Tag
Research shows that large language models' ability to confirm user beliefs depends on phrasing, with accuracy varying across epistemic expressions due to task confusion where models default to fact-checking.