Detecting LLM Hallucination Through Layer-wise Information Deficiency: Analysis of Ambiguous Prompts and Unanswerable Questions

TL;DR

This paper introduces a layer-wise information analysis method to detect hallucinations in large language models, especially when processing ambiguous or unanswerable prompts, enhancing model reliability without retraining.

Contribution

The paper proposes a novel, test-time layer-wise information flow analysis technique to identify hallucinations in LLMs, focusing on information deficiencies across layers.

Findings

Layer-wise information deficiencies correlate with hallucinations.

Tracking cross-layer information dynamics improves detection robustness.

Method works without additional training or model modifications.

Abstract

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.