chance into an Agent's Next Decision?

Computer Science > Machine Learning arXiv:2610.33029 (cs) [Submitted on 12 Oct 2026] Title: Lessons from Luck: Do LLM-Written Lessons Carry Chance into an Agent’s Next Decision? Authors: Kazuki Nakayashiki.

计算机科学 > 机器学习 arXiv:2610.33029 (cs) [提交于 2026 年 10 月 12 日] 标题：运气带来的教训：大语言模型编写的经验教训是否会将偶然性带入智能体的下一次决策？作者：Kazuki Nakayashiki。

Abstract: Self-improving agents write lessons from the outcomes of their runs; when outcomes are noisy, a lesson can carry chance into later decisions. We measure one link of this chain in a registered study. In 256 synthetic worlds, each a recurring task with two interchangeable tools of recorded reliability, a scripted previous run used the more reliable tool (a good decision) or the less reliable one (a bad decision), with an assigned outcome: success or a transient failure.

摘要：自我改进型智能体会根据运行结果总结经验教训；当结果存在噪声时，这些教训可能会将偶然性带入后续决策中。我们在一项注册研究中测量了这一链条中的一个环节。在 256 个合成世界中，每个世界都是一项重复性任务，包含两种可靠性已知的可互换工具。脚本化的前一次运行使用了更可靠的工具（正确的决策）或较不可靠的工具（错误的决策），并被分配了成功或短暂失败的结果。

A model wrote a lesson from the run with ReasoningBank’s memory-writing instruction; a fresh session of the same model, holding the authoritative log and the previous result, chose the next run’s first tool with one memory item: that lesson, the lesson from the other outcome, an irrelevant item, or an unreasoned directive. The protocol was frozen, deposited and timestamped before any confirmatory call; 11 LLM endpoints gave 90,084 answers.

模型使用 ReasoningBank 的记忆写入指令根据运行情况编写了教训；同一模型的全新会话在持有权威日志和前一次结果的情况下，通过一项记忆条目选择下一次运行的首选工具：该教训、来自另一种结果的教训、无关项或无推理指令。该协议在任何确认性调用之前已冻结、存档并加盖时间戳；11 个大语言模型端点共给出了 90,084 个答案。

After a chance failure of the better tool, the panel’s lessons carried the chance (registered reading luck-carrying; interaction +0.112, 95% interval [+0.097, +0.126]): readers named the worse tool in 11.6% of next runs with the lesson written from that failure, against 0.4% with an irrelevant item and 0.4% with the lesson written from a success. The effect was +0.042 where the log stated that failures are independent and +0.181 where it did not (descriptive).

在较好工具发生偶然性故障后，专家组总结的教训确实携带了偶然性（注册读数为运气携带；交互作用 +0.112，95% 置信区间 [+0.097, +0.126]）：在使用从该故障中总结出的教训进行下一次运行时，阅读者有 11.6% 的概率选择了较差的工具，而使用无关项或从成功中总结出的教训时，该比例仅为 0.4%。在日志声明故障是独立的情况下，该效应为 +0.042，而在未声明的情况下为 +0.181（描述性统计）。

After a lucky success of the worse tool, the interaction was bounded below 0.10 (+0.002), a bound, not evidence of no harm. In exploratory readings, 6 endpoints carried the chance (up to +0.352); 5 were bounded below 0.10. In secondary comparisons, a three-sentence decision-quality note in the writing instruction lowered the worse-tool rate after a chance failure to 7.5%, and one instruction for both outcomes with a no-change option to 3.1%; neither met the registered harm-reduction rule at every such endpoint.

在较差工具因运气而成功后，交互作用被限制在 0.10 以下（+0.002），这只是一个界限，并非没有危害的证据。在探索性读数中，6 个端点携带了偶然性（最高达 +0.352）；5 个端点被限制在 0.10 以下。在二次比较中，写入指令中包含的三句决策质量说明将偶然性故障后选择较差工具的比例降低至 7.5%，而包含“无变化”选项的针对两种结果的统一指令将其降低至 3.1%；但两者均未在每个此类端点上达到注册的危害降低规则。