Tag
This paper analyzes hallucination detection in LLMs, proposing a max-pooling approach that improves efficiency by eliminating costly semantic consistency computations while maintaining competitive performance.