Reproduce “RSI Arena: Towards Recursive Self-Improvement with Human Feedback” 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
Yue Huang, Yanbo Wang, Lichi Li, Yuanyuan Chen, Andreas Haupt, Xiaonan Luo, Zheyuan Liu, Jihye Choi and 10 more
RSI Arena tests whether agents can train assistants using user feedback. Ten agents started from one base model; people compared anonymous answers, and agents received votes involving them. Eight of ten improved their style-adjusted rating against a common reference, while median benchmark progress changed by two points. Stage 2 involved resources and a separate arena, so feedback’s effect was not isolated.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.