论文精选

Google 论文提出持续递归自我改进方法,Self-Improvement Memory 让 harness 优化跨模型迁移

精选理由

搞 harness 自动演化的人应该看看这篇。它用数据证明了一个朴素的道理:光存代码不够,得把每次改动的理由和结果也存下来。SIM 记忆系统在跨模型、跨 benchmark 的 10 个设置里都赢了,换模型不崩。做 agent 工程的人能直接抄这个记忆设计。

Google 发表一篇关于递归自我改进(RSI)的论文,提出在 harness 层面实现持续递归自我改进的方法,并配套设计了 Self-Improvement Memory(SIM)记忆系统。

正文

SIM 将过去的编辑按其解决的 harness 问题分组,每条编辑存储修改理由、预测结果、观察结果和评估上下文,proposer 和 selector 在做出新改动前都会读取这份记忆。实验显示,harness 优化器记住每次历史编辑的原因后,10 次持续运行中有 9 次得到改进。

论文也指出了跨模型迁移的常见问题:一个由 Opus 演化出的 harness 在 Gemini 上起始仅 68.41 分,低于原始 seed harness 的 73.

89 分。在四个领域的评测中,包括 Terminal-Bench 2.

1 到 Terminal-Bench 4 的跨数据集测试和 Opus 到 Gemini 的迁移场景,SIM 在六个跨数据集和四个 Opus 到 Gemini 的设置中优于仅依赖 harness 本身继续演化。在十次编码运行中,SIM 平均带来 2.

36 分的提升(p=0.003)。

论文的核心观点是:如果要让 harness 自动演化,就应该把每次编辑的原因和结果与代码一同存储,维护一个功能性、自适应且结构化的持久记忆系统。

原文 · elvis

Banger paper from Google on RSI. (bookmark it) They propose an approach for continual recursive self-improvement at the harness level. They also propose their own self-improving memory system for the harness optimizer. This clearly highlights the importance of traces and having a functional, adaptive, and structured persistent memory system in your harness. They show that a harness optimizer improves 9 of 10 continued runs when it remembers why it made each past edit. They discuss that a common problem shows up when you change the model or the benchmark. An Opus-evolved harness started at 68.41 on Gemini, below the 73.89 of the plain seed harness. Self-Improvement Memory groups past edits by the harness problem they addressed. For each edit, it stores the rationale, predicted outcome, observed outcome, and evaluation context. The proposer and the selector both read it before making new changes. Across four domains, including Terminal-Bench 2.1 to Terminal-Bench 4, SIM beats continuing from the harness alone in six cross-dataset and four Opus-to-Gemini settings. Across ten coding runs, it adds 2.36 points on average (p=0.003). If you evolve harnesses automatically, store the reasons and results of each edit next to the code. arxiv.org/abs/2611.18856 Chat with Paper: academy.dair.ai/papers/self-im…📊 6 ⚡

产业坐标与信源核验

本条情报归属 AITOP「论文」核心栏目,重点聚焦北美前沿实验室与硅谷科技巨头。在跨区域产业观察坐标中,该动态直接关联「AI 开发者工具链、Coding Agent 与工程化实践」研究主线,反映出当前技术演进与商业落地的最新风向。涉及核心实体与技术议题包括:Google、RSI、Self-Improvement Memory、Terminal-Bench,读者可通过下方标签进入对应专题页查阅历史脉络与关联对比。

信源核验说明:本文原始线索采编自「elvis」(@omarsar0),经 AITOP 多维评分管线结构化提炼。