Large Language Models Can Mislead Humans When Providing Incorrect Explanations for Claims

Large language models can provide convincing but incorrect explanations that lead humans to over-rely on them, resulting in lower accuracy compared to just showing retrieved evidence.