Reproduce “Blind Variation, Selective Retention: Improving Creative Idea Generation in AI Scientists” 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
Guowei Xu, Jiaru Zou, Mete Erdogan, Haotian Ye, Pan Lu, James Zou
BVSR reduces task-specific information during idea generation and uses outcomes from tested proposals to screen candidates before verification. At a 50% discard rate, it retains 79% of top-5% proposals by creativity and 82% by utility, separately. In SimpleTES, BVSR reaches higher final scores on all five evaluated tasks with fewer verified proposals; this does not establish elapsed-time savings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.