Google Researchers Propose Dream-RSI: Recursive Self-Improvement via Offline Replay Simulators
mikeflache · x · 2026-09-18
A new arXiv paper, Dream-RSI: Recursive Self-Improvement through Evolving Worlds (by @zhengtoong et al., with Google), introduces a framework to accelerate autonomous scientific and mathematical discovery:
- It converts an agent's historical exploration data into a low-cost, offline replay simulator.
- The agent tests and refines search strategies through this simulated "dreaming" process before deploying them online.
- The result is efficient recursive self-improvement without burning excessive compute on live model calls.
Related event: Google's Dream-RSI Lets Agents Self-Improve by Replaying Their Own History(7 posts)→
More from Research
- Index pretraining lifts humanoid zero-shot success from 8% to 56% — coreylynch · 2026-09-18
- A 'Life Diary' Eval Could Be the Toughest Test Yet for Continual Learning in LLMs — JohnnyNi13 · 2026-09-18
- Pretraining on Human Behavior Reportedly Boosts Task Success from 9% to 56% — Dr_Singularity · 2026-09-18
- Index pretraining lifts Helix 2.5 zero-shot success from 8% to 56%, generating 50 min of data per second — coreylynch · 2026-09-18
- LLMs got good at text and stayed bad at tables — and it's not just a training-data problem — FamiliarSlide7685 · 2026-09-18
- IFM releases K2-Horizon-7B, a diffusion-augmented LLM claiming lossless 5,200 tokens/s — Zulfiqaar · 2026-09-18