A Head to Predict and a Head to Question: Pre-trained Uncertainty Quantification Heads for Hallucination Detection in LLM Outputs

TL;DR

This paper introduces pre-trained uncertainty quantification heads for LLMs, significantly improving hallucination detection accuracy and generalization across domains and languages by leveraging Transformer architecture and attention features.

Contribution

The authors propose supervised UQ heads for LLMs that outperform unsupervised methods, enhancing hallucination detection and generalization across languages and domains.

Findings

State-of-the-art claim-level hallucination detection performance

Robustness across in-domain and out-of-domain prompts

Strong generalization to unseen languages

Abstract

Large Language Models (LLMs) have the tendency to hallucinate, i.e., to sporadically generate false or fabricated information. This presents a major challenge, as hallucinations often appear highly convincing and users generally lack the tools to detect them. Uncertainty quantification (UQ) provides a framework for assessing the reliability of model outputs, aiding in the identification of potential hallucinations. In this work, we introduce pre-trained UQ heads: supervised auxiliary modules for LLMs that substantially enhance their ability to capture uncertainty compared to unsupervised UQ methods. Their strong performance stems from the powerful Transformer architecture in their design and informative features derived from LLM attention maps. Experimental evaluation shows that these heads are highly robust and achieve state-of-the-art performance in claim-level hallucination detection across both in-domain and out-of-domain prompts. Moreover, these modules demonstrate strong generalization to languages they were not explicitly trained on. We pre-train a collection of UQ heads for popular LLM series, including Mistral, Llama, and Gemma 2. We publicly release both the code and the pre-trained heads.