Better Think Thrice: Learning to Reason Causally with Double Counterfactual Consistency

Victoria Lin, Xinnuo Xu, Rachel Lawrence +4

Despite their strong performance on reasoning benchmarks, large language models (LLMs) have proven brittle when presented with counterfactual questions, suggesting weaknesses in th…