RSI-Anything: Unlocking Recursive Self-Improvement with World Models
arXiv cs.CL · 10 Oct 2026, 03:50 AEST
RSI-Anything is a framework designed to enhance recursive self-improvement using predictive world models. It shows significant performance gains across various domains, including code repair and laboratory planning.
Key facts
- Recursive self-improvement (RSI) aims to transform interaction experience into persistent improvements, yet external feedback is costly and world-model predictions may become unreliable as policies evolve.
- Across diverse computational and scientific domains, RSI-Anything demonstrates substantial gains.
- In code repair, with the coder fixed, success increases from 26.3% with a static verifier to 29.2% through predictive selection and 31.2% after world-model adaptation, at comparable test usage.
- Graphical user interface (GUI) evaluations showed higher task completion, while average actions decrease by 39.4%.
- In reaction optimization, RSI-Anything achieves 96.15% mean best observed yield versus 82.16% for Bayesian optimization under the same 40-experiment budget.
Read aloud
RSI-Anything is a framework designed to enhance recursive self-improvement using predictive world models. It shows significant performance gains across various domains, including code repair and laboratory planning. Recursive self-improvement RSI aims to transform interaction experience into persistent improvements, yet external feedback is costly and world-model predictions may become unreliable as policies evolve. Across diverse computational and scientific domains, RSI-Anything demonstrates substantial gains. In code repair, with the coder fixed, success increases from twenty six point three percent with a static verifier to twenty nine point two percent through predictive selection and thirty one point two percent after world-model adaptation, at comparable test usage. Graphical user interface GUI evaluations showed higher task completion, while average actions decrease by thirty nine point four percent. In reaction optimization, RSI-Anything achieves ninety six point one five percent mean best observed yield versus eighty two point one six percent for Bayesian optimization under the same forty-experiment budget.
Sources in this cluster
- RSI-Anything: Unlocking Recursive Self-Improvement with World Models
arXiv cs.CL · 10 Oct 2026, 03:50 AEST