Reproduce “Contrastive Discovery: Open-Ended Scientific Discovery over Competing Explanations” 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
Ziang Liu, James J. Kim, Yijia Dai, Jennifer J. Sun
Contrastive Discovery tests scientific questions by comparing alternative explanations, eliciting beliefs before and after evidence, rather than testing a claim against an undifferentiated negation. Across six tabular observational datasets, it averaged 161.9 resolved questions among the top 200 candidates, versus 97.2 for pointwise surprisal, under the same 200-validation budget. Results use a GPT-4o-based agent scaffold and do not establish causality.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.