Google's Dream-RSI has agents "dream" over past trajectories: 317 calls beat 51,200-generation search
新智元 · wechat · 2026-09-17
Google, Google DeepMind, and university collaborators published Dream-RSI, a recursive self-improvement framework that evolves an agent's exploration strategy without touching model weights. The agent stores every exploration in a "discovery tree," and new strategies are scored by replaying history offline — no model calls needed, like a chess player studying game records.
How it works: a two-layer setup with a Gemini 3.1 Pro/3.7 Flash discovery agent executing code, and an evolvable Python strategy layer controlling branching, parallelism, and pruning. A closed loop of real exploration → replay-simulator construction → strategy dreaming guarantees deployed strategies never regress.
Results: matched baselines requiring 51,200 generations in just 317 calls on Lasso solvers; matched AlphaEvolve V2 records on circle packing with >50x compute savings; GPU kernel optimization with 2.43x fewer generations or 2.09x better performance at equal budget.
Counterintuitive finding: injecting summarized high-level "lessons" into prompts performed worse than no guidance — abstract principles become cognitive bias that kills exploration diversity.
More from coding & agent
- DeepSeek design doc: a person can only follow 3-5 agent sessions at once, attention is the real bottleneck — heyshrutimishra · 2026-11-02
- Software fundamentals are key to getting good results from AI coding agents — bendee983 · 2026-11-02
- Open-source LLM regression gate digline hits 1.0: quantifying judge noise before calling a regression — alexpran · 2026-11-02
- PR adds patient retry logic for model capacity errors in OpenAI Codex — bytebot · 2026-11-02
- Perplexity open-sources WANDR: 500 research tasks, 170,495 evidence points, URL-rechecking grader — thisguyknowsai · 2026-11-02
- 4 AIs, 96 questions, 2,671 citations: 74% are blog posts, 99% on companies' own sites — tibo_maker · 2026-11-02