Reproduce “RSI-Master: Structuring Experiments to Guide Autonomous Model Improvement” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Yaxin Du, Xiyuan Yang, Zhifan Zhou, Yujie Ge, Cheng Wang, Jiajun Wang, Sijie Chen, Zehui Liu and 5 more
RSI-Master uses a constrained interface for data development, training, and evaluation, with persistent records linking experiments, checkpoints, results, and reviews. A research DAG lets Workers explore directions while Reviewers assess evidence and the Main Agent redirects work. On PostTrainBench, starting from Qwen3-4B-Base, it scored 54.49 versus 46.53 for Kimi Agent Swarm, with 0.0% recorded hacking under the audit definition.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.