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
- Dev generates a full software promo video in pure code with GPT-6 Astra, no generative models — op7418 · 2026-09-17
- Frida, an iOS Personal Life Agent, Opens Second TestFlight Batch — Scobleizer · 2026-09-17
- Doubao Seed-2.1-pro-0915 hands-on: tool-calling score jumps 38.4% to 64.8%, coding cost down 40% — vista8 · 2026-09-17
- Dev claims harness makes agent computer use consume identical tokens to regular tool use — TejasKumar_ · 2026-09-17
- Spreadsheet agents should deliver editable drafts, not just approval summaries — Mountain_Athlete1350 · 2026-09-17
- Score Studio Launches as All-in-One Vision AI Platform for Annotation, Training and Deployment — markjeffrey · 2026-09-17