A five-level framework argues most AI self-improvement today sits at L1-L2, real RSI starts at L5
青稞AI · wechat · 2026-09-17
A new paper, "The Last AI Built by Humans: Toward Genuine Recursive Self-Improvement," maps RSI on five levels by how much autonomy a system has over its own improvement — from executing changes (L1) to recursively improving the improvement process itself (L5). It formalizes a unified loop (experience → modification → verification → retention) and argues current Self-Refine/Self-Reward methods mostly sit at L1/L2, with verification, stability and open-environment challenges blocking L3–L5.
More from AGI Musings
- vikhyat predicts alternating compute oversupply and undersupply on the road to AGI — TheZachMueller · 2026-11-02
- Sam Altman: AI has surpassed top scientists, and 2026 will be a turning point year — kimmonismus · 2026-11-02
- DeepSeek design doc: a person can only follow 3-5 agent sessions at once, attention is the real bottleneck — heyshrutimishra · 2026-11-02
- AI researcher: we have no reliable way to tell whether an AI system is conscious — dioscuri · 2026-11-02
- Toby Ord: track both what models can and can't do to see real capability gaps — tobyordoxford · 2026-11-02
- Forward Deployed Verifier will soon be as valuable as FDE, argues poster — vaibhavbetter · 2026-11-02