Large language models struggle to accurately infer causal relationships from statistical correlations, even when provided with complete information about the correlations.
