Dream-RSI: agents self-improve policies by 'dreaming' over replayed exploration histories
inductionheads · x · 2026-09-16
A study shared by Steve Hsu proposes Dream-RSI, a recursive self-improvement loop: the agent continuously collects discovery histories via online exploration, builds replay simulators from them, refines meta-exploration strategies through 'dreaming,' then redeploys the improved policy online. Experiments show gains in both discovery effectiveness and efficiency across several settings.
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