Realizing LLMs' Causal Potential Requires Science-Grounded, Novel Benchmarks

Ashutosh Srivastava, Lokesh Nagalapatti, Gautam Jajoo +3

Recent claims of strong performance by Large Language Models (LLMs) on causal discovery are undermined by a key flaw: many evaluations rely on benchmarks likely included in pretrai…