The Anatomy of Uncertainty in LLMs

TL;DR

This paper introduces a framework to decompose LLM uncertainty into input ambiguity, knowledge gaps, and decoding randomness, enabling better understanding, auditing, and targeted improvements for more reliable language models.

Contribution

It proposes a novel uncertainty decomposition framework that dissects LLM uncertainty into three semantic components, advancing beyond traditional single-score methods.

Findings

Uncertainty components vary with model size and task.

Decomposition helps identify sources of hallucinations.

Framework improves auditing and reliability of LLMs.

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.