Revisiting Hallucination Detection with Effective Rank-based Uncertainty
This addresses the fundamental challenge of ensuring trustworthy deployment of LLMs by improving hallucination detection, representing a new paradigm in the field.
The paper tackles the problem of detecting hallucinations in large language models by proposing a method that quantifies uncertainty using the effective rank of hidden states from multiple outputs and layers, achieving effective detection and robust generalization across scenarios.
Detecting hallucinations in large language models (LLMs) remains a fundamental challenge for their trustworthy deployment. Going beyond basic uncertainty-driven hallucination detection frameworks, we propose a simple yet powerful method that quantifies uncertainty by measuring the effective rank of hidden states derived from multiple model outputs and different layers. Grounded in the spectral analysis of representations, our approach provides interpretable insights into the model's internal reasoning process through semantic variations, while requiring no extra knowledge or additional modules, thus offering a combination of theoretical elegance and practical efficiency. Meanwhile, we theoretically demonstrate the necessity of quantifying uncertainty both internally (representations of a single response) and externally (different responses), providing a justification for using representations among different layers and responses from LLMs to detect hallucinations. Extensive experiments demonstrate that our method effectively detects hallucinations and generalizes robustly across various scenarios, contributing to a new paradigm of hallucination detection for LLM truthfulness.