Reproduce “The AI Scientist-v2: Workshop-Level Automated Scientific Discovery via Agentic Tree Search” 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
Yutaro Yamada, Robert Tjarko Lange, Cong Lu, Shengran Hu, Chris Lu, Jakob Foerster, Jeff Clune, David Ha
The authors present AI Scientist-v2, a system that turns research ideas into experiments and manuscripts without a human-authored code template. An agent generates code, branches through staged experiments, and evaluates saved experiment states. Of three selected submissions to an ICLR workshop, one passed the review-score threshold with scores of 6, 6, and 7, but no meta-review established acceptance or reliability.
Key insights
- Successful peer review demonstrates feasibility of fully autonomous scientific research
- Tree search enables more systematic experimentation through specialized nodes for different research tasks
- Vision-language models can effectively evaluate and improve research artifacts
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.