Na Hyun Kim

Humans and LLMs Split Theory-Building Strengths

A blinded comparison of 25 LLMs and 73 scientists finds stronger individual AI performance but more prediction-efficient, aggregable human theories.