Reproduce “TIGER: Learning from Researcher Taste to Generate Tailored Research Ideas” 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
Shane Gu, Yufeng Lin, Dixi Yao, Xuchen Gong, Chenhao Tan
TIGER trains individualized research-idea generators by learning each researcher’s preferences from candidate-idea ratings, then using a matched reward model during policy training without collecting ratings. The study evaluates five researchers across 400 requests. In comparisons, all five groups preferred their matched TIGER policy over GPT-4.1 in a majority of judgments. These results measure preference alignment, not scientific merit or impact.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.