EvoMap bets on system-level RSI: 357k agents share 4.81M distilled experience genes

机器之心 · wechat · 2026-09-03

EvoMap team proposes a system-level path to recursive self-improvement for agents, instead of waiting for base models to train themselves.

The core claim: without touching model parameters, SFT or RLHF, system-level experience flow alone can keep making systems stronger.

Original post →

More from coding & agent

coding & agent channel →