AI language models rely on word cues, not real causal reasoning

Original: Caliper: Probing Lexical Anchors versus Causal Structure in LLMs

Source: arxiv.org ↗