LLM Evaluation

LLM: Large Language Model

Follow this topic Ask about this topic

Momentum

194 papers in the last four weeks, up 87% on the four weeks before. 1.9% of all new papers.

Jul 13Week of Sep 28

Latest papers 1,631

Sep 15, 2026cs.LG

When Confidence Signals Disagree: Local and Global Confidence in Autoregressive Language Models

Modern predictive systems expose multiple quantities that are commonly interpreted as measures of confidence. However, these quantities can summarize different aspects of the predictive process. This distinction matters when confidence is used to evaluate reliability or inform downstream oversight and control. We investigate whether different confidence readouts are empirically interchangeable in an autoregressive language model by comparing local confidence, defined from the probability of the greedy-selected answer token, with global confidence, defined from modal-answer frequency under repeated sampling. Across MMLU and ARC Challenge, the two signals are weakly correlated and differ substantially in their association with correctness: global confidence is moderately associated with correctness, whereas local confidence shows little association. We further test whether question-level disagreement between the signals is associated with sampling instability. On ARC, larger local--global confidence gaps are associated with higher answer entropy, more distinct sampled answers, and lower modal-answer concentration. The gap--entropy association persists when disagreement and instability are estimated from disjoint stochastic samples, indicating that it is not explained by shared finite-sample variation. The corresponding relationship is substantially weaker on MMLU, where only 4% of questions exhibit sampling instability. These results show that confidence readouts derived from the same predictive system are not empirically interchangeable and that their disagreement can provide a diagnostic of unstable sampling behavior. Confidence should therefore be treated as an explicitly defined measurement rather than as a single intrinsic scalar property of a model, particularly when it is used to inform downstream evaluation, oversight, or control.

Julio C. Amador Diaz Lopez

SiftyML, LTD London, United Kingdom

LLM Evaluation Confidence Estimation in Language Models LLM Uncertainty Estimation

Sep 15, 2026cs.CL

Challenges of Auditing: Variability in Outputs of Large Language Models for Health

People increasingly use frontier AI models for health advice, but via different access modes (e.g., ChatGPT, ChatGPT Health, APIs) with varying settings. Here, we find systematic differences across access modes. Because evaluations typically rely on APIs while consumers interact through chatbot interfaces, these discrepancies limit evaluation validity. Our findings underscore an urgent need for model providers to enable faithful replication of consumer experiences and settings for rigorous audits.

Yuan Pu, Yewon Chang, Furong Jia +4

Department of Computer Science, Duke University, 308 Research Drive, Durham, 27708, NC. · Department of Medicine, Duke University, 40 Duke Medicine Circle, Durham, 27710, NC. · Geriatrics and Extended Care, Durham VA Health System, 508 Fulton Street, Durham, 27705, NC. +2

LLM Evaluation LLM Auditing Clinical Language Model Evaluation

Sep 15, 2026cs.CL

Competence-Preserving Resume Perturbations Expose Presentation Sensitivity in LLM Screening

Resume screeners must infer job-relevant competence from resumes whose presentation can vary substantially in wording, structure, stylistic polish, and document extraction quality. Ideally, such surface variation should not change decisions when the underlying qualification evidence is unchanged. We introduce a controlled audit of this property, constructing occupation-grounded candidate profiles at controlled competence levels and rendering each profile into multiple resume presentations. A deterministic validation gate excludes variants that alter the underlying evidence before scoring. Across six open instruction-tuned LLM conditions, we find a clear disconnect between screening validity and presentation stability. Llama-3.1-8B with its native chat template achieves the strongest validity (

0.781

) yet reverses

29.6\%

of matched pairwise decisions under competence-preserving presentation changes; Mistral-7B-v0.3 reaches validity

0.644

with a

41.4\%

flip rate. Native chat formatting improves validity for several chat-tuned models but does not remove this instability. These results show that resume-screening evaluations should assess not only whether a system identifies stronger candidates, but also whether those decisions remain stable when the same competence evidence is presented differently.

Qiangju Chen, Yang Xiao

1Macquarie University · 2The University of Melbourne

LLM Evaluation Language Model Robustness Language Model Auditing

Sep 14, 2026cs.CL

Before You Poll with LLMs: A Deliberative Diagnostic Framework

Can LLMs reason through new information like humans, or do they merely retrieve cached opinions? This is critical for silicon sampling, where LLM personas simulate public opinion at scale. Current evaluations test only whether personas hold the right opinions -- a static snapshot. But opinion research increasingly depends on dynamic fidelity: whether personas update beliefs in response to new arguments, as humans do during deliberation. No existing benchmark tests this. We introduce the Deliberative Polling Diagnostic Framework, which compares human and LLM belief shifts after identical informational interventions. Grounded in deliberative polling, it surfaces failures invisible to static evaluation: models that produce plausible partisan opinions can still misrepresent how those opinions change. Applying the framework to five frontier models using data from America in One Room (526 personas, 72 questions), we find that every model fails, each in a unique manner. GPT-5.1 exhibits reversal: its personas become more hostile toward the opposing party after balanced information, while humans become less so. This reversal is selective (80% on outgroup vs. 26% on policy questions) and symmetric across partisan identities. Gemini 2.0 Flash, Claude Sonnet 4.5, and Llama 3.3 70B exhibit overshoot, shifting correctly but at 5-7x human magnitude. DeepSeek V3 exhibits rigidity with near-zero change. Targeted ablations reveal that policy content triggers these failures and that they are identity-specific: GPT-5.1 reverses on outgroup questions but overshoots on ingroup; Gemini shows the inverse. We term this signature self-sycophancy: conformity to the model's internal stereotype of the persona rather than reasoning from the information provided. Our framework offers a concrete protocol: run the deliberative diagnostic before trusting LLM personas to mimic revised beliefs.

Ahmed Wali, Hassaan Tayyab

Lahore University of Management Sciences

LLM Evaluation Political Bias in Language Models Belief Updating in LLMs

Sep 14, 2026cs.AI

Are LLMs Good Financial User Simulators? A Preliminary Study

Large language models (LLMs) are increasingly used as user simulators, but their ability to reproduce evolving individual financial decisions remains unclear. We present a preliminary study in a controlled paper-trading environment with 120 volunteers. Participants used non-redeemable virtual funds under real-time market conditions; no real brokerage accounts, real-money positions, or real transaction records were accessed. Given only information available before a prediction cutoff, a simulator predicts the participant's next-trading-day action, traded security, and transaction quantity. We evaluate temporally aligned rolling predictions and compare settings with and without point-in-time market information. Market context improves action and ticker prediction in the controlled ablation, while transaction sizing remains difficult. We also observe systematic behavioral compression: models overproduce hold actions, underpredict sell decisions, and simplify multi-security transactions. These results provide an initial empirical characterization and motivate larger-scale evaluation of individual, temporal, and portfolio-level behavioral fidelity.

Jiajie He, Jiangyuan Hong, Dongling Ni +2

1Hithink Research · 2Nanyang Technological University · 3McMaster University

LLM Evaluation Human Behavior Simulation User Simulation Financial Decision-Making

Sep 14, 2026cs.CL

Can We Trust the Judges? Validation of Factuality Evaluation Methods via Answer Perturbation

Evaluating the factual correctness of large language models (LLMs) is vital for many applications. But are our evaluation tools themselves trustworthy? Despite the rise of factuality-based metrics, their sensitivity and reliability remain underexplored. This paper introduces a meta-evaluation framework that systematically tests these metrics using controlled corruptions of gold standard answers. Our method generates ranked outputs with known degrees of degradation to probe how metrics capture nuanced changes in truthfulness. Our experiments reveal that pipeline-based methods, such as the RAGAS's factual correctness metric, better track degradation than LLM-as-judge approaches. We also propose a new variant of the factual correctness metric that provides a competitive and cost-efficient.

Sarra Gharsallah, Adele Robaldo, Mariia Tokareva +5

Data Science Department, EURECOM, Campus SophiaTech, Biot, France

LLM Evaluation LLM-as-a-Judge Factual Consistency Evaluation

Sep 14, 2026cs.CL

Authorship attribution and aesthetic evaluation of AI poetry: a case study with Haiku

This paper investigates the generation and human evaluation of Japanese haiku by contemporary Large Language Models (LLMs), focusing on authorship perception and aesthetic judgment within a constrained poetic form. Using a few-shot prompting strategy, Japanese haiku were generated across a heterogeneous set of large language models, including open- and closed-source systems, medium-scale and large-scale architectures, models with native or adapted Japanese support, and multilingual proprietary models. These AI-generated haiku were combined with human-written ones and presented in a questionnaire distributed to students at Japanese universities in Tokyo. The survey assessed whether respondents could distinguish between AI-generated and human-written haiku and which cues informed their judgments. Recognition accuracy varied across models. GPT-5, Gemini 2.5, and StableLM-7B performed at approximately chance level (approx 0.50), whereas LLM-JP, Gemma-2B, and LLaMA-2 showed moderate detectability (approx 0.59-0.67). However, recognition was strongly item-dependent. Ratings of fluency, coherence, poeticness, and related aesthetic dimensions predicted perceived humanness but not correct classification, indicating an attribution bias linked to aesthetic evaluation and revealing a dissociation between aesthetic evaluation and true authorship detection. The extended analysis additionally examines generation-constraint adherence, participant-level characteristics, and exploratory LLM-based evaluations of haiku authorship. Overall, the findings suggest that as LLMs improve, surface-level creative plausibility may reduce reliable human discrimination within constrained poetic settings.

Livia Oddi, Simone Scardapane, Toru Sugimoto +1

DIET, Sapienza University of Rome, Rome, Italy · Shibaura Institute of Technology, Tokyo, Japan · DIAG, Sapienza University of Rome, Rome, Italy

LLM Evaluation AI-Generated Text Detection Text Generation

Sep 14, 2026cs.CL

Temperature Fragility and the Conditional Benefits of Truncation Sampling

Large language models generate text by sampling each token from a predicted distribution, and a temperature parameter sets how far the draw strays from the most probable tokens. Truncation samplers such as top-p and min-p discard the least probable tokens before the draw, so that sampling at high temperature stays coherent. Their reported accuracy gains come from temperatures of 1.5 to 3, while the defaults of deployed systems cluster between 0.6 and 1.0. Whether they change accuracy at those defaults, and for which models, has not been measured. We test thirteen open-weight models on GSM8K and MMLU-Pro at temperatures 0.7, 1.0, and 1.3 in one controlled pipeline, ten of them under eight decoding configurations. Six of the thirteen models lose 17 to 38 accuracy points on MMLU-Pro between 0.7 and 1.3, and the other seven lose at most 10. The lost accuracy comes from generations that run to the token limit or never state an answer. These results suggest that truncation samplers improve accuracy primarily when higher temperatures substantially degrade model performance. Where accuracy remains stable across temperatures, none of the tested truncation samplers improves on plain temperature sampling.

Francesco La Rosa

University of Edinburgh

LLM Evaluation Language Model Decoding

Sep 14, 2026cs.CL

Turkish MMLU Pro: Traceable Option Augmentation and Its Validity Limits in Turkish Multiple-Choice Evaluation

Adding answer options can lower multiple-choice scores without improving assessment validity. Turkish MMLU Pro examines this distinction using 12,000 Turkish-source questions across 58 sections. Each question retains its stem, five original options and source key, and receives five options copied from other questions in the same section. Sentence-embedding retrieval proposes candidates; a language model selects existing identifiers. Deterministic verification reconstructs all 60,000 additions. A 25-model calibration exposes scoring and generation-budget effects. Five evaluations produce source-key accuracies of 34.8%-81.4%. On 981 shared questions, one API-served model falls from 93.7% with five choices to 83.1% with ten; 102 of 115 lost correct responses select borrowed options. The decrease is 24.4 percentage points on heuristically flagged negative stems and 5.9 points elsewhere. A completed human-checked audit of 200 sampled questions, with undocumented reviewer tool use, yields 47 and 31 multiple-answer judgments across the two record sets, 25 of the latter unresolved. These records support concern about ambiguity, while their dependence and incomplete reviewer-method documentation limit validation. Because order and labels also change, the paired comparison measures augmentation as implemented. The contribution is a traceable construction and an analysis of its validity limits, not evidence that lower ten-choice scores measure knowledge better.

M. Ali Bayram

Department of Computer Engineering, Yıldız Technical University İstanbul, Turkey

Multilingual Language Model Evaluation LLM Evaluation Benchmark Construction Multiple-Choice Question Answering

Sep 14, 2026cs.AI

Empirical Evaluation of Open-Source Large Language Models for Retrieval-Augmented Generation in ESG Domain

Environmental, Social, and Governance (ESG) reporting is critical for corporate accountability, with Large Language Models (LLMs) and Retrieval-Augmented Generation (RAG) offering strong potential to automate KPI extraction. However, open-source LLM performance in domain-specific ESG tasks remains insufficiently understood. This paper evaluates open-source LLMs in ESG contexts using a structured framework and evaluation resource based on 498 real-world ESG reports from EU-listed companies (2010-2024). We evaluate seven open-source models (2B to 30B parameters) -- glm-4.7-flash, nemotron-3-nano:4b, qwen3:4b-instruct, gemma3:4b, gemma4:e4b, gemma4:e2b, and ministral-3:8b -- using 100 persona-based synthetic QA pairs covering ESG information needs. System performance is assessed via RAGAS metrics, including contextual recall, precision, relevance, faithfulness, answer relevancy, and factual correctness. Results show notable performance variations across architectures. Retrieval performance is strong across models (context recall around 0.58-0.61, context precision around 0.78-0.81, context relevance 0.965-0.985). Generation diverges most on faithfulness (0.607-0.822) and least on answer relevancy (0.760-0.881): glm-4.7-flash leads in faithfulness (0.822), qwen3 in factual correctness (0.449), and ministral-3 in answer relevancy (0.881). Low overall factual correctness (0.387-0.449) highlights the need for domain-specific fine-tuning. This work provides data-driven guidance for deploying open-source models in ESG reporting.

Motaz Saad, Anna Borrelli, Ivan Gentile +3

University of Salento · University of Naples Federico II · IFAB Foundation

LLM Evaluation LLM Grounding LLM Information Extraction RAG Evaluation

Sep 14, 2026cs.CL

SALUTE: Benchmarking and Adapting LLMs for the Defense Domain

Defense is a knowledge-intensive domain that requires precise understanding of specialized terminology, doctrinal concepts, operational procedures, and evolving military events. Although recent work has explored language technologies for military applications, existing efforts remain fragmented: they are often task-specific, rely on limited adaptation pipelines, or lack comprehensive defense-domain evaluation. In this paper, we present SALUTE, an end-to-end framework for benchmarking and adapting LLMs for the defense domain. SALUTE integrates Salute-Corpus, a curated corpus from open-access U.S. military doctrine and government documents; Salute-Conv, a grounded instruction dataset from doctrinal sources and decade-long defense news; Salute-Pref, a defense-aware preference dataset; and Salute-Bench, a rigorously filtered benchmark for evaluating defense-domain understanding and reasoning over doctrine and defense news. Based on these resources, we train Salute-LLM through multi-stage post-training with continual pretraining, supervised fine-tuning, and preference alignment. Extensive experiments show that Salute-LLM achieves strong defense-domain performance while retaining competitive general capabilities, demonstrating the effectiveness of SALUTE as an end-to-end framework for defense-domain LLM adaptation.

Hyeongcheol Park, Sumin In, Suyeon Myeong +5

Korea University, Republic of Korea · Sungkyunkwan University, Republic of Korea · Hanwha Systems, Republic of Korea

LLM Evaluation Domain-Adaptive Pretraining Domain Adaptation for LLMs

Sep 14, 2026cs.LG

Can We Trust LLM Judges: A Study of Capability-Dependent Biases and Multi-Judge Ensemble for Bias Calibration

LLMs are increasingly used as automated judges for model training and evaluation, yet individual judges exhibit systematic biases that undermine reliability. Much of prior work has studied biases in pairwise LLM-as-a-judge settings; in this paper, we focus on absolute scoring tasks, which mirror more realistic use cases. Across four benchmarks and six models (36 judge-examinee pairs), we show that a model's task accuracy strongly predicts its judging accuracy (Pearson

r \geq 0.90

on most models) and inversely predicts its directional bias (

r \leq -0.83

), but that accuracy alone does not ensure fair evaluation: more capable examinee models consistently receive more lenient judgments from all judges (

r \geq 0.83

). To address this, we propose calibrated weighted majority voting (WMV), an ensemble evaluation method that aggregates multiple LLM judges weighted by online estimates of their false-positive and false-negative rates. We introduce a disagreement-based estimator that derives these error rates purely from inter-judge agreement patterns, requiring no ground-truth labels or task metadata. In a simulated experiment with shifting task distributions, our label-free WMV tracks an oracle with perfect error-rate knowledge to within 0.5 percentage points on average, outperforming both individual judges and unweighted majority voting. These results demonstrate that principled multi-judge calibration can simultaneously improve accuracy and correct for systematic leniency without requiring labeled data, offering a scalable path to reliable automated evaluation as model capabilities increase.

Gemma Zhang, Prachi Badarayani, Asmi Kumar +2

Microsoft · Massachusetts Institute of Technology

LLM Evaluation LLM-as-a-Judge Language Model Calibration LLM Reliability Language Model Bias Evaluation Bias in LLM-as-a-Judge+1

Sep 14, 2026cs.CL

ORQA: An Occupation-Realistic Question and Answer Framework for LLM Professional Knowledge

We present ORQA, a method for testing occupation-level knowledge in large language models. Prior methods either map abstract LLM skills to occupations via task definitions or utilize expert knowledge which is difficult to obtain at scale and expensive. ORQA complements both of these methods by connecting O*NET occupations to trusted occupation-specific websites (such as regulatory agencies, licensing bodies, professional organizations, and government publications) and converting these into source-traceable question-answer pairs. A combination of an automated pipeline and human review produces a set of high quality questions about occupations. The question set created via our method covers 116 occupations from all 21 major groups in the SOC, with 480 questions sourced from 187 different websites. Each question is designed to probe a real-world skill question that is relevant to the occupation in question. We test 15 state-of-the-art frontier and open-weight models via this method. Claude Opus 4.6, GPT-5.4 and Claude Sonnet 4.6 all perform the best at approximately 58-62% while smaller open-weight models achieve approximately 33-41% performance. Performance varies significantly across occupations. Healthcare-related occupations achieve the highest performance (78%) while Office and Administrative Support achieve approximately 40%. Performance on individual occupations (e.g. Sheet Metal Workers and Fish and Game Wardens) is essentially zero. We also find that open-ended questions and weighting by wage bill do not significantly affect the ranking of models on this benchmark. We believe that leveraging existing trusted occupation-specific information to test LLM knowledge in professional domains may be a scalable and useful method for evaluating occupation-level AI performance in the future. Results and data are available at orqabench.org.

Shreyas Krishnan, Serina Chang, Abhishek Nagaraj

University of California, Berkeley · National Bureau of Economic Research (NBER)

LLM Evaluation Factual Knowledge in Language Models Knowledge-Intensive QA

Sep 14, 2026cs.CL

Zipbench: Low-Cost Framework for Compressing Comprehensive Benchmarks of Large Language Models

Comprehensive benchmark suites are essential for improving large language models (LLMs), but many widely used benchmarks are redundant, making evaluation unnecessarily expensive. Although recent benchmark compression methods (BCMs) can mitigate this cost, many strong BCMs rely on large collections of per-sample evaluation results from numerous LLMs to identify representative samples. Building such collections is also expensive unless they are already public, making these methods difficult to extend to newly released benchmarks. To address this challenge, we present ZipBench, a simple and low-cost BCM with theoretical error and rank-consistency guarantees. ZipBench evaluates only a small set of anchor LLMs, synthesizes pseudo evaluation results to broaden coverage, learns compact sample representations, and selects a small yet representative subset. Building on it, we create ZipBench Zoo, a collection of compact versions of 100+ benchmark proxies spanning text, multimodal, and agent tasks. These benchmark achieve mean absolute errors of 0.002--0.02 and average Spearman correlations of ~0.98 with the full benchmarks. Overall, ZipBench reduces the cost of both LLM evaluation and compact benchmark construction, lowering the barrier to broad LLM research for compute-constrained researchers. The code has been released in https://github.com/MilkThink-Lab/ZipBench.

Zhongzhan Huang, Junxin Li, Guoming Ling +3

Bosch Research · Sun Yat-sen University

Sep 14, 2026cs.CL

Judging by the Cover: Cleaning LLM Truthfulness Benchmarks to Avoid Surface-Level Feature Leakage

Binary-choice truth benchmarks ask models to choose between a correct and an incorrect answer, but if the two answers differ systematically in surface-level features, models can exceed chance without performing the intended reasoning. We show that this failure mode is detectable and can be exploited by downstream classifiers. In TruthfulQA, a simple six-feature logistic classifier achieves substantial accuracy in separating correct from incorrect answers. We further show that similar surface-level artifacts are present in additional benchmarks. To counteract this, we developed a general mechanism to clean them by removing the most leakage-reinforcing pairs. We release a version of TruthfulQA with surface-feature leakage reduced close to chance and provide a mechanism, Audit-Prune, so that the datasets can be cleaned before release.

Foad Namjoo, Remy Ogasawara, Amirali Abdullah +3

University of Utah · Thoughtworks · University of Massachusetts Amherst +1

Data Leakage LLM Evaluation Question Answering LLM Honesty

Sep 14, 2026cs.AI

How Good Are Frontier Models at Physics? Expert Re-Grading Reveals Broken Evaluations and Near-Saturation of Leading Benchmarks

Low reported scores on leading physics benchmarks, including those featured in the Artificial Analysis Intelligence Index (2026), suggest that frontier language models still struggle with advanced physics, a demanding test of their scientific reasoning and quantitative problem-solving abilities. Yet this impression does not always align with domain experts' experiences using these models in their work. We revisit these reported findings by evaluating frontier models on six widely used physics benchmarks and auditing them with experts, focusing on text-only problems with verifiable final answers. For each subfield of physics, faculty and graduate researchers with relevant expertise carefully review problem statements, reference solutions, and model responses to distinguish genuine model errors from grader errors, incorrect reference solutions, and ambiguous or underspecified questions. Most audited cases initially evaluated as incorrect reflect these benchmarking issues rather than errors in the models' physics reasoning. We then ask experts to address these benchmarking issues by correcting erroneous reference solutions and repairing or excluding flawed questions. We find that GPT-5.6-Sol's measured mean@4 rises from 47.3% to 78.7% on HLE-Physics and from 61.0% to 87.2% on CMT-Benchmark, while its corrected pass@4 reaches 94.4% on the 54 retained CritPt challenges. Corrected scores are computed on the retained evaluation subsets following expert review. Scores on the audited subsets of UGPhysics, PRISM-Physics, and PHYBench also rise substantially after correction. These findings suggest that current benchmarks substantially understate frontier models' ability to solve well-posed physics problems. Near-saturation on these closed-ended tasks highlights the need for more demanding, expert-validated evaluations.

Ali Ansari, Haoran Sun, Andy Zeyi Liu +48

LLM Evaluation Physical Reasoning Benchmark Validity

Sep 14, 2026cs.CL

Biomedical Reference Generation Remains Unreliable across 26 Large Language Models

Background. Large language models are increasingly used to help write biomedical text but may fabricate references to nonexistent work. How often large language models do so is not well characterized. Methods. We prompted 26 language models from eight developers (2023 to 2026) to supply a missing reference for each of 69 biomedical passages across ten domains. References were classified as verifiable (real paper with a resolving identifier), partial matches (real paper without a resolving identifier), fabricated (no matching indexed paper), or declined (the model refused to supply a reference). A reference was considered correct in every evaluated bibliographic field only when it was verifiable and its journal, year, and listed authors matched those of the cited paper. Results. Fabrication ranged from 10.2% (Claude Opus 4.8, which declined 52.1% of prompts) to 98.4% (Ministral 3B, which produced no verifiable reference). Claude Opus 4.6 and Claude Sonnet 4.5 produced similar proportions of verifiable references (77.6% and 76.6%) but named authors correctly in 78.7% and 28.7% of author-evaluable verifiable references, respectively, and were correct in every evaluated field in 54.6% and 19.9% of responses. GPT-5.5 was correct in every field in 48.1%. Across all models, 55.4% of responses were fabricated and 14.9% were correct in every field. Among the five tested models first released in 2026, the corresponding proportions were 35.3% and 31.8%, respectively. Conclusions. Fabrication remained common, and no model was correct in every evaluated bibliographic field in more than 54.6% of responses. Models that identify real papers may still misstate their metadata, so references produced with model assistance require verification before use.

Maxim Topaz, Zhihong Zhang, Nir Roguin +3

School of Nursing, Columbia University, New York, NY, USA · Data Science Institute, Columbia University, New York, NY, USA · VNS Health, New York, NY, USA +4

LLM Evaluation Hallucination in Language Models LLM Reliability Text Generation

Sep 14, 2026cs.CL

One Example Is Enough to Pass Fairness Benchmarks: Rethinking Fairness Evaluation for Aligned LLMs

Warning: This submission studies stereotypes and biases, and contains toxic and offensive examples, used for illustration purposes only. Fairness benchmarks such as BBQ have become the de facto standard for fairness evaluation across major model families. We argue that these benchmarks are too easy to support their role: training Qwen 2.5 7B Base with Group Relative Policy Optimization (GRPO) on a single BBQ example, or placing that example in context as a one-shot demonstration for in-context learning (ICL), lifts mean BBQ accuracy from 79.9% to 92.9% and 99.0%, respectively, closing 80% of the gap to its large-scale RLHF counterpart (96.1%) with GRPO, and surpassing it with ICL. These effects generalize across model families. A cross-conditioning analysis shows the improvement is carried by the reasoning traces generated by the model, and one example suffices to elicit a category-agnostic ``missing evidence'' reasoning pattern. We argue that BBQ-style multiple-choice abstention benchmarks measure a single structural cue, and a model that solves them does not thereby become fair. We call for evaluation suites that cover a broader spectrum of fairness alignment.

Naihao Deng, Samee Arif, Shuaichen Chang +2

University of Michigan · The Ohio State University · University of Aberdeen +1

LLM Evaluation LLM Alignment Social Bias in Language Models Language Model Bias Evaluation

Sep 13, 2026cs.CL

A primer on evaluation methods for large language models in healthcare

Large language models (LLMs) have a growing range of applications in medicine, and their evaluation is critical for ensuring they provide benefit and not harm. This evaluation can be more challenging than traditional machine learning for many reasons, including probabilistic and open-ended outputs, and behavior that shifts with prompt design and accumulated context. This review covers four key areas of LLM evaluation: principles of study design, statistical methods, capability evaluation and clinical context evaluation. Capability evaluation considers different benchmarks, including multiple-choice, agentic and multi-turn benchmarks, alongside operational metrics like token usage. Clinical context evaluation addresses establishing accuracy of free text outputs, such as human review and LLM-as-a-judge, and clinical trial approaches. Across sections, we describe underlying concepts and potential pitfalls, while emphasizing the importance of aligning evaluation methods with the research question. Together, this article aims to provide a pragmatic basis for designing and executing rigorous evaluations of healthcare LLMs.

Suzannah E McKinney, Phuc Vu, Samuel A Justice +7

Mass General Brigham AI, Mass General Brigham, Boston, MA, USA · Generative AI Laboratory, School of Informatics, University of Edinburgh, Edinburgh, UK · University of Edinburgh Centre for Cardiovascular Science, Edinburgh, UK +5

LLM Evaluation Clinical Language Model Evaluation Medical LLMs

Sep 13, 2026cs.AI

Crypto Accounting Bench: Evaluating Frontier and Open-Weight Models on Crypto-Asset Accounting Tasks

We introduce Crypto Accounting Bench (CAB), a benchmark for assessing whether frontier and open-weight language models can reconstruct the complete journal entry that an organization actually posted for a crypto-asset transaction. CAB contains 118 evaluation tasks drawn from 7 pseudonymized organizations. Each task combines transaction mechanics, asset quantities and base-currency values, wallet and legal-entity context, counterparty evidence, related transaction legs, recurrence, tax-lot evidence, and the organization's complete chart of accounts. The target is a balanced structured entry with every required account, side, amount, currency, and full-precision asset quantity. We evaluate 12 models spanning proprietary frontier systems and open-weight releases over 3 independent attempts per task, producing 4,248 trajectories. We report 3 metrics: Mean Score, Best@3, and Pass@3. Pass@3 is the fraction of tasks with at least 1 of 3 attempts that satisfies every rubric criterion and required gate. The leading model reaches 77.43% Mean Score, while the best Pass@3 is 56.78%. Deterministic diagnostics, read from each task's best of 3 attempts and macro-averaged across the 12 models, show higher base-amount agreement (97.8%) than deciding-account accuracy (56.3%). Together with the failure analysis, these results identify account selection and complete-entry composition as the main remaining challenges on CAB.

Kareem Khattab, Omar Khattab, Mohamed Ibrahem

Entendre Finance

LLM Evaluation Text Generation

Sep 12, 2026cs.CL

Measuring the Creativity of Frontier LLMs in Automated Research

Frontier LLMs are increasingly capable of conducting automated research, yet their creativity in this setting has not been systematically evaluated. We propose a set of metrics to evaluate creativity along the two dimensions of valueness and novelty. Valueness assesses whether each proposed idea is useful, while novelty is evaluated from three perspectives: whether the same idea has appeared before (Exact-Match P-Novelty), whether the modified variable or variable combination has been explored before (Variable-level P-Novelty), which reflects the breadth of research-space exploration, and whether the proposed idea is explicitly attributed to external knowledge in the model's reasoning (H-Novelty). Our evaluation shows that the models achieve relatively similar Valueness and Exact-Match P-Novelty scores, while differing substantially in Variable-level P-Novelty. H-Novelty is also consistently high among the models for which it can be evaluated. Notably, further correlation and idea-level performance analyses reveal a strong positive correlation between Variable-level P-Novelty and research performance.

Yiheng Zhao, Mengzhuo Chen, Chengming Hu +3

Concordia University, Montreal, Canada · Independent Researcher · Mcgill University, Montreal, Canada +2

Creativity Assessment LLM Evaluation Scientific Hypothesis Generation Research Idea Generation Large Language Model-Guided Scientific Discovery

Sep 12, 2026cs.AI

IBBench-Light: A Paired Evaluation of Task-Conditioned Responses to External Directives

An external record may contain a procedure to apply or text to read, depending on the user's request. IBBench-Light tests both uses against the same record. Twelve semantic bases yield 144 matched pairs per model; four quantized instruction models produced 1,152 archived greedy responses. Paired exact-contract accuracy (PECA) requires both members to satisfy their output contracts. Qwen succeeds on 132 execute and 109 process prompts, but only 97 complete pairs, showing what marginal averages omit. We audit literal-target exposure and case normalization, then add 1,722 logged CPU generations to test directive-absent controls, twelve additional semantic bases, within-base wording changes, and generation stopping. In the pinned Phi rerun, changing the end-of-sequence (EOS) set changes exact paired success from 0/144 to 62/144. A bounded IHEval comparison uses the same SmolLM2 checkpoint and output budget while preserving its published instruction roles and scorer. The benchmark measures conditional task and output-contract success. Its task margins and paired count need to be read together with the stopping policy.

Kainan Zhou, Zhaoyi Li, Janet Sung +2

Google LLC Mountain View, CA, USA

LLM Evaluation Benchmark Design Instruction Following

Sep 12, 2026cs.CL

Does Linguistic Structure Enrichment Enhance Coherence Assessment? Not With Current Architectures

Recent advances in large language models have transformed human-computer interaction. Despite their fluency, these models often produce texts that are grammatically correct but semantically incoherent, containing contradictions or disruptions in logical flow. This work investigates whether enriching text with syntactic and rhetorical information can improve incoherence prediction. Our experiments and analysis show that plain texts achieved higher accuracy because the added information was structurally and syntactically incompatible with the language model's architecture. Additionally, to demonstrate the practical importance of coherence assessment, we performed zero-shot experiments on a Brazilian disinformation dataset, suggesting that textual coherence can serve as a proxy for detecting misleading content. Code and models are available at https://github.com/ittozzamV/cohereclassifier.

Victor Mazzotti, Luiz Pereira, Marina Bitencourt dos Santos +4

Instituto de Matemática, Estatística e Computação Científica (IMECC) · Instituto de Computação (IC) · Instituto de Estudos da Linguagem (IEL) +1

Sep 12, 2026cs.AI

NovGauge: A Fine-Grained Benchmark for Diagnosing LLMs' Capability in Paper Novelty Assessment

Large language models (LLMs) are increasingly used in peer review at major AI conferences, yet novelty remains a persistent weak point. Existing benchmarks assess novelty as a single holistic score, making it difficult to diagnose which dimension a model misjudges or whether its evidence is faithful. We present NovGauge, a human-anchored benchmark for fine-grained novelty assessment diagnosis. The benchmark contains 619 paper pairs and 50 multi-paper sets, drawn from two expert sources: ICLR reviewer overlap claims and survey co-citations. Instances are independently labeled along three dimensions: task, problem, and method, capturing application goals, technical challenges, and solution approaches. We propose a cascading diagnostic pipeline that verifies per-dimension correctness, evidence grounding, and logical support. Evaluation of 18 LLMs shows hallucination rates ranging from 0% to 39% across dimensions, and among non-hallucinated correct-positive judgments, over 70% cite evidence fails to logically support the stated reason. The best-performing model, GPT-5.5, achieves 43-72% Verified F1 across dimensions, while most models retain less than half of their raw F1 after faithfulness verification. These results suggest that current LLMs remain far from reliable scientific novelty assessment, particularly when correctness is conditioned on faithful evidence grounding.

Guoqiang Zhang, Kexin Tan, Ming Zhang +12

College of Computer Science and Artificial Intelligence, Fudan University, Shanghai, China · Graduate School of Engineering, The University of Tokyo, Tokyo, Japan · AtomInnoLab 4Atom Infinite Pte. Ltd., Singapore +1

LLM Evaluation LLM Answer Verification LLM Hallucination Detection

Sep 12, 2026cs.CL

Recognizing Is Not Reversing: A Controlled Inversion Test of Fact-Preserving News Framing

Large language models (LLMs) are increasingly used to analyze and rewrite news, yet current framing studies mainly evaluate generation, detection, or whether rewritten text appears more neutral. They do not directly show whether a model can undo a known framing transformation while keeping the facts fixed. We introduce a controlled inversion test over three established textual realizations of framing: evaluative lexis, agency realization, and information salience. Across 60 news articles and three intervention strengths, this yields 540 paired variants with preserved atomic facts and recorded edits. Across Qwen, DeepSeek, and Kimi, factual preservation remains near 0.84, whereas intervention reversal is 0.044--0.068. Even when both framing type and direction are recognized correctly, pooled reversal reaches 0.071. These results reveal a clear separation between factual fidelity, framing recognition, and framing inversion: recognizing how an article is framed does not imply that the framing can be undone.

Yi Liu

University of Science and Technology of China

LLM Evaluation Framing Effects in Language Models

Sep 11, 2026cs.CL

Rethinking Verbalized Confidence for LLM-as-a-Judge: A Compatibility Shift on Post-2025 Proprietary Models

Verbalized confidence, long dismissed as overconfident, coarse, and prone to round-number clustering, is now the more robust soft-scoring mechanism for LLM-as-a-Judge on top-tier proprietary models. Across SummEval, AggreFact, and HelpSteer2, spanning up to 18 LLMs, we show that the standard advice to prefer log-probabilities no longer holds on post-2025 models, where verbalized confidence is the better signal. We call this a compatibility shift. On top of a standard verbalized-confidence baseline, we introduce two new ingredients: an overconfidence advisory and self-debate. Together they improve calibration, score-distribution spread, and robustness to task subjectivity. We further observe a generation effect: post-2025 models accommodate these two additions with little balanced-accuracy cost, whereas pre-2025 models pay a measurable penalty. Compared with logprob-based G-Eval, verbalized confidence is the more subjectivity-robust soft signal on GPT-family top-tier releases. The shift is invisible under accuracy-only reporting. Rather than defaulting to hard predictions, we recommend broader use of soft scoring in LLM-as-a-Judge. More broadly, verbalized confidence has moved from a weaker substitute for logprobs to a practical soft-scoring mechanism for contemporary LLM judges.

Yu-Chung Hsiao

Cisco Systems

LLM Evaluation LLM-as-a-Judge Confidence Estimation in Language Models Language Model Calibration Language Model Robustness Metacognition in Language Models

Sep 11, 2026cs.CL

The widening evaluation gap in medical large language model research 2023 to 2026

Large language models are superseded every few quarters; clinical evidence takes years. We asked whether medical research is keeping pace with the systems it evaluates. PubMed returned 11,628 records for January 2023 to June 2026 across fourteen clinical domains, growing 45-fold; 2.5% used a randomised, controlled or prospective design. Evaluation lag, from a study's newest named model release to its own publication, widened from 1.33 to 6.08 quarters. Because discontinued models age mechanically, we benchmarked this against a counterfactual holding model composition fixed: migration to newer systems offset only 56% of the drift (95% CI 50-65). Randomised trials evaluated models a median 4.6 quarters older than other designs (P = 3 x 10^-19), yet among studies naming a model still under development no design differed from any other; 62% of randomised trials evaluated a discontinued family. Rigour and currency are in tension, and that tension reflects model selection rather than research timelines.

Raad Bin Tareaf, Murad Al-Rajab, Samia Loucif

XU Exponential University of Applied Sciences, Data Science and Artificial Intelligence Cluster, Marlene-Dietrich-Allee 12B, 14482 Potsdam, Germany · College of Engineering, Abu Dhabi University, Abu Dhabi, United Arab Emirates · College of Technological Innovation, Zayed University, Abu Dhabi, United Arab Emirates

LLM Evaluation Clinical Language Model Evaluation Medical LLMs

Sep 10, 2026cs.CL

On the Impact of Anonymization on the Performance of Large Language Models

As large language models are increasingly deployed in sensitive domains, anonymizing input data to protect personally identifiable information has become a critical practice. However, the impact of this anonymization on model utility is not well understood. This paper presents a systematic empirical study of the trade-off between privacy and performance. We evaluate five prominent language models across eleven diverse benchmarks, comparing their performance on original versus pseudonymized inputs. Our results reveal that while anonymization generally degrades performance, the effect is highly nuanced. We find that more capable models, such as Qwen2.5-72B and GPT-4o mini, suffer the largest performance drops, suggesting a stronger reliance on specific entity information. The impact is also task-dependent: performance on TruthfulQA improves with anonymization, while retrieval-focused tasks like RGB experience a catastrophic decline. Further experiments show that reversible anonymization techniques that preserve entity uniqueness significantly outperform irreversible ones like redaction, and that explicitly prompting models about anonymization offers no discernible benefit. We conclude that anonymization is not a one-size-fits-all solution and must be co-designed with the model and task in mind to balance privacy and utility effectively. Our findings provide a crucial baseline for developing more robust, privacy-aware AI systems.

Tobias Deußer, Max Hahnbück, Lorenz Sparrenberg +3

University of Bonn, Bonn, Germany · 2Fraunhofer IAIS, Sankt Augustin, Germany · 3Lamarr Institute for Machine Learning and Artificial Intelligence, Bonn, Germany +1

LLM Evaluation Privacy-Preserving Language Models Privacy-Utility Tradeoff

Sep 10, 2026cs.AI

SemVerBench: Benchmarking LLM Comprehension of Version-Constraint Resolution Semantics

Large language model (LLM) coding agents constantly decide whether a version satisfies a constraint such as ^1.2.3 or >=2.0,<3, yet their grasp of version-constraint semantics has never been measured directly. We introduce SemVerBench, the first benchmark of LLM version-constraint resolution semantics across three ecosystems (npm, PEP 440, Cargo): 240 machine-checkable items with unique answers, built author-neutrally from four balanced sources (each ecosystem's official test suite plus three frontier LLM proposers) and labeled by a non-circular two-implementation oracle. Evaluating six frontier models, we find systematic, predictable per-mechanism blind spots: a partial-comparator carry rule (>1.2 means >=1.3.0) traps every model on Cargo (near 60%), and although standard PEP 440 prefix matching is universal, on zero-pad/post-release corner cases GPT-5.1 collapses (0/26) while Claude stays at 97-100% (verified on a 67-item oracle-validated set). Opus significantly outperforms all other models, and Sonnet outperforms the OpenAI models (McNemar). The failures look more like an activation/application gap than a knowledge gap: injecting the rule or a light correct hint recovers most errors, whereas interval decomposition does not, and models are at ceiling on the basic forms of the same rules. An author-stratified analysis finds no statistically significant self-favoritism. Because the task is verifiable and a free, 100%-correct resolver exists, tool delegation reaches ~100%: coding agents should delegate version resolution to a resolver rather than reason about versions in-head.

Qibai Chen, Zeming Liu

Independent Researcher United States · Brown University United States

Software Engineering LLM Evaluation Software Engineering Benchmarks

Sep 10, 2026cs.CL

Quantifying Logical Consistency in Transformers via Query-Key Alignment

Large language models (LLMs) have demonstrated impressive performance in various natural language processing tasks, yet their ability to perform multi-step logical reasoning remains an open challenge. Although Chain-of-Thought prompting has improved logical reasoning by enabling models to generate intermediate steps, it lacks mechanisms to assess the coherence of these logical transitions. In this paper, we propose a novel, lightweight evaluation strategy for logical reasoning that uses query-key alignments inside transformer attention heads. By computing a single forward pass and extracting a "QK-score" from carefully chosen heads, our method reveals latent representations that reliably separate valid from invalid inferences, offering a scalable alternative to traditional ablation-based techniques. We also provide an empirical validation on multiple logical reasoning benchmarks, demonstrating improved robustness of our evaluation method against distractors and increased reasoning depth. The experiments were conducted on a diverse set of models, ranging from 1.5B to 70B parameters.

Eduard Tulchinskii, Anastasia Voznyuk, Laida Kushnareva +4

Skolkovo Institute of Science and Technology · AI Foundation and Algorithm Lab · Moscow Institute of Physics and Technology +2

Transformer Interpretability LLM Evaluation LLM Reasoning