Rohan Paul· @rohanpaul_ai · X·· 2026-09-06AI 评分43
AI 导读
自我改进的智能体需要一种有效的方式来记住什么方法有效。 SkillGLoW 表明，智能体应记住可复用的任务解决方式，而非每一个过往任务，从而以 3.6 倍更精简的库获得 17.2 分的提升。 自我改进的智能体在存储共享流程时，可以用更少的记忆实现更好的表现。 任务特定细节根据当前任务重建，而非永久存储。 它还在真实执行中测试记忆更新，并拒绝使智能体变差的更改。
正文 · 原文
Self-improving agents need a useful way to remember what worked.
SkillGLoW shows agents should remember reusable ways of solving tasks, not every past task, gaining 17.2 points with a 3.6× more compact library.
Self-improving agents can remember less and perform better when they store shared procedures.
The task-specific details are rebuilt from the current task instead of stored permanently.
It also tests memory updates in real execution and rejects changes that make the agent worse.
来源：Rohan Paul · x.com