The Consistency Hypothesis in Uncertainty Quantification for Large Language Models

• URL: http://arxiv.org/abs/2506.21849v1
• Date: Fri, 27 Jun 2025 01:53:15 GMT
• Title: The Consistency Hypothesis in Uncertainty Quantification for Large Language Models
• Abstract summary: Black-box uncertainty quantification (UQ) methods, relying solely on model API access, have gained popularity due to their practical benefits.<n>In this paper, we examine the implicit assumption behind several UQ methods, which use generation consistency as a proxy for confidence.<n>We propose data-free black-box UQ methods that aggregate similarities between generations for confidence estimation.
• Score: 22.60039074743706
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Estimating the confidence of large language model (LLM) outputs is essential for real-world applications requiring high user trust. Black-box uncertainty quantification (UQ) methods, relying solely on model API access, have gained popularity due to their practical benefits. In this paper, we examine the implicit assumption behind several UQ methods, which use generation consistency as a proxy for confidence, an idea we formalize as the consistency hypothesis. We introduce three mathematical statements with corresponding statistical tests to capture variations of this hypothesis and metrics to evaluate LLM output conformity across tasks. Our empirical investigation, spanning 8 benchmark datasets and 3 tasks (question answering, text summarization, and text-to-SQL), highlights the prevalence of the hypothesis under different settings. Among the statements, we highlight the `Sim-Any' hypothesis as the most actionable, and demonstrate how it can be leveraged by proposing data-free black-box UQ methods that aggregate similarities between generations for confidence estimation. These approaches can outperform the closest baselines, showcasing the practical value of the empirically observed consistency hypothesis.

Related papers

• The Origins of Stochasticity: Comprehensive Investigations on Uncertainty Quantification for Large Language Models [12.213066436465601]
  We propose a granular uncertainty taxonomy that attributes uncertainty into input-level, parameter-level, token-level, and decoding-process sources.<n>We introduce a comprehensive evaluation framework covering diverse generation settings and metrics.<n>Our experimental results demonstrate that (i) the effectiveness of UQ methods is sensitive to task types and generation settings; (ii) consensus-based methods consistently outperform other UQ approaches; and (iii) larger model scales correlate with lower uncertainty estimates.
  arXiv  Detail & Related papers  (2026-06-22T03:05:04Z)
• Conformal Reliability: A New Evaluation Metric for Conditional Generation [62.761166941396844]
  Conditional generative models have recently remarkable success in various applications.<n>We propose a novel evaluation metric called reliability based on prediction set at a pre-specified confidence level.<n>We introduce Conformal ReLiability (CReL), a framework that can (i) construct a prediction set and (ii) accurately optimize the reliability score within the constructed prediction set.
  arXiv  Detail & Related papers  (2026-05-29T03:52:44Z)
• The Illusion of Certainty: Uncertainty quantification for LLMs fails under ambiguity [48.899855816199484]
  We introduce MAQA* and AmbigQA*, the first ambiguous question-answering (QA) datasets equipped with ground-truth answer distributions.<n>We show that predictive-distribution and ensemble-based estimators are fundamentally limited under ambiguity.
  arXiv  Detail & Related papers  (2025-11-06T14:46:35Z)
• SIMBA UQ: Similarity-Based Aggregation for Uncertainty Quantification in Large Language Models [17.805673311465295]
  Uncertainty quantification (UQ) provides measures of uncertainty.<n>Black-box UQ methods do not require access to internal model information.<n>We propose a high-level non-verbalized similarity-based aggregation framework.
  arXiv  Detail & Related papers  (2025-10-10T17:22:53Z)
• Uncertainty as Feature Gaps: Epistemic Uncertainty Quantification of LLMs in Contextual Question-Answering [29.4458902836278]
  We introduce a task-agnostic, token-level uncertainty measure defined as the cross-entropy between the predictive distribution of the given model and the unknown true distribution.<n>We derive an upper bound for uncertainty and show that it can be interpreted as semantic feature gaps in the given model's hidden representations.<n>We apply this generic framework to the contextual QA task and hypothesize that three features approximate this gap: context-reliance, context comprehension, and honesty.
  arXiv  Detail & Related papers  (2025-10-03T02:09:25Z)
• Confidence in Large Language Model Evaluation: A Bayesian Approach to Limited-Sample Challenges [13.526258635654882]
  This study introduces a Bayesian approach for large language models (LLMs) capability assessment.<n>We treat model capabilities as latent variables and leverage a curated query set to induce discriminative responses.<n> Experimental evaluations with GPT-series models demonstrate that the proposed method achieves superior discrimination compared to conventional evaluation methods.
  arXiv  Detail & Related papers  (2025-04-30T04:24:50Z)
• Uncertainty Quantification for LLMs through Minimum Bayes Risk: Bridging Confidence and Consistency [66.96286531087549]
  Uncertainty quantification (UQ) methods for Large Language Models (LLMs) encompass a variety of approaches.<n>We propose a novel approach to integrating model confidence with output consistency, resulting in a family of efficient and robust UQ methods.<n>We evaluate our approach across various tasks such as question answering, abstractive summarization, and machine translation.
  arXiv  Detail & Related papers  (2025-02-07T14:30:12Z)
• Distribution-Free Calibration of Statistical Confidence Sets [2.283561089098417]
  We introduce two novel methods, TRUST and TRUST++, for calibrating confidence sets to achieve distribution-free conditional coverage.<n>We demonstrate that our methods outperform existing approaches, particularly in small-sample regimes.
  arXiv  Detail & Related papers  (2024-11-28T20:45:59Z)
• Cycles of Thought: Measuring LLM Confidence through Stable Explanations [53.15438489398938]
  Large language models (LLMs) can reach and even surpass human-level accuracy on a variety of benchmarks, but their overconfidence in incorrect responses is still a well-documented failure mode. We propose a framework for measuring an LLM's uncertainty with respect to the distribution of generated explanations for an answer.
  arXiv  Detail & Related papers  (2024-06-05T16:35:30Z)
• Prototype-based Aleatoric Uncertainty Quantification for Cross-modal Retrieval [139.21955930418815]
  Cross-modal Retrieval methods build similarity relations between vision and language modalities by jointly learning a common representation space. However, the predictions are often unreliable due to the Aleatoric uncertainty, which is induced by low-quality data, e.g., corrupt images, fast-paced videos, and non-detailed texts. We propose a novel Prototype-based Aleatoric Uncertainty Quantification (PAU) framework to provide trustworthy predictions by quantifying the uncertainty arisen from the inherent data ambiguity.
  arXiv  Detail & Related papers  (2023-09-29T09:41:19Z)
• Advancing Counterfactual Inference through Nonlinear Quantile Regression [77.28323341329461]
  We propose a framework for efficient and effective counterfactual inference implemented with neural networks. The proposed approach enhances the capacity to generalize estimated counterfactual outcomes to unseen data. Empirical results conducted on multiple datasets offer compelling support for our theoretical assertions.
  arXiv  Detail & Related papers  (2023-06-09T08:30:51Z)
• Data Association Aware POMDP Planning with Hypothesis Pruning Performance Guarantees [7.928094304325113]
  We introduce a pruning-based approach for planning with ambiguous data associations. Our key contribution is to derive bounds between the value function based on the complete set of hypotheses and the value function based on a pruned-subset of the hypotheses. We demonstrate how these bounds can both be used to certify any pruning in retrospect and propose a novel approach to determine which hypotheses to prune in order to ensure a predefined limit on the loss.
  arXiv  Detail & Related papers  (2023-03-03T18:35:01Z)
• Logical Satisfiability of Counterfactuals for Faithful Explanations in NLI [60.142926537264714]
  We introduce the methodology of Faithfulness-through-Counterfactuals. It generates a counterfactual hypothesis based on the logical predicates expressed in the explanation. It then evaluates if the model's prediction on the counterfactual is consistent with that expressed logic.
  arXiv  Detail & Related papers  (2022-05-25T03:40:59Z)
• Meta-Learned Confidence for Few-shot Learning [60.6086305523402]
  A popular transductive inference technique for few-shot metric-based approaches, is to update the prototype of each class with the mean of the most confident query examples. We propose to meta-learn the confidence for each query sample, to assign optimal weights to unlabeled queries. We validate our few-shot learning model with meta-learned confidence on four benchmark datasets.
  arXiv  Detail & Related papers  (2020-02-27T10:22:17Z)
• Confidence Sets and Hypothesis Testing in a Likelihood-Free Inference Setting [5.145741425164947]
  $texttACORE$ is a frequentist approach to LFI that first formulates the classical likelihood ratio test (LRT) as a parametrized classification problem. $texttACORE$ is based on the key observation that the statistic, the rejection probability of the test, and the coverage of the confidence set are conditional distribution functions.
  arXiv  Detail & Related papers  (2020-02-24T17:34:49Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.