The Anatomy of Uncertainty in LLMs

2026-03-25

AI Summary

This work addresses the challenge of interpreting the sources of uncertainty in large language models (LLMs), which hinders their reliable deployment. We propose the first framework that systematically decomposes LLM uncertainty into three interpretable semantic components: input ambiguity, knowledge gaps, and decoding stochasticity—thereby transcending the limitations of the traditional aleatoric–epistemic dichotomy. Through controlled experiments and multi-dimensional uncertainty quantification across varying model scales and task settings, we analyze the relative dominance of each component. Our results demonstrate that this decomposition effectively elucidates the origins of uncertainty, substantially enhancing capabilities in hallucination detection and model reliability auditing.

Technology Category

• Machine Learning: Calibration & Uncertainty Quantification
• Reasoning under Uncertainty: Uncertainty Representations
• Natural Language Processing: (Large) Language Models

Application Category

• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs

Abstract

Understanding why a large language model (LLM) is uncertain about the response is important for their reliable deployment. Current approaches, which either provide a single uncertainty score or rely on the classical aleatoric-epistemic dichotomy, fail to offer actionable insights for improving the generative model. Recent studies have also shown that such methods are not enough for understanding uncertainty in LLMs. In this work, we advocate for an uncertainty decomposition framework that dissects LLM uncertainty into three distinct semantic components: (i) input ambiguity, arising from ambiguous prompts; (ii) knowledge gaps, caused by insufficient parametric evidence; and (iii) decoding randomness, stemming from stochastic sampling. Through a series of experiments we demonstrate that the dominance of these components can shift across model size and task. Our framework provides a better understanding to audit LLM reliability and detect hallucinations, paving the way for targeted interventions and more trustworthy systems.

Problem

• uncertainty
• large language models
• hallucination
• reliability
• interpretability

Innovation

• uncertainty decomposition
• large language models
• input ambiguity
• knowledge gaps
• decoding randomness