Illusions of Confidence? Diagnosing LLM Truthfulness via Neighborhood Consistency

Haoming Xu, Ningyuan Zhao, Yunzhi Yao +7

As Large Language Models (LLMs) are increasingly deployed in real-world settings, correctness alone is insufficient. Reliable deployment requires maintaining truthful beliefs under…