Stanford study finds AI chatbots struggle to distinguish belief from fact consistently 24 language models, including ChatGPT, answered 13,000 questions on beliefs and facts All models failed to identify false beliefs, with accuracy dropping significantly in tests
