CLUE: Concept-Level Uncertainty Estimation for Large Language Models

TL;DR

CLUE introduces a new framework for estimating uncertainty at the concept level in large language models, enhancing interpretability and utility in tasks like hallucination detection and story generation.

Contribution

This paper presents a novel concept-level uncertainty estimation method for LLMs, addressing the limitations of sequence-level approaches and improving interpretability.

Findings

CLUE provides more interpretable uncertainty estimates than sentence-level methods.

CLUE effectively detects hallucinations in LLM outputs.

CLUE enhances the quality of story generation by identifying uncertain concepts.

Abstract

Large Language Models (LLMs) have demonstrated remarkable proficiency in various natural language generation (NLG) tasks. Previous studies suggest that LLMs' generation process involves uncertainty. However, existing approaches to uncertainty estimation mainly focus on sequence-level uncertainty, overlooking individual pieces of information within sequences. These methods fall short in separately assessing the uncertainty of each component in a sequence. In response, we propose a novel framework for Concept-Level Uncertainty Estimation (CLUE) for LLMs. We leverage LLMs to convert output sequences into concept-level representations, breaking down sequences into individual concepts and measuring the uncertainty of each concept separately. We conduct experiments to demonstrate that CLUE can provide more interpretable uncertainty estimation results compared with sentence-level uncertainty, and could be a useful tool for various tasks such as hallucination detection and story generation.