Reproduce “BabyTheorist: A Benchmark for Learning to Theorize the World from Observation Alone” 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
Doojin Baek, Junyeob Baek, Mingyu Jo, Hosung Lee, Taegu Kang, Gyubin Lee, Sungjin Ahn
BabyTheorist tests whether learners acquire reusable rules from ungrouped visual examples and apply them to new cases. It trains on unlabeled before-and-after grids generated by hidden programs, then evaluates transfer across a curriculum. Seven representative baselines stop within the first two of eight levels; NEO reaches Level 2 in Single modes but fails Level 1 in Multi modes.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.