RSI-Anything: Unlocking Recursive Self-Improvement with World Models
RSI-Anything uses predictive world models to recursively improve policies across code, GUI, lab, and chemistry tasks.
RSI-Anything builds on JEPA-Anything, using predictive world models to choose actions and learning data while compatibility checks decide which updated policy-model pairs are retained. With the coder fixed, code-repair success rises from 26.3% to 31.2%, and GUI evaluations reduce average actions by 39.4%. Laboratory-planning scores more than double from 21.80% to 46.01%, and reaction optimization reaches 96.15% mean best yield versus 82.16% for Bayesian optimization under 40 experiments.
- Predictive world models select actions and training data for retained updates.
- Code-repair success rises from 26.3% to 31.2% after adaptation.
- GUI runs cut average actions by 39.4% while improving completion.
- Lab-planning scores climb from 21.80% to 46.01% over five versions.
- Reaction optimization reaches 96.15% yield versus 82.16% for Bayesian optimization.
- AI models
- RSI-AnythingJEPA-Anything
From the original article
1 min read at arxiv.orgRecursive 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. We introduce RSI-Anything, a domain-general framework built on JEPA-Anything that uses predictive world models to guide action and learning-data selection.
By Taoyong Cui · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.