Efficient semantic uncertainty quantification in language models via diversity-steered sampling

TL;DR

This paper presents a diversity-steered sampling method that improves the efficiency of semantic uncertainty estimation in large language models, especially for question answering tasks, by reducing redundant outputs and covering more semantic space with fewer samples.

Contribution

It introduces a novel diversity-steered sampler that enhances sample efficiency and uncertainty estimation in LLMs without requiring gradient access or extensive modifications.

Findings

Matches or surpasses baseline methods in QA benchmarks

Covers more semantic clusters with fewer samples

No gradient access needed for implementation

Abstract

Accurately estimating semantic aleatoric and epistemic uncertainties in large language models (LLMs) is particularly challenging in free-form question answering (QA), where obtaining stable estimates often requires many expensive generations. We introduce a diversity-steered sampler that discourages semantically redundant outputs during decoding, covers both autoregressive and masked diffusion paradigms, and yields substantial sample-efficiency gains. The key idea is to inject a continuous semantic-similarity penalty into the model's proposal distribution using a natural language inference (NLI) model lightly finetuned on partial prefixes or intermediate diffusion states. We debias downstream uncertainty estimates with importance reweighting and shrink their variance with control variates. Across four QA benchmarks, our method matches or surpasses baselines while covering more semantic clusters with the same number of samples. Being modular and requiring no gradient access to the base LLM, the framework promises to serve as a drop-in enhancement for uncertainty estimation in risk-sensitive model deployments.