Tag
This paper introduces InnerExpert, a method that leverages Mixture-of-Experts signals for per-token hallucination detection in LLMs, achieving superior performance over existing techniques.