DERELAB: Probing Defeasible Reasoning and Confirmation Bias in LLMs with a Generative Benchmark

Jayanta Sadhu, Sayem Shahad, Kenneth Marino

Defeasible reasoning is a type of reasoning where inferences are drawn from plausible current evidence, but can be retracted upon the introduction of newer evidence. Although recen…