RSI-Anything: a general recursive self-improvement framework across domains
jiqizhixin · x · 2026-10-19
PhAI Labs presents RSI-Anything, a general recursive self-improvement (RSI) framework built on JEPA-Anything (ArXiv 2609.20800). It pairs a predictive world model with a self-improvement loop: predictions are checked against real feedback to adapt the predictor and transfer improved strategies across domains. Results: code repair success rises from 26.3% to 31.2%; 18.4% relative improvement on OSWorld GUI tasks; a 4B-parameter policy improves over five generations from 21.80% to 46.01% planning score, slightly surpassing a reference built from larger model combinations.
More from Research
- Embedding arithmetic demo: cartoon − animation − comic = diagram — mayfer · 2026-10-23
- Omnii Genome: Single Generalist Model Outperforms AlphaGenome Across All Benchmarks — JosephJacks_ · 2026-10-23
- Domingos Cites New Result Claiming RL Is 'Basically Useless' vs Supervised Learning — pmddomingos · 2026-10-23
- Word Arithmetic on Embeddings: "Bathroom Minus Toilet Equals Kitchen" Beats SAE Probing — mayfer · 2026-10-23
- Wolfram Opens Rule 30 Prizes to AIs: $30,000 for Three Open Problems — cephaloform · 2026-10-23
- Stanford and Databricks use AI agents to reproduce medical findings and test analytical robustness — ChengleiSi · 2026-10-23