Language Model Generation Evaluation

Follow this topic Ask about this topic

Momentum

30 papers in the last four weeks, up 100% on the four weeks before. 0.3% of all new papers.

Jul 13Week of Sep 28

Latest papers 209

May 14, 2026cs.CL

Dimension-Level Intent Fidelity Evaluation for Large Language Models: Evidence from Structured Prompt Ablation

Holistic evaluation scores capture overall output quality but do not distinguish whether a model reproduced the structural form of a user's request from whether it preserved the user's specific intent. We propose a dimension-level intent fidelity evaluation framework, applied here through a structured prompt ablation study across 2,880 outputs spanning three languages, three task domains, and six LLMs, that separately measures structural recovery and intent fidelity for each semantic dimension. This framework reveals a systematic structural-fidelity split: among Chinese-language outputs with complete paired scores, 25.7% received perfect holistic alignment scores (GA=5) while exhibiting measurable dimensional intent deficits; among English-language outputs, this proportion rose to 58.6%. Human evaluation confirmed that these split-zone outputs represent genuine quality deficits and that dimensional fidelity scores track human judgements more reliably than holistic scores do. A public-private decomposition of 2,520 ablation cells characterises when models successfully compensate for missing intent and when they fail, while proxy annotation distinguishes prior inferability from default recoverability. A weight-perturbation experiment shows that moderate misalignment is typically absorbed, whereas severe dimensional inversion is consistently harmful. These findings demonstrate that dimension-level intent fidelity evaluation is a necessary complement to holistic assessment when evaluating LLM outputs for user-specific tasks.

GAng Peng

Huizhou Lateni AI Technology Co., Ltd., Huizhou, China · Huizhou University, Huizhou, China

Multilingual Language Model Evaluation LLM Evaluation Language Model Generation Evaluation Instruction Following

May 14, 2026cs.LG

NodeSynth: Socially Aligned Synthetic Data for AI Evaluation

Recent advancements in generative AI facilitate large-scale synthetic data generation for model evaluation. However, without targeted approaches, these datasets often lack the sociotechnical nuance required for sensitive domains. We introduce NodeSynth, an evidence-grounded methodology that generates socially relevant synthetic queries by leveraging a fine-tuned taxonomy generator (TaG) anchored in real-world evidence. Evaluated against four mainstream LLMs (e.g., Claude 4.5 Haiku), NodeSynth elicited failure rates up to five times higher than human-authored benchmarks. Ablation studies confirm that our granular taxonomic expansion significantly drives these failure rates, while independent validation reveals critical deficiencies in prominent guard models (e.g., Llama-Guard-3). We open-source our end-to-end research prototype and datasets to enable scalable, high-stakes model evaluation and targeted safety interventions (https://github.com/google-research/nodesynth).

Qazi Mamunur Rashid, Xuan Yang, Zhengzhe Yang +5

Google Research · Google · Google Deepmind

Language Model Generation Evaluation Language Model Safety Evaluation LLM Auditing Synthetic Data Generation Synthetic Benchmark Generation Synthetic Data Generation for Language Models+1

May 13, 2026cs.CL

RTLC -- Research, Teach-to-Learn, Critique: A three-stage prompting paradigm inspired by the Feynman Learning Technique that lifts LLM-as-judge accuracy on JudgeBench with no fine-tuning

LLM-as-a-judge is now the default measurement instrument for open-ended generation, but on the public JudgeBench benchmark even strong instruction-tuned judges barely scrape past random on objective-correctness pairwise items. We introduce RTLC, a three-stage prompting recipe -- Research, Teach-to-Learn, Critique -- that promotes a single black-box LLM into an ensemble-of-thought judge with no fine-tuning, retrieval, or external tools. Stage 1 wraps the input in a fixed pedagogical scaffold porting the Feynman Learning Technique (study

\to

teach

\to

find gaps

\to

simplify) into LLM prompting. Stage 2 draws N=10 independent candidate verdicts at temperature 0.4. Stage 3 acts as its own critic, cross-comparing the candidate set against the original question to emit one critiqued verdict at temperature 0. On JudgeBench-GPT (350 hard pairwise items), Claude 3.7 Sonnet's pairwise accuracy climbs from 64.6% (single-shot vanilla prompt) to 78.6% (RTLC critique-of-10) -- an absolute 14.0-percentage-point gain. RTLC also beats N=10 self-consistency majority voting (77.7%) and a zero-shot first candidate (74.0%). A clean three-step ablation attributes +9.4 pp to the Teach-to-Learn scaffold, +3.7 pp to N=10 marginalisation, and +0.9 pp to explicit critique. We discuss the cost-accuracy frontier (RTLC sits above self-consistency at every working point), the error-budget breakdown across the four JudgeBench categories (knowledge, reasoning, math, coding), and how RTLC composes orthogonally with post-hoc judge-score calibration, with the two interventions compounding multiplicatively in practice.

Andrea Morandi

Cisco Systems, Inc.

LLM Evaluation Pairwise Comparison LLM-as-a-Judge Language Model Generation Evaluation

May 12, 2026cs.CL

Predicting Disagreement with Human Raters in LLM-as-a-Judge Difficulty Assessment without Using Generation-Time Probability Signals

Automatic generation of educational materials using large language models (LLMs) is becoming increasingly common, but assigning difficulty levels to such materials still requires substantial human effort. LLM-as-a-Judge has therefore attracted attention, yet disagreement with human raters remains a major challenge. We propose a method for predicting which LLM-generated difficulty ratings are likely to disagree with human raters, so that such cases can be sent for re-rating. Unlike prior approaches, our method does not rely on generation-time probability signals, which must be collected during rating generation and are often difficult to compare across LLMs. Instead, exploiting the fact that difficulty is an ordinal scale, we use a separate embedding space, such as ModernBERT, and identify disagreement candidates based on the geometric consistency of the rating set. Experiments on English CEFR-based sentence difficulty assessment with GPT-OSS-120B and Qwen3-235B-A22B showed that the proposed method achieved higher AUC for predicting disagreement with human raters than probability-based baselines.

Yo Ehara

Tokyo Gakugei University

Annotator Disagreement LLM-as-a-Judge Language Model Generation Evaluation Educational Assessment

May 12, 2026cs.CL

DiffScore: Text Evaluation Beyond Autoregressive Likelihood

Autoregressive language models are widely used for text evaluation, however, their left-to-right factorization introduces positional bias, i.e., early tokens are scored with only leftward context, conflating architectural asymmetry with true text quality. We propose masked reconstruction as an alternative paradigm, where every token is scored using full bidirectional context. We introduce DiffScore, an evaluation framework built on Masked Large Diffusion Language Models. By measuring text recoverability across continuous masking rates, DiffScore eliminates positional bias and naturally establishes an evaluation hierarchy from local fluency to global coherence. We further provide diagnostic tools unavailable to autoregressive frameworks: multi-timestep quality profiles that decompose scores across masking rates, and bidirectional PMI decomposition that disentangles fluency from faithfulness. Experiments across ten benchmarks show that DiffScore consistently outperforms autoregressive baselines in both zero-shot and fine-tuned settings. The code is released at: https://github.com/wenlai-lavine/DiffScore.

Wen Lai, Yingli Shen, Dingnan Jin +4

1Ant Group · 2Tsinghua University · 3Technical University of Munich

LLM Evaluation Language Model Generation Evaluation Automated Evaluation Diffusion Language Models

May 11, 2026cs.LG

VERDI: Single-Call Confidence Estimation for Verification-Based LLM Judges via Decomposed Inference

LLM-as-Judge systems are widely deployed for automated evaluation, yet practitioners lack reliable methods to know when a judge's verdict should be trusted. Token log-probabilities, the standard post-hoc confidence signal, are unavailable for many commercial LLMs and, even when accessible, saturate above 0.999 with structured JSON output. We introduce VERDI (VERification-Decomposed Inference), a method that extracts confidence from the reasoning trace a structured judge already produces, with no additional inference calls. VERDI decomposes each verification-style evaluation into sub-checks and derives three structural signals: Step-Verdict Alignment, Claim-Level Margin, and Evidence Grounding Score. We combine them with Platt-scaled logistic regression. On three public benchmarks, VERDI achieves AUROC 0.72-0.91 on GPT-4.1-mini and 0.66-0.80 on GPT-5.4-mini. On Qwen3.5-4B/9B/27B, where answer-token logprobs are anti-calibrated (higher confidence on errors, AUROC 0.32-0.49), VERDI achieves 0.56-0.70. We additionally validate on a production system with eight rubrics (AUROC 0.73-0.88 on factual rubrics), demonstrate cross-model transfer (AUROC 0.66-0.69), and show that a 33M-parameter NLI (Natural Language Inference) model provides a scalable alternative to regex extraction.

Jasmine Qi, Danylo Dantsev, Muyang Sun

Indeed Inc.

Post-Hoc Calibration LLM-as-a-Judge LLM Answer Verification Language Model Generation Evaluation Confidence Estimation in Language Models LLM Reliability+1

May 11, 2026cs.AI

LLARS: Enabling Domain Expert & Developer Collaboration for LLM Prompting, Generation and Evaluation

We demonstrate LLARS (LLM Assisted Research System), an open-source platform that bridges the gap between domain experts and developers for building LLM-based systems. It integrates three tightly connected modules into an end-to-end pipeline: Collaborative Prompt Engineering for real-time co-authoring with version control and instant LLM testing, Batch Generation for configurable output production across user-selected prompts

\times

models

\times

data with cost control, and Hybrid Evaluation where human and LLM evaluators jointly assess outputs through diverse assessment methods, with live agreement metrics and provenance analysis to identify the best model-prompt combination for a given use case. New prompts and models are automatically available for batch generation and completed batches can be turned into evaluation scenarios with a single click. Interviews with six domain experts and three developers in online counselling confirmed that LLARS feels intuitive, saves considerable time by keeping everything in one place and makes interdisciplinary collaboration seamless.

Philipp Steigerwald, Mara Stieler, Jennifer Burghardt +2

Technische Hochschule Nürnberg Georg Simon Ohm, Nürnberg, Germany · Faculty of Computer Science, Centre for Artificial Intelligence (KIZ) · Faculty of Social Sciences, Institute for E-Counselling

Language Model Generation Evaluation LLM Prompting Human-in-the-Loop Evaluation Prompt Engineering

May 11, 2026cs.CY

StereoTales: A Multilingual Framework for Open-Ended Stereotype Discovery in LLMs

Multilingual studies of social bias in open-ended LLM generation remain limited: most existing benchmarks are English-centric, template-based, or restricted to recognizing pre-specified stereotypes. We introduce StereoTales, a multilingual dataset and evaluation pipeline for systematically studying the emergence of social bias in open-ended LLM generation. The dataset covers 10 languages and 79 socio-demographic attributes, and comprises over 650k stories generated by 23 recent LLMs, each annotated with the socio-demographic profile of the protagonist across 19 dimensions. From these, we apply statistical tests to identify more than 1{,}500 over-represented associations, which we then rate for harmfulness through both a panel of humans (N = 247) and the same LLMs. We report three main findings. \textbf{(i)} Every model we evaluate emits consequential harmful stereotypes in open-ended generation, regardless of size or capabilities, and these associations are largely shared across providers rather than isolated misbehaviors. \textbf{(ii)} Prompt language strongly shapes which stereotypes appear: rather than transferring as a shared set of biases, harmful associations adapt culturally to the prompt language and amplify bias against locally salient protected groups. \textbf{(iii)} Human and LLM harmfulness judgments are broadly aligned (Spearman

ρ=0.62

), with disagreements concentrating on specific attribute classes rather than specific providers. To support further analyses, we release the evaluation code and the dataset, including model generations, attribute annotations, and harmfulness ratings.

Pierre Le Jeune, Étienne Duchesne, Weixuan Xiao +4

Giskard AI · École Normale Supérieure, INSERM, Paris

Open-Ended Generation Multilingual Language Model Evaluation Language Model Generation Evaluation Social Bias in Language Models Cultural Bias in Language Models Language Model Bias Evaluation

May 11, 2026cs.AI

Active Testing of Large Language Models via Approximate Neyman Allocation

Large language models (LLMs) require reliable evaluation from pre-training to test-time scaling, making evaluation a recurring rather than one-off cost. As model scales grow and target tasks increasingly demand expert annotators, both the compute and labeling costs needed for each evaluation rise rapidly. Active testing aims to alleviate this bottleneck by approximating the evaluation result from a small but informative subset of the evaluation pool. However, existing approaches primarily target classification and break down on generative tasks. We introduce a novel active testing algorithm tailored to generative tasks. Our method leverages semantic entropy from surrogate models to stratify the evaluation pool and then conducts approximate Neyman allocation based on signals extracted from these surrogates. Across multiple language and multimodal benchmarks and a range of surrogate-target model pairs, our method significantly improves on baselines and closely tracks Oracle-Neyman, delivering up to 28% MSE reduction over Uniform Sampling and an average of 22.9% budget savings.

Zeli Liu, Jiancheng Zhang, Cong Liu +1

University of California, Riverside

LLM Evaluation Language Model Generation Evaluation Semantic Entropy Generative AI Evaluation

May 8, 2026cs.CL

NARRA-Gym for Evaluating Interactive Narrative Agents

Interactive narrative tasks require LLMs to sustain a coherent, evolving story while adapting to a user over multiple turns. However, suitable benchmarks for this setting are limited: existing evaluations often focus on static prompts, isolated story generations, or post-hoc ratings, and therefore miss whether models can jointly manage story generation, long-context state and pacing, character simulation, empathic personalization, and story-grounded artifacts. We introduce NARRA-Gym, an executable evaluation environment that turns a sparse emotional seed into a complete interactive story episode and logs the full model-in-the-loop trajectory, including story construction, memory updates, planning, pacing interventions, and optional artifact synthesis. We evaluate nine frontier LLMs using a controlled LLM-as-judge sweep over eight benchmark personas and a human evaluation in which participants rate customized model outputs. Our results show substantial variation across models, personas, and evaluation dimensions: models that produce fluent stories can still fail on robustness, user experience, or resistance-sensitive personalization. These findings suggest that interactive narrative offers a useful benchmark for evaluating long-horizon, user-adaptive LLM behavior beyond isolated story quality.

Yue Huang, Yuchen Ma, Jiayi Ye +14

University of Notre Dame · 2LMU Munich · 11Munich Center for Machine Learning +8

Language Model Generation Evaluation Interactive Storytelling LLM Agent Evaluation Long-Horizon LLM Agents

May 7, 2026cs.LG

Delulu: A Verified Multi-Lingual Benchmark for Code Hallucination Detection in Fill-in-the-Middle Tasks

Large Language Models for code generation frequently produce hallucinations in Fill-in-the-Middle (FIM) tasks -- plausible but incorrect completions such as invented API methods, invalid parameters, undefined variables, or non-existent imports. These failures pass superficial review yet introduce runtime errors. We introduce Delulu, a verified multi-lingual benchmark of 1,951 FIM samples across 7 languages and 4 hallucination types. Samples are curated through an adversarial pipeline: a frontier LLM generates plausible hallucinations, four diverse judge models evaluate them, embedding-based clustering mines progressively harder examples, self-contained Docker containers verify that golden completions compile while hallucinated variants produce the expected runtime error, and a final human-expert review removes any remaining biased or trivially decidable samples. We evaluate 11 open-weight FIM models from five families spanning 0.5B-32B parameters: a six-point Qwen2.5-Coder scaling slate, plus a cross-family slate (CodeLlama, DeepSeek-Coder-V2, StarCoder2). The strongest model reaches only 84.5% pass@1, no family exceeds 0.77 Edit Similarity, and every family produces hallucination-aligned completions on a non-trivial share of samples, confirming that the difficulty exposed by Delulu is task-intrinsic rather than family-specific. We release the benchmark, containers, and evaluation framework at https://github.com/microsoft/delulu.

Mahdi Erfanian, Nelson Daniel Troncoso, Aashna Garg +4

University of Illinois Chicago · Microsoft

Language Model Error Detection Language Model Generation Evaluation Hallucination in Language Models Code Generation LLM Reliability LLM Hallucination Detection+2

May 7, 2026cs.CL

ReasonLab: A Controlled and Auditable Evaluation of Prompting Techniques for Multiple-Choice QA

Probing the capabilities of Large Language Models (LLMs) and building robust solutions for Multiple-Choice Question Answering (MCQA) remain central challenges in natural language understanding. Furthermore, the rapid proliferation of LLMs has created the implicit assumption that more sophisticated prompting techniques yield better performance. Several studies claim such gains, but report them under differing models, prompt wordings and answer-extraction rules, so the gains cannot be attributed to the technique alone. We address this gap with ReasonLab, an evaluation framework in which the prompting technique is a first-class experimental variable alongside the model and the dataset, and which retains every generation for inspection. Using ReasonLab we conduct a controlled study of 8 prompting techniques across 10 MCQA datasets, 27 model configurations and 480,927 evaluations at temperature 0. We find that the prompting technique is a minor determinant of accuracy: on configurations without a reasoning budget the reasoning triggers improve on direct prompting by only 3.92 to 4.69 pp and are indistinguishable from one another, and on configurations with reasoning enabled no technique differs by more than 0.51 pp. Self-Generate is the only technique with a consistent effect, a reduction of 2.95 pp. We further investigate three phenomena: (1) the comparison of models on a common set of datasets, where model size does not predict accuracy, (2) the trade-offs across thinking budgets, where enabling reasoning is worth up to 12.74 pp whereas an eightfold budget increase adds only 0.48 to 2.10 pp, and (3) the variation in dataset difficulty, with 60% of benchmarks below 70% accuracy and a 43.9 pp spread from easiest to hardest. These results suggest that, for MCQA, the prompting technique is a minor lever compared with enabling model reasoning, and that substantial headroom remains.

Inder Preet, Shuxin Lin, Dhaval Patel

IBM, Dublin · IBM, New York

LLM Evaluation Language Model Generation Evaluation Multiple-Choice Question Answering LLM Prompting Large Reasoning Models

May 7, 2026cs.AI

Beyond Fixed Benchmarks and Worst-Case Attacks: Dynamic Boundary Evaluation for Language Models

Evaluating large language models (LLMs) today rests on fixed benchmarks that apply the same set of items to any model, producing ceiling and floor effects that mask capability gaps. We argue that the most informative evaluation signal lies at the boundary, where the per-prompt pass probability is near

0.5

under random-sampling decoding strategy, and propose Dynamic Boundary Evaluation (DBE), which actively locates each model's boundary and places it on a globally comparable difficulty scale. DBE delivers three artifacts: (i) a calibrated item bank covering safety, capability, and truthfulness, with per-item difficulty labels validated across

9

reference LLMs; (ii) Skill-Guided Boundary Search (SGBS), a search algorithm that finds boundary items for a given target LLM using only API-level query access; and (iii) an evaluation protocol that places a new LLM on a unified ability scale and grows the evaluation set adaptively when the target falls outside the bank's coverage. We instantiate DBE on four categories spanning safety (harmful request refusal, over-refusal), capability (constrained instruction following), and truthfulness (multi-turn sycophancy resistance). The resulting evaluation covers a broader model spectrum without saturation while remaining compatible with existing datasets.

Haoxiang Wang, Da Yu, Huishuai Zhang

Peking University · Google Research · Wangxuan Institute of Computer Technology, Peking University State Key Laboratory of General Artificial Intelligence

LLM Evaluation Language Model Generation Evaluation Language Model Safety Evaluation

May 6, 2026cs.CL

Automatically Finding and Validating Unexpected Side-Effects of Interventions on Language Models

We present an automated, contrastive evaluation pipeline for auditing the behavioral impact of interventions on large language models. Given a base model

M_1

and an intervention model

M_2

, our method compares their free-form, multi-token generations across aligned prompt contexts and produces human-readable, statistically validated natural-language hypotheses describing how the models differ, along with recurring themes that summarize patterns across validated hypotheses. We evaluate the approach in synthetic setting by injecting known behavioral changes and showing that the pipeline reliably recovers them. We then apply it to three real-world interventions, reasoning distillation, knowledge editing and unlearning, demonstrating that the method surfaces both intended and unexpected behavioral shifts, distinguishes large from subtle interventions, and does not hallucinate differences when effects are absent or misaligned with the prompt bank. Overall, the pipeline provides a statistically grounded and interpretable tool for post-hoc auditing of intervention-induced changes in model behavior.

Quintin Pope, Ajay Hayagreeve Balaji, Jacques Thibodeau +1

Oregon State University · Independent Researcher

Language Model Generation Evaluation LLM Auditing Causal Interventions in Language Models

May 6, 2026cs.CL

StoryAlign: Evaluating and Training Reward Models for Story Generation

Story generation aims to automatically produce coherent, structured, and engaging narratives. Although large language models (LLMs) have significantly advanced text generation, stories generated by LLMs still diverge from human-authored works regarding complex narrative structure and human-aligned preferences. A key reason is the absence of effective modeling of human story preferences, which are inherently subjective and under-explored. In this work, we systematically evaluate the modeling of human story preferences and introduce StoryRMB, the first benchmark for assessing reward models on story preferences. StoryRMB contains

1,133

high-quality, human-verified instances, each consisting of a prompt, one chosen story, and three rejected stories. We find existing reward models struggle to select human-preferred stories, with the best model achieving only

66.3\%

accuracy. To address this limitation, we construct roughly

100,000

high-quality story preference pairs across diverse domains and develop StoryReward, an advanced reward model for story preference trained on this dataset. StoryReward achieves state-of-the-art (SoTA) performance on StoryRMB, outperforming much larger models. We also adopt StoryReward in downstream test-time scaling applications for best-of-n (BoN) story selection and find that it generally chooses stories better aligned with human preferences. We will release our dataset, model, and code to facilitate future research. Related code and data are available at https://github.com/THU-KEG/StoryReward.

Haotian Xia, Hao Peng, Yunjia Qi +4

Department of Computer Science and Technology, Tsinghua University

Reward Modeling Language Model Generation Evaluation Human Preference Modeling Text Generation Reward Model Evaluation

May 6, 2026cs.CL

RaguTeam at SemEval-2026 Task 8: Meno and Friends in a Judge-Orchestrated LLM Ensemble for Faithful Multi-Turn Response Generation

We present our winning system for Task~~B (generation with reference passages) in SemEval-2026 Task~~8: MTRAGEval. Our method is a heterogeneous ensemble of seven LLMs with two prompting variants, where a GPT-4o-mini judge selects the best candidate per instance. We ranked 1st out of 26 teams, achieving a conditioned harmonic mean of 0.7827 and outperforming the strongest baseline (gpt-oss-120b, 0.6390). Ablations show that diversity in model families, scales, and prompting strategies is essential, with the ensemble consistently beating any single model. We also introduce Meno-Lite-0.1, a 7B domain-adapted model with a strong cost--performance trade-off, and analyse MTRAGEval, highlighting annotation limitations and directions for improvement. Our code is publicly available: https://github.com/RaguTeam/ragu_mtrag_semeval

Ivan Bondarenko, Roman Derunets, Oleg Sedukhin +3

Novosibirsk State University

Language Model Generation Evaluation RAG Evaluation Language Model Ensembles

May 6, 2026cs.CL

DoGMaTiQ: Automated Generation of Question-and-Answer Nuggets for Report Evaluation

Evaluation of long-form, citation-backed reports has lately received significant attention due to the wide-scale adoption of retrieval-augmented generation (RAG) systems. Core to many evaluation frameworks is the use of atomic facts, or nuggets, to assess a report's coverage of query-relevant information attested in the underlying collection. While nuggets have traditionally been represented as short statements, recent work has used question-answer (QA) representations, enabling fine-grained evaluations that decouple the information need (i.e. the question) from the potentially diverse content that satisfies it (i.e. its answers). A persistent challenge for nugget-based evaluation is the need to manually curate sets of nuggets for each topic in a test collection -- a laborious process that scales poorly to novel information needs. This challenge is acute in cross-lingual settings, where information is found in multilingual source documents. Accordingly, we introduce DoGMaTiQ, a pipeline for generating high-quality QA-based nugget sets in three stages: (1) document-grounded nugget generation, (2) paraphrase clustering, and (3) nugget subselection based on principled quality criteria. We integrate DoGMaTiQ nuggets with AutoArgue -- a recent nugget-based evaluation framework -- to enable fully automatic evaluation of generated reports. We conduct extensive experiments on two cross-lingual TREC shared tasks, NeuCLIR and RAGTIME, showing strong rank correlations with both human-in-the-loop and fully manual judgments. Finally, detailed analysis of our pipeline reveals that a strong LLM nugget generator is key, and that the system rankings induced by DoGMaTiQ are robust to outlier systems. We facilitate future research in report evaluation by publicly releasing our code and artifacts at https://github.com/manestay/dogmatiq.

Bryan Li, William Walden, Yu Hou +6

Google Inc. · University of Pennsylvania · Johns Hopkins University +3

Language Model Generation Evaluation RAG Evaluation Automated Evaluation IR Evaluation

May 5, 2026cs.AI

EvoLM: Self-Evolving Language Models through Co-Evolved Discriminative Rubrics

Language models encode substantial evaluative knowledge from pretraining, yet current post-training methods rely on external supervision (human annotations, proprietary models, or scalar reward models) to produce reward signals. Each imposes a ceiling. Human judgment cannot supervise capabilities beyond its own, proprietary APIs create dependencies, and verifiable rewards cover only domains with ground-truth answers. Self-improvement from a model's own evaluative capacity is a reward source that scales with the model itself, yet remains largely untapped by current methods. We introduce EVOLM, a post-training method that structures this capacity into explicit discriminative rubrics and uses them as training signal. EVOLM trains two capabilities within a single language model in alternation: (1) a rubric generator producing instance-specific evaluation criteria optimized for discriminative utility, which maximizes a small frozen judge's ability to distinguish preferred from dispreferred responses; and (2) a policy trained using those rubric-conditioned scores as reward. All preference signals are constructed from the policy's own outputs via temporal contrast with earlier checkpoints, requiring no human annotation or external supervision. EVOLM trains a Qwen3-8B model to generate rubrics that outperform GPT-4.1 on RewardBench-2 by 25.7%. The co-trained policy achieves 69.3% average on the OLMo3-Adapt suite, outperforming policies trained with GPT-4.1 prompted rubrics by 3.9% and with the state-of-the-art 8B reward model SkyWork-RM by 16%. Overall, EVOLM demonstrates that structuring a model's evaluative capacity into co-evolving discriminative rubrics enables self-improvement without external supervision.

Shuyue Stella Li, Rui Xin, Teng Xiao +8

University of Washington · 2Allen Institute for AI · University of Pennsylvania

RL for Language Models LLM Evaluation Language Model Generation Evaluation Self-Training for Language Models Language Model Self-Assessment Language Model Post-Training+1

May 4, 2026cs.CL

Beating the Style Detector: Three Hours of Agentic Research on the AI-Text Arms Race

Reproducing an empirical NLP study used to take weeks. Given the released data and a modern agentic-research harness, we redo every experiment of a recent ACL,2026 study on personal-style post-editing of LLM drafts -- and add three new ones -- with the human investigator acting only as a reviewer-in-the-loop. We reproduce all seven preregistered hypotheses and recover the paper's headline correlation between perceived self-similarity and embedding-measured self-similarity to three decimal places (

r{=}{+}0.244

,

p{<}10^{-8}

,

n{=}648

). Under a leakage-free held-out protocol, GPT-5.5 and Claude,Opus,4.7 close

71

--

75\,\%

of the style gap to the same-author ceiling on

324

paired tasks, against

24\,\%

for the human post-edit, and beat the human post-edit on

\sim$$80\,\%

of tasks. We then frame the same data as an AI-text detection arms race. A leave-authors-out linear SVM on LUAR-MUD embeddings reaches AUC

0.93

--

1.00

across approaches; six diagnostics show that GPT-5.5 detection is mostly a length confound while Opus detection is a genuine stylistic signature. Given

T{=}20

feedback iterations against the frozen detector, an Opus agent flips two of five held-out test mimics to the human half-space and shrinks every margin by an order of magnitude. With moderate effort against a known detector, a frontier LLM can already efficiently lower its own AI-detection probability. All code,

648

mimic drafts, trained detectors, diagnostics, and adversarial trajectories are released.

Andreas Maier, Moritz Zaiss, Siming Bayer

Pattern Recognition Lab, Friedrich-Alexander-Universität Erlangen-Nürnberg, Erlangen, Germany · Institute of Neuroradiology, University Hospital Erlangen, Friedrich-Alexander-Universität Erlangen-Nürnberg, Erlangen, Germany

AI-Generated Text Detection Language Model Generation Evaluation Robustness of Machine-Generated Text Detection Text Generation Adversarial Text Attacks

May 4, 2026cs.CL

A multilingual hallucination benchmark: MultiWikiQHalluA

Most hallucination evaluations focus on English, leaving it unclear whether findings transfer to lower-resource languages. We investigate faithfulness hallucinations, defined as model-generated content that is fluent and plausible but diverges from the provided input or is internally inconsistent. Leveraging the multilingual MultiWikiQA dataset, we utilize the LettuceDetect framework to create synthetic hallucination datasets for 306 languages, from which we train token-level hallucination classifiers for 30 European languages. In this work, we present evaluations of model hallucinations on a selection of languages: English, Danish, German, and Icelandic. Using these classifiers, we evaluate the hallucination rates for Qwen3-0.6B, Qwen3-14B, Gemma-3-12B-IT, cogito-v1-preview-qwen-32B, and cogito-v1-preview-llama-70B. Our classifiers reveal notably higher hallucination rates for Qwen3-0.6B (up to 60% of answers containing at least one hallucination, peaking in Icelandic) and generally lower rates for larger models, with cogito-v1-preview-qwen-32B and cogito-v1-preview-llama-70B performing best on most languages. Hallucination rates are consistently higher for lower-resource languages, particularly Icelandic.

Freja Thoresen, Dan Saattrup Smart

Alexandra Institute · Rued Langgaards Vej 7, 2300 København

Multilingual Language Models Multilingual Language Model Evaluation Language Model Error Detection Language Model Generation Evaluation Hallucination in Language Models LLM Reliability+1

May 2, 2026cs.CL

Prosa: Rubric-Based Evaluation of LLMs on Real User Chats in Brazilian Portuguese

Rankings produced by holistic LLM-as-a-judge scoring are sensitive to the bias of the chosen judge model. We show that switching to binary rubric scoring with multi-judge filtering removes this sensitivity: decomposing the judgement matters more than the judge model itself. To support this claim, we introduce Prosa, the first real user multi-turn Brazilian Portuguese chat benchmark: 1,000 WildChat conversations scored by three judges from three model families on 16 models. Under filtered rubric scoring the three judges agree on every one of the 16 ranks, whereas under holistic scoring they agree on only 7 of 16. Additionally, the rubric filtering pipeline increases the average score gap between neighbouring models by 47%, thereby improving Prosa's discriminative power. Evaluating a new model on Prosa costs approximately $2.1 when using Gemini 3 Flash as the judge. We release the benchmark and the filtering code to ensure that future models can be assessed under identical conditions. These artifacts also make our rubric-based scoring method reusable beyond Prosa, supporting other open-ended evaluation settings.

Roseval Malaquias Junior, Giovana Kerche Bonás, Thales Sales Almeida +6

Maritaca AI · Jusbrasil

LLM Evaluation LLM-as-a-Judge Language Model Generation Evaluation Multi-Turn Dialogue Evaluation Rubric-Based Evaluation Bias in LLM-as-a-Judge

May 2, 2026cs.LG

The Partial Testimony of Logs: Evaluation of Language Model Generation under Confounded Model Choice

Offline evaluation of language models from usage logs is biased when model choice is confounded: the same user-side factors that influence which model is used can also influence how its output is judged, so raw comparisons of logged scores mix self-selected populations rather than estimating a common quantity of interest. A small randomized experiment can break this bias by overriding model choice, but in practice such experiments are scarce and costly. We study a three-source design that combines a large confounded observational log (OBS) for scale, a small randomized experiment (EXP) for unconfounded scoring, and an offline simulator (SIM) that replays candidate models on cached contexts. Our main result is an identification theorem showing that the randomized experiment and the simulator are together enough to recover causal model values; the observational log enters only afterward, to reduce estimation error rather than to make the causal comparison valid. Six estimator families are evaluated in a controlled semi-synthetic validation and in two real-task cached benchmarks for summarization and coding. No family dominates every regime; relative performance depends on the amount of unbiased EXP supervision and on how closely the target reward aligns with OBS-derived structure.

Jikai Jin, Vasilis Syrgkanis

LLM Evaluation Off-Policy Evaluation Language Model Generation Evaluation

Apr 30, 2026cs.AI

Optimization before Evaluation: Evaluation with Unoptimised Prompts Can be Misleading

Current Large Language Model (LLM) evaluation frameworks utilize the same static prompt template across all models under evaluation. This differs from the common industry practice of using prompt optimization (PO) techniques to optimize the prompt for each model to maximize application performance. In this paper, we investigate the effect of PO towards LLM evaluations. Our results on public academic and internal industry benchmarks show that PO greatly affects the final ranking of models. This highlights the importance of practitioners performing PO per model when conducting evaluations to choose the best model for a given task.

Nicholas Sadjoli, Tim Siefken, Atin Ghosh +2

SAP · Stanford University

LLM Evaluation Language Model Generation Evaluation Prompt Optimization

Apr 30, 2026cs.CL

From Coarse to Fine: Benchmarking and Reward Modeling for Writing-Centric Generation Tasks

Large language models have achieved remarkable progress in text generation but still struggle with generative writing tasks. In terms of evaluation, existing benchmarks evaluate writing reward models coarsely and fail to measure performance from the perspective of specific requirements. In terms of training, existing training methods either use LLM-as-a-judge approaches or train coarse-grained reward models, lacking fine-grained requirement-adherence reward modeling. To address these issues, we propose a fine-grained evaluation pipeline WEval for writing reward models and a fine-grained reinforcement learning training framework WRL. The evaluation data of WEval covers multiple task categories and requirement types, enabling systematic evaluation of writing reward models by measuring the correlation between the rankings of the reward model and gold rankings. WRL constructs positive and negative samples by selectively dropping instruction requirements, allowing for more precise reward model training. Experiments show that our models achieve substantial improvements across various writing benchmarks and exhibit strong generalization. The code and data are publicly available at https://github.com/Rainier-rq1/From_Coarse_to_Fine.

Qingyu Ren, Tianjun Pan, Xingzhou Chen +1

Shanghai Key Laboratory of Data Science, College of Computer Science and Artificial Intelligence, Fudan University · 2Shanghai Artificial Intelligence Laboratory

Reward Modeling RL for Language Models LLM Evaluation Language Model Generation Evaluation Instruction Following Reward Model Evaluation

Apr 29, 2026cs.CL

Instruction Complexity Induces Positional Collapse in Adversarial LLM Evaluation

When instructed to underperform on multiple-choice evaluations, do language models engage with question content or fall back on positional shortcuts? We map the boundary between these regimes using a six-condition adversarial instruction-specificity gradient administered to two instruction-tuned LLMs (Llama-3-8B and Llama-3.1-8B) on 2,000 MMLU-Pro items. Distributional screening (response-position entropy) and an independent content-engagement criterion (difficulty-accuracy correlation) jointly characterise each condition. The gradient reveals three regimes rather than a monotonic transition. Vague adversarial instructions produce moderate accuracy reduction with preserved content engagement. Standard sandbagging and capability-imitation instructions produce positional entropy collapse with partial content engagement. A two-step answer-aware avoidance instruction produces extreme positional collapse, with near-total concentration on a single response position (99.9% and 87.4%) and no measurable content sensitivity. This was the only multi-step instruction tested, and it produced the most extreme shortcut. The attractor position matches each model's content-absent null-prompt default. The effect replicates across both models and four academic domains. Distributional collapse and content engagement can co-occur (50% concordance between screening criteria), indicating that entropy-based screening and difficulty-based content assessment capture partially independent dimensions of response validity. Results suggest that instruction complexity can determine whether adversarial compliance uses content-aware or content-blind mechanisms in small instruction-tuned LLMs under greedy decoding.

Jon-Paul Cacioli

Independent Researcher, Melbourne, Australia

LLM Evaluation Language Model Generation Evaluation Adversarial Attacks on LLMs

Apr 29, 2026cs.CL

Theory-Grounded Evaluation Exposes the Authorship Gap in LLM Personalization

Stylistic personalization - making LLMs write in a specific individual's style, rather than merely adapting to task preferences - lacks evaluation grounded in authorship science. We show that grounding evaluation in authorship verification theory transforms what benchmarks can measure. Drawing on three measurement traditions - LUAR, a trained authorship verification model; an LLM-as-judge with decoupled trait matching; and classical function-word stylometrics - we evaluate four inference-time personalization methods across 50 authors and 1,000 generations. The theory-grounded metric, LUAR, provides what ad hoc alternatives cannot: calibrated baselines, with a human ceiling of 0.756 and a cross-author floor of 0.626, that give scores absolute meaning. All methods score below this floor, from 0.484 to 0.508, exposing an authorship gap invisible to uncalibrated metrics. The three metrics produce near-zero pairwise correlations, with absolute r less than 0.07, confirming that without theoretical grounding, metric choice determines conclusions: an LLM judge declares a clear winner while LUAR finds no meaningful differentiation. These findings demonstrate the theory-benchmark cycle in action: authorship theory exposes evaluation failures that ad hoc benchmarks miss.

Yash Ganpat Sawant

Independent AI Researcher

LLM Evaluation LLM Personalization Language Model Generation Evaluation Authorship Attribution

Apr 28, 2026cs.CR

The Surprising Universality of LLM Outputs: A Real-Time Verification Primitive

We report a striking statistical regularity in frontier LLM outputs that enables a CPU-only scoring primitive running at 2.6 microseconds per token, with estimated latency up to 100,000

\times

(five orders of magnitude) below existing sampling-based detectors. Across six contemporary models from five independent vendors, two generation sizes, and five held-out domains, token rank-frequency distributions converge to the same two-parameter Mandelbrot ranking distribution, with 34 of 36 model-by-domain fits exceeding

R^{2} = 0.94

and 35 of 36 favoring Mandelbrot over Zipf by AIC. The shared family does not collapse the models into statistical duplicates. Fitted Mandelbrot parameters remain cleanly separable between models: the cross-model spread in

q

(1.63 to 3.69) exceeds its per-model bootstrap standard deviation (0.03 to 0.10) by more than an order of magnitude, yielding tens of standard deviations of separation per few thousand output tokens. Two capabilities follow. First, statistical model fingerprinting: text from a vendor-delivered LLM can be tested against its claimed model family without cryptographic watermarks or access to model internals, supporting provenance verification and silent-substitution audits. Second, a model-agnostic reference distribution for black-box output assessment, from which we derive a single-pass scoring primitive that composes with model log probabilities when available and degrades to a rank-only mode usable on closed APIs. Pilot results on FRANK, TruthfulQA, and HaluEval map where the primitive helps (lexical anomalies, unsupported entities) and where it structurally cannot (reasoning errors in domain-appropriate vocabulary). We position the primitive as a first-pass triage layer in compound evaluation stacks, not as a replacement for sampling-based or source-conditioned verifiers.

Alex Bogdan, Adrian de Valois-Franklin

Evolutionairy AI Toronto, Canada

LLM Evaluation Language Model Generation Evaluation LLM Auditing Language Model Fingerprinting Anomaly Detection

Apr 28, 2026cs.CR

MGTEVAL: An Interactive Platform for Systemtic Evaluation of Machine-Generated Text Detectors

We present MGTEVAL, an extensible platform for systematic evaluation of Machine-Generated Text (MGT) detectors. Despite rapid progress in MGT detection, existing evaluations are often fragmented across datasets, preprocessing, attacks, and metrics, making results hard to compare and reproduce. MGTEVAL organizes the workflow into four components: Dataset Building, Dataset Attack, Detector Training, and Performance Evaluation. It supports constructing custom benchmarks by generating MGT with configurable LLMs, applying 12 text attacks to test sets, training detectors via a unified interface, and reporting effectiveness, robustness, and efficiency. The platform provides both command-line and Web-based interfaces for user-friendly experimentation without code rewriting.

Yuanfan Li, Qi Zhou, Chengzhengxu Li +5

Faculty of Electronic and Information Engineering, Xi’an Jiaotong University · Queen Mary University of London

AI-Generated Text Detection Language Model Generation Evaluation Language Model Robustness Robustness of Machine-Generated Text Detection

Apr 28, 2026cs.CL

LongSumEval: Question-Answering Based Evaluation and Feedback-Driven Refinement for Long Document Summarization

Evaluating long document summaries remains the primary bottleneck in summarization research. Existing metrics correlate weakly with human judgments and produce aggregate scores without explaining deficiencies or guiding improvement, preventing effective refinement in applications requiring verifiable accuracy. We introduce LongSumEval, a unified framework bridging evaluation and generation through structured question-answering feedback. The framework operationalizes summary quality as answerability and factual alignment of question-answer pairs, generating interpretable scores and actionable feedback that identifies coverage gaps and factual inconsistencies. This resolves the misalignment where evaluation operates independently of generation objectives. Meta-evaluation of our QA-based evaluation module across seven benchmarks demonstrates substantially stronger agreement with human judgments compared to established metrics. Structured feedback enables significant quality improvements through self-refinement without retraining. By demonstrating that evaluation feedback can serve as executable instructions for generation, this work establishes a generalizable paradigm for aligning assessment with improvement, with direct implications for controllable text generation requiring verifiable accuracy and transparent quality control. All code and datasets will be released in GitHub for reproducibility.

Huyen Nguyen, Haoxuan Zhang, Yang Zhang +2

Cigna Group Evernorth Health Services Austin, Texas, USA · dept. of Information Science University of North Texas Denton, Texas, USA · dept. of Data Science University of North Texas Denton, Texas, USA

Language Model Generation Evaluation Factual Consistency Evaluation Text Summarization Summarization Evaluation Automated Evaluation

Apr 26, 2026cs.CL

JudgeSense: A Benchmark for Prompt Sensitivity in LLM-as-a-Judge Systems

Large language models are widely adopted as automated evaluation judges, yet the stability of their verdicts under semantically equivalent prompt rephrasings remains largely unexamined. We conduct a systematic empirical study of prompt-induced decision instability across multiple evaluation tasks and judge architectures. To facilitate this analysis, we release JudgeSense, a benchmark comprising hand-validated prompt-paraphrase pairs spanning factuality, coherence, relevance, and preference, drawn from established NLP benchmarks and accompanied by comprehensive decision logs. The benchmark enables the measurement of judge stability across equivalent prompts, allowing researchers to assess whether stability correlates with model scale or instruction-tuning, and to identify which tasks are most sensitive to prompt wording. Our evaluation reveals that coherence remains the primary task for distinguishing judge behavior, while factuality judgments demonstrate high stability under standard conditions. Pairwise evaluation tasks consistently exhibit position bias. Crucially, we find that model scale is not a reliable proxy for consistency; notably, as an interesting result in our analysis, the largest and newest models are not the most consistent.

Rohith Reddy Bellibatlu, Edward Raff, Wenbin Zhang

Florida International University, Miami, FL · CrowdStrike, New York, USA · University of Maryland Baltimore County, Baltimore, USA

LLM Evaluation LLM-as-a-Judge Language Model Generation Evaluation