Detecting LLM Hallucination Through Layer-wise Information Deficiency: Analysis of Ambiguous Prompts and Unanswerable Questions
This addresses the risk of inaccurate LLM responses in safety-critical domains by providing a test-time detection method for ambiguous or insufficient inputs.
The paper tackles the problem of detecting hallucinations in large language models (LLMs) by analyzing information flow across model layers, revealing that hallucinations correlate with deficiencies in inter-layer transmissions and achieving robust detection without requiring model retraining.
Large language models (LLMs) frequently generate confident yet inaccurate responses, introducing significant risks for deployment in safety-critical domains. We present a novel, test-time approach to detecting model hallucination through systematic analysis of information flow across model layers. We target cases when LLMs process inputs with ambiguous or insufficient context. Our investigation reveals that hallucination manifests as usable information deficiencies in inter-layer transmissions. While existing approaches primarily focus on final-layer output analysis, we demonstrate that tracking cross-layer information dynamics ($\mathcal{L}$I) provides robust indicators of model reliability, accounting for both information gain and loss during computation. $\mathcal{L}$I integrates easily with pretrained LLMs without requiring additional training or architectural modifications.