Large language models often give confident answers even when the question can’t be answered. New research shows these models ...