论文

自改进记忆成为持续学习智能体研究热点,Meta 与 Google 同周发文

精选理由

搞智能体记忆的同学该看看这篇梳理。Meta 让模型自己写记忆策略再用 LoRA 微调进权重,RIME 直接用 RL 训练写记忆,两条路线对比很清楚,文末还附了一串相关论文链接。

研究博主 Elvis Saravia 推荐了一组关于自改进记忆(self-improving memory)的论文,指出该方向正成为持续学习智能体的关键组件。仅一周内,Meta 和 Google 各自发布了相关研究。

正文

Meta 的做法分两步:先让模型搜索更好的记忆指令,即通过提示词让模型改写自己的记忆策略(记什么、怎么记、回忆什么),再用 SFT 配 LoRA 把表现最好的记忆习惯微调进模型权重。同期提到的 RIME 则走另一条路,跳过指令搜索,直接用 RL 训练单个模型同时完成任务和写记忆,并对有助于后续任务的笔记给奖励。

文中还列出 Google、Continual-ARC、Adaptive Memory Selection、MemCredit 等多项相关工作,适合研究递归自改进和持续学习的开发者追踪。

原文 · elvis

Highly recommended read. Self-improving memory is starting to look like an important component for continual-learning agents. So pay close attention to this research direction if you are studying recursive self-improvement. Just last week, I featured a few papers on the topic, and this week both Meta and Google published on it. (find more references towards the end) Meta lets a model design its own memory habits. Then they train those habits into the model weights. The goal is to allow the system to learn the memory design (what to memorize, how to memorize, and what to recall). This work is actually in the same vein as RIME, which I wrote some notes on last week: x.com/omarsar0/statu… Meta first searches for better memory instructions (prompts the model to rewrite itself), then fine-tunes the model (SFT with LoRA) on runs that used the best ones. RIME, on the other hand, skips the search and uses RL to train one model to solve tasks and write memories, rewarding notes that help on later tasks. It also trains the model to explore so it collects useful information. Different ways to address the problem. More useful links below: Google Self-Improving Memory for Continual Learning - x.com/omarsar0/statu… How to Train Memory Writing for Continual Learning - x.com/omarsar0/statu… Can LLM Agents Organize Their Own Memory - x.com/omarsar0/statu… Continual-ARC - x.com/omarsar0/statu… Adaptive Memory Selection for LLM Agents - x.com/omarsar0/statu… MemCredit - x.com/omarsar0/statu…📊 8 ⚡

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:记忆机制、持续学习、Meta、RIME,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。