RSI-Anything: Unlocking Recursive Self-Improvement with World Models
RSI-Anything uses world models to select actions and updates, improving code repair, GUI tasks, and lab planning.
RSI-Anything, built on JEPA-Anything, uses predictive world models to choose actions and learning data and retains only compatible policy-model updates. With the coder fixed, code-repair success moves from 26.3% with a static verifier to 29.2% via predictive selection and 31.2% after world-model adaptation. GUI task completion improves while average actions fall 39.4%, and laboratory-planning scores rise from 21.80% to 46.01% over five policy versions. Under a 40-experiment budget, mean best observed reaction yield is 96.15% versus 82.16% for Bayesian optimization.
- With the coder fixed, code-repair success rises from 26.3% to 31.2%.
- GUI evaluations reduce average actions by 39.4% while improving completion.
- Laboratory-planning scores rise from 21.80% to 46.01% across five versions.
- Reaction optimization reaches 96.15% yield versus 82.16% for Bayesian optimization.
- Products
- RSI-AnythingJEPA-Anything
From the original article
1 min read at huggingface.co
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. 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, Pinxian Zeng, Wanghan Xu · huggingface.co
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.