RSI-Anything

Headlines · 2

1. Oct 12
   PhAI Labs releases RSI-Anything world-model framework for recursive self-improvement

   On October 12, PhAI Labs published a technical report, "RSI-Anything: Unlocking Recursive Self-Improvement with World Models," proposing a cross-domain framework in which a world model predicts the outcomes of candidate actions and real feedback is used to update both the policy and the predictor, extending its earlier JEPA-Anything prediction-learning line. Reported results include code repair on CodeContests with ReasonFlux-Coder-7B held fixed (success rate 26.3% to 31.2%), a roughly 18.4% relative gain on the OSWorld GUI benchmark, laboratory planning on LabHorizon where a 4B policy model improved from 21.80% to 46.01% over five generations, and Suzuki–Miyaura chemistry optimization reaching a 96.15% average best observed yield after 40 experiments versus 82.16% for Bayesian optimization. The report also describes an immunology workflow that built an immune world model and produced a candidate combination-intervention hypothesis involving SIRPα inhibition and IL-36γ axis modulation, which remains untested.

   机器之心 · 🔥 0
2. Oct 11
   OpenRSI Index benchmarks whether AI agents can improve real ML research

   OpenRSI Index, launched by the OpenRSI Foundation, has drawn attention on X, where its announcement post has passed 133,000 views. The benchmark draws tasks from real open-source research projects such as Marin, Qwen and GPIC, handing agents existing models, code, data and training recipes to improve, spanning pretraining, post-training and image generation, with the site listing representative single runs at roughly 5,815 to 20,864 H100-hours. Evaluation uses an RSI-Harness that separates the agent's workspace from a clean judging environment, pausing and snapshotting the workspace for scoring; advisers include Luke Zettlemoyer, Karthik Narasimhan, Yejin Choi and Dawn Song, and a companion RSI-Anything pipeline lets researchers turn their own open problems into runnable RSI tasks.

   机器之心 · 🔥 0

Experience and discussion from the community

Share my RSI-Anything experienceAsk about RSI-Anything