Reproduce “Evolving in Thought Space: Training a Small Model at Test Time Unlocks Better Discoveries” 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

Chonghe Jiang, Ao Qu, Siyuan Liu, Ruoyun Ma, Zijian Zhou, Dingyi Zhuang, Bo Liu, Han Zheng and 4 more

Guidance-TTT trains a compact model to propose strategic changes to promising programs, while a stronger frozen model implements them and a verifier scores the results. The guide uses task and solution summaries plus scores, not source code. On Polyomino, it scored 91.89 versus CORAL’s 89.40, a reported comparison under different web-access conditions, not a matched test.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.