Evaluation Pitfalls and Sparsity Limitations in LLM-based Confidence Estimates for Classification

• URL: http://arxiv.org/abs/2608.04899v1
• Date: Wed, 05 Aug 2026 14:24:34 GMT
• Title: Evaluation Pitfalls and Sparsity Limitations in LLM-based Confidence Estimates for Classification
• Abstract summary: Common approaches such as verbalization produce extremely sparse outputs.<n>Qwen3-32B verbalizes only eight unique confidence values on SST-2, with over half being exactly 95%.<n>We address sparsity and the best AUARC (+2.3 points over vanilla verbalization) without incurring additional inference cost.
• Score: 1.5977080543655726
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Confidence estimation is essential when LLMs are used for classification, indicating when predictions can be trusted. However, common approaches such as verbalization produce extremely sparse outputs. For instance, Qwen3-32B verbalizes only eight unique confidence values on SST-2, with over half being exactly 95%, a pattern we observe consistently across four datasets and two LLMs. Besides limiting practical utility, we show that this sparsity critically affects evaluation: the choice of interpolation in area under the accuracy-rejection curve (AUARC) dramatically alters rankings, with consistency sampling dropping from best to worst under stepwise versus linear interpolation. We advocate for standardizing stepwise interpolation for a fairer comparison. Under such a fair evaluation, we find that weighting verbalized digits by token probabilities, a method we term verbalization logprobs, addresses sparsity and achieves the best AUARC (+2.3 points over vanilla verbalization) without incurring additional inference cost.

Related papers

• Membership Inference via Pairwise Likelihood Ratios [3.0593723329631075]
  Membership inference attacks (MIAs) are the standard tool for auditing the privacy risks of machine learning models.<n>We propose Pairwise Likelihood MIA (PL-MIA), a unified method that combines a Gaussian likelihood-ratio statistic with population calibration and the Cauchy combination test.<n>We obtain $p$-values from pairwise comparisons between the query point and reference points not used for training, and aggregate these continuous signals using the Cauchy combination test.
  arXiv  Detail & Related papers  (2026-09-11T02:38:07Z)
• A Finetuned SpeechLLM for Joint Multi-Granular L2 Assessment and Natural-Language Rationales [8.300738063140129]
  We propose a rubric-guided SpeechLLM for multi-aspect, multi-granular assessment.<n>The model jointly predicts ordinal labels at the sentence-level (accuracy, fluency, prosody), word/phoneme-level accuracy, and generates a natural-language rationale in the same response.
  arXiv  Detail & Related papers  (2026-06-08T13:27:40Z)
• CauTion: Knowing When to Trust LLMs for Ensemble Causal Discovery [51.07538881798502]
  Large language models (LLMs) offer a promising source of domain knowledge to complement statistical inference.<n>We propose CauTion, a framework that reliably integrates LLM domain knowledge into an ensemble of statistical causal discovery algorithms.<n>CauTion consistently outperforms both data-centric and LLM-augmented baselines.
  arXiv  Detail & Related papers  (2026-06-02T13:07:43Z)
• Evaluating LLMs When They Do Not Know the Answer: Statistical Evaluation of Mathematical Reasoning via Comparative Signals [18.612081365101464]
  We develop a framework that combines standard labeled outcomes with pairwise comparison signals obtained by having models judge auxiliary reasoning chains.<n>Across simulations, our one-step estimator substantially improves ranking accuracy with gains increasing as model output noise grows.<n>Experiments on GPQA Diamond, AIME 2025 and GSM8K further demonstrate more precise performance estimation and more reliable model rankings.
  arXiv  Detail & Related papers  (2026-02-03T03:40:01Z)
• Dependence-Aware Label Aggregation for LLM-as-a-Judge via Ising Models [55.94503936470247]
  Large-scale AI evaluation increasingly relies on aggregating binary judgments from $K$ annotators, including judges.<n>Most classical methods assume annotators are conditionally independent given the true label $Yin0,1$, an assumption often violated by LLM judges.<n>We study label aggregation through a hierarchy of dependence-aware models based on Ising graphical models and latent factors.
  arXiv  Detail & Related papers  (2026-01-29T21:26:50Z)
• CLUE: Non-parametric Verification from Experience via Hidden-State Clustering [64.50919789875233]
  We show that correctness of a solution is encoded as a geometrically separable signature within the trajectory of hidden activations.<n>ClUE consistently outperforms LLM-as-a-judge baselines and matches or exceeds modern confidence-based methods in reranking candidates.
  arXiv  Detail & Related papers  (2025-10-02T02:14:33Z)
• Reference-Free Rating of LLM Responses via Latent Information [53.463883683503106]
  We study the common practice of asking a judge model to assign Likert-scale scores to free-text responses.<n>We then propose and evaluate Latent Judges, which derive scalar ratings from internal model signals.<n>Across a broad suite of pairwise and single-rating benchmarks, latent methods match or surpass standard prompting.
  arXiv  Detail & Related papers  (2025-09-29T12:15:52Z)
• TrustJudge: Inconsistencies of LLM-as-a-Judge and How to Alleviate Them [58.04324690859212]
  Large Language Models (LLMs) as automated evaluators (LLM-as-a-judge) has revealed critical inconsistencies in current evaluation frameworks.<n>We identify two fundamental types of inconsistencies: Score-Comparison Inconsistency and Pairwise Transitivity Inconsistency.<n>We propose TrustJudge, a probabilistic framework that addresses these limitations through two key innovations.
  arXiv  Detail & Related papers  (2025-09-25T13:04:29Z)
• The Comparative Trap: Pairwise Comparisons Amplifies Biased Preferences of LLM Evaluators [31.520403357740317]
  Large language models (LLMs) are increasingly used as evaluators for natural language generation tasks.<n>LLMs display biased preferences, such as favoring verbosity and authoritative tones.<n>We introduce PRePair, which integrates pointwise reasoning within a pairwise framework.
  arXiv  Detail & Related papers  (2024-06-18T06:43:04Z)
• TeLeS: Temporal Lexeme Similarity Score to Estimate Confidence in End-to-End ASR [1.8477401359673709]
  Class-probability-based confidence scores do not accurately represent quality of overconfident ASR predictions. We propose a novel Temporal-Lexeme Similarity (TeLeS) confidence score to train Confidence Estimation Model (CEM) We conduct experiments with ASR models trained in three languages, namely Hindi, Tamil, and Kannada, with varying training data sizes.
  arXiv  Detail & Related papers  (2024-01-06T16:29:13Z)
• Scalable Personalised Item Ranking through Parametric Density Estimation [53.44830012414444]
  Learning from implicit feedback is challenging because of the difficult nature of the one-class problem. Most conventional methods use a pairwise ranking approach and negative samplers to cope with the one-class problem. We propose a learning-to-rank approach, which achieves convergence speed comparable to the pointwise counterpart.
  arXiv  Detail & Related papers  (2021-05-11T03:38:16Z)

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.