Generative AI Evaluation

Follow this topic Ask about this topic

The field asks whether generated text, images, audio, video, and other artifacts meet users’ goals—not just whether they resemble a reference. Common approaches include reference and embedding metrics, distribution distances such as FID and MMD, rubric-based LLM judges, pairwise preference comparisons, and expert or user studies. Task-specific benchmarks increasingly test dimensions such as faithfulness, constraint following, multimodal synchronization, robustness, and performance in complete workflows. A central challenge is that automated scores can disagree with people or miss important failures, so evaluations need clearer constructs, uncertainty estimates, human validation, and evidence of real-world utility.

Latest papers

314 papers38 papers in the last four weeks, up 3% on the four weeks before. 0.2% of all new papers.

Mar 17, 2026

Answer Bubbles: Information Exposure in AI-Mediated Search

Generative search systems are increasingly replacing link-based retrieval with AI-generated summaries, yet little is known about how these systems differ in sources, language, and fidelity to cited material. We examine responses to 11,000 real search queries across five systems---vanilla GPT, Search GPT, Perplexity Search with Grok, Google AI Overviews, and traditional Google Search---at three levels: source diversity, linguistic characterization of the generated summary, and source-summary fidelity. We find that generative search systems exhibit significant source-selection biases in their citations, favoring certain sources over others. Incorporating search also selectively attenuates epistemic markers, reducing hedging by up to 60% while preserving confidence language in the AI-generated summaries. At the same time, AI summaries further compound the citation biases: Wikipedia and longer sources are disproportionately overrepresented, whereas cited social media content and negatively framed sources are substantially underrepresented. Our findings highlight the potential for answer bubbles, in which identical queries yield structurally different information realities across systems, with implications for user trust, source visibility, and the transparency of AI-mediated information access.

Michelle Huang, Agam Goyal, Koustuv Saha +1

Siebel School of Computing and Data Science University of Illinois Urbana-Champaign, Urbana, IL, USA

Generative Retrieval Information Retrieval Algorithmic Bias Generative AI Evaluation

Mar 16, 2026

AdaQE-CG: Adaptive Query Expansion for Web-Scale Generative AI Model and Data Card Generation

Transparent and standardized documentation is essential for building trustworthy generative AI (GAI) systems. However, existing automated methods for generating model and data cards still face three major challenges: (i) static templates, as most systems rely on fixed query templates that cannot adapt to diverse paper structures or evolving documentation requirements; (ii) information scarcity, since web-scale repositories such as Hugging Face often contain incomplete or inconsistent metadata, leading to missing or noisy information; and (iii) lack of benchmarks, as the absence of standardized datasets and evaluation protocols hinders fair and reproducible assessment of documentation quality. To address these limitations, we propose AdaQE-CG, an Adaptive Query Expansion for Card Generation framework that combines dynamic information extraction with cross-card knowledge transfer. Its Intra-Paper Extraction via Context-Aware Query Expansion (IPE-QE) module iteratively refines extraction queries to recover richer and more complete information from scientific papers and repositories, while its Inter-Card Completion using the MetaGAI Pool (ICC-MP) module fills missing fields by transferring semantically relevant content from similar cards in a curated dataset. In addition, we introduce MetaGAI-Bench, the first large-scale, expert-annotated benchmark for evaluating GAI documentation. Comprehensive experiments across five quality dimensions show that AdaQE-CG substantially outperforms existing approaches, exceeds human-authored data cards, and approaches human-level quality for model cards. Code, prompts, and data are publicly available at: https://github.com/haoxuan-unt2024/AdaQE-CG.

Haoxuan Zhang, Ruochi Li, Zhenni Liang +7

Document Information Extraction Generative AI Evaluation Query Expansion

Mar 14, 2026

Evaluating Prompt Robustness in Text-to-Audio Systems for Adaptive Virtual Agents and Game Soundtracks

Recent text-to-audio models enable adaptive game soundtracks, but small prompt changes can cause abrupt musical shifts. We evaluate MusicGen-small, MusicGen-large, and Stable Audio 2.5 under Minimal Lexical Substitution, Intensity Shifts, and Structural Rephrasing using log-Mel distance, MFCC/chroma-DTW, and CLAP similarity. Stable Audio 2.5 achieves the lowest pooled acoustic distances and the highest audio-audio CLAP similarity under structural rephrasing, while MusicGen-large has the highest audio-audio CLAP similarity under lexical substitutions and intensity shifts. Stable Audio 2.5 also shows the greatest between-seed variation in prompt-to-audio alignment, demonstrating the importance of multi-seed robustness evaluation for adaptive game audio.

Jiahui Wu, Mei Si

Rensselaer Polytechnic Institute Troy, New York, USA

Music Generation Evaluation Text-to-Audio Generation Music Generation Generative AI Evaluation

Mar 12, 2026

The Perfection Paradox: From Architect to Curator in AI-Assisted API Design

Enterprise API design is often bottlenecked by the tension between rapid feature delivery and the rigorous maintenance of usability standards. We present an industrial case study evaluating an AI-assisted design workflow trained on API Improvement Proposals (AIPs). Through a controlled study with 16 industry experts, we compared AI-generated API specifications against human-authored ones. While quantitative results indicated AI superiority in 10 of 11 usability dimensions and an 87% reduction in authoring time, qualitative analysis revealed a paradox: experts frequently misidentified AI work as human (19% accuracy) yet described the designs as unsettlingly "perfect." We characterize this as a "Perfection Paradox" -- where hyper-consistency signals a lack of pragmatic human judgment. We discuss the implications of this perfection paradox, proposing a shift in the human designer's role from the "drafter" of specifications to the "curator" of AI-generated patterns.

Mak Ahmad, Andrew Macvean, JJ Geewax +1

Google, San Francisco, CA, USA · Google, Seattle, USA · Google, Singapore +1

Software Engineering Human-AI Co-Creation Generative AI Evaluation

Mar 12, 2026

GPT4o-Receipt: A Dataset and Human Study for AI-Generated Document Forensics

Can humans detect AI-generated financial documents better than machines? We present GPT4o-Receipt, a benchmark of 1,235 receipt images pairing GPT-4o-generated receipts with authentic ones from established datasets, evaluated by five state-of-the-art multimodal LLMs and a 30-annotator crowdsourced perceptual study. Our findings reveal a striking paradox: humans are better at seeing AI artifacts, yet worse at detecting AI documents. Human annotators exhibit the largest visual discrimination gap of any evaluator, yet their binary detection F1 falls well below Claude Sonnet 4 and below Gemini 2.5 Flash. This paradox resolves once the mechanism is understood: the dominant forensic signals in AI-generated receipts are arithmetic errors -- invisible to visual inspection but systematically verifiable by LLMs. Humans cannot perceive that a subtotal is incorrect; LLMs verify it in milliseconds. Beyond the human--LLM comparison, our five-model evaluation reveals dramatic performance disparities and calibration differences that render simple accuracy metrics insufficient for detector selection. GPT4o-Receipt, the evaluation framework, and all results are released publicly to support future research in AI document forensics.

Yan Zhang, Simiao Ren, Ankit Raj +6

Digital Image Forensics AI-Generated Text Detection Generative AI Evaluation Document Image Understanding

Mar 11, 2026

The Orthogonal Vulnerabilities of Generative AI Watermarks: A Comparative Empirical Benchmark of Spatial and Latent Provenance

As open-weights generative AI rapidly proliferates, the ability to synthesize hyper-realistic media has introduced profound challenges to digital trust. Automated disinformation and AI-generated imagery have made robust digital provenance a critical cybersecurity imperative. Currently, state-of-the-art invisible watermarks operate within one of two primary mathematical manifolds: the spatial domain (post-generation pixel embedding) or the latent domain (pre-generation frequency embedding). While existing literature frequently evaluates these models against isolated, classical distortions, there is a critical lack of rigorous, comparative benchmarking against modern generative AI editing tools. In this study, we empirically evaluate two leading representative paradigms, RivaGAN (Spatial) and Tree-Ring (Latent), utilizing an automated Attack Simulation Engine across 30 intensity intervals of geometric and generative perturbations. We formalize an "Adversarial Evasion Region" (AER) framework to measure cryptographic degradation against semantic visual retention (OpenCLIP > 75.0). Our statistical analysis (

n=100

per interval,

MOE = \pm 3.92\%

) reveals that these domains possess mutually exclusive, mathematically orthogonal vulnerabilities. Spatial watermarks experience severe cryptographic degradation under algorithmic pixel-rewriting (exhibiting a 67.47% AER evasion rate under Img2Img translation), whereas latent watermarks exhibit profound fragility against geometric misalignment (yielding a 43.20% AER evasion rate under static cropping). By proving that single-domain watermarking is fundamentally insufficient against modern adversarial toolsets, this research exposes a systemic vulnerability in current digital provenance standards and establishes the foundational exigence for future multi-domain cryptographic architectures.

Jesse Yu, Nicholas Wei

Millburn High School · Williamsville East High School

AI-Generated Image Watermarking Robust Watermarking Watermark Removal Generative AI Evaluation Diffusion Model Watermarking

Mar 10, 2026

Diagnosing and Repairing Citation Failures in Generative Engine Optimization

Generative Engine Optimization (GEO) aims to improve content visibility in AI-generated responses. However, existing methods measure contribution-how much a document influences a response-rather than citation, the mechanism that actually drives traffic back to creators. Also, these methods apply generic rewriting rules uniformly, failing to diagnose why individual document are not cited. This paper introduces a diagnostic approach to GEO that asks why a document fails to be cited and intervenes accordingly. We develop a unified framework comprising: (1) the first taxonomy of citation failure modes spanning different stages of a citation pipeline; (2) AgentGEO, an agentic system that diagnoses failures using this taxonomy, selects targeted repairs from a corresponding tool library, and iterates until citation is achieved; and (3) a document-centric benchmark evaluating whether optimizations generalize across held-out queries. AgentGEO achieves over 40% relative improvement in citation rates while modifying only 5% of content, compared to 25% for baselines. Our analysis reveals that generic optimization can harm long-tail content and some documents face challenges that optimization alone cannot fully address-findings with implications for equitable visibility in AI-mediated information access.

Zhihua Tian, Yuhan Chen, Yao Tang +2

Virginia Tech · Zhejiang University

Generative Engine Optimization Generative AI Evaluation

Mar 10, 2026

Evaluate-as-Action: Self-Evaluated Process Rewards for Retrieval-Augmented Agents

Retrieval-augmented agents can query external evidence, yet their reliability in multi-step reasoning remains limited: noisy retrieval may derail multi-hop question answering, while outcome-only reinforcement learning provides credit signals that are too coarse to optimize intermediate steps. We propose EvalAct (Evaluate-as-Action), which converts implicit retrieval quality assessment into an explicit action and enforces a coupled Search-to-Evaluate protocol so that each retrieval is immediately followed by a structured evaluation score, yielding process signals aligned with the interaction trajectory. To leverage these signals, we introduce Process-Calibrated Advantage Rescaling (PCAR), a GRPO-based optimization method that rescales advantages at the segment level according to evaluation scores, emphasizing reliable segments while updating uncertain ones conservatively. Experiments on seven open-domain QA benchmarks show that EvalAct achieves the best average accuracy, with the largest gains on multi-hop tasks, and ablations verify that the explicit evaluation loop drives the primary improvements while PCAR provides consistent additional benefits.

Jiangming Shu, Yuxiang Zhang, Ye Ma +2

School of Computer Science and Technology, Beijing Jiaotong University · Hithink Research

Retrieval-Augmented Reasoning Agentic RAG RL for LLM Agents Process Reward Models Multi-Hop QA Generative AI Evaluation

Mar 9, 2026

Quantifying Uncertainty in AI Visibility: A Statistical Framework for Generative Search Measurement

AI-powered answer engines are inherently non-deterministic: identical queries submitted at different times can produce different responses and cite different sources. Despite this stochastic behavior, current approaches to measuring domain visibility in generative search typically rely on single-run point estimates of citation share and prevalence, implicitly treating them as fixed values. This paper argues that citation visibility metrics should be treated as sample estimators of an underlying response distribution rather than fixed values. We conduct an empirical study of citation variability across three generative search platforms--Perplexity Search, OpenAI SearchGPT, and Google Gemini--using repeated sampling across three consumer product topics. Two sampling regimes are employed: daily collections over nine days and high-frequency sampling at ten-minute intervals. We show that citation distributions follow a power-law form and exhibit substantial variability across repeated samples. Bootstrap confidence intervals reveal that many apparent differences between domains fall within the noise floor of the measurement process. Distribution-wide rank stability analysis further demonstrates that citation rankings are unstable across samples, not only among top-ranked domains but throughout the frequently cited domain set. These findings demonstrate that single-run visibility metrics provide a misleadingly precise picture of domain performance in generative search. We argue that citation visibility must be reported with uncertainty estimates and provide practical guidance for sample sizes required to achieve interpretable confidence intervals.

Ronald Sielinski

IQRush

Generative AI Evaluation Uncertainty Quantification

Mar 7, 2026

PresentBench: A Fine-Grained Rubric-Based Benchmark for Slide Generation

Slides serve as a critical medium for conveying information in presentation-oriented scenarios such as academia, education, and business. Despite their importance, creating high-quality slide decks remains time-consuming and cognitively demanding. Recent advances in generative models, such as Nano Banana Pro, have made automated slide generation increasingly feasible. However, existing evaluations of slide generation are often coarse-grained and rely on holistic judgments, making it difficult to accurately assess model capabilities or track meaningful advances in the field. In practice, the lack of fine-grained, verifiable evaluation criteria poses a critical bottleneck for both research and real-world deployment. In this paper, we propose PresentBench, a fine-grained, rubric-based benchmark for evaluating automated real-world slide generation. It contains 238 evaluation instances, each supplemented with background materials required for slide creation. Moreover, we manually design an average of 54.1 checklist items per instance, each formulated as a binary question, to enable fine-grained, instance-specific evaluation of the generated slide decks. Extensive experiments show that PresentBench provides more reliable evaluation results than existing methods, and exhibits significantly stronger alignment with human preferences. Furthermore, our benchmark reveals that NotebookLM significantly outperforms other slide generation methods, highlighting substantial recent progress in this domain.

Xin-Sheng Chen, Jiayu Zhu, Pei-lin Li +3

Tsinghua University

Benchmark Design Generative AI Evaluation Automated Evaluation Rubric-Based Evaluation

Mar 6, 2026

Evaluating Human-AI Safety: A Framework for Measuring Harmful Capability Uplift

Current frontier AI safety evaluations emphasize static benchmarks, third-party annotations, and red-teaming. In this position paper, we argue that AI safety research should focus on human-centered evaluations that measure harmful capability uplift: the marginal increase in a user's ability to cause harm with a frontier model beyond what conventional tools already enable. We frame harmful capability uplift as a core AI safety metric, ground it in prior social science research, and provide concrete methodological guidance for systematic measurement. We conclude with actionable steps for developers, researchers, funders, and regulators to make harmful capability uplift evaluation a standard practice.

Michelle Vaccaro, Jaeyoon Song, Abdullah Almaatouq +1

Massachusetts Institute of Technology, Cambridge, USA

AI Safety Evaluation LLM Safety Evaluation Generative AI Evaluation Human-Centered AI

Mar 5, 2026

FireBench: Evaluating Instruction Following in Enterprise and API-Driven LLM Applications

Instruction following is critical for LLMs deployed in enterprise and API-driven settings, where strict adherence to output formats, content constraints, and procedural requirements is essential for enabling reliable LLM-assisted workflows. However, existing instruction following benchmarks predominantly evaluate natural language generation constraints that reflect the needs of chat assistants rather than enterprise users. To bridge this gap, we introduce FireBench, an LLM instruction following benchmark grounded in real-world enterprise and API usage patterns. FireBench evaluates six core capability dimensions across diverse applications including information extraction, customer support, and coding agents, comprising over 2,400 samples. We evaluate 11 LLMs and present key findings on their instruction following behavior in enterprise scenarios. We open-source FireBench at fire-bench.com to help users assess model suitability, support model developers in diagnosing performance, and invite community contributions.

Yunfan Zhang, Yijie Bei, Jetashree Ravi +1

Columbia University · Fireworks AI

LLM Evaluation Instruction Following Generative AI Evaluation Language Modeling

Mar 4, 2026

Struggle Premium : How Human Effort and Imperfection Drive Perceived Value in the Age of AI

As AI enters creative practice, audiences face growing uncertainty in judging authenticity and value. This study examines the Struggle Premium, the added value attributed to perceived human effort, by analyzing how visible effort cues influence evaluations of human- and AI-generated creative works. We surveyed 70 university students, focusing on process videos, time documentation, written explanations, and imperfections. Process-oriented cues, especially videos and time spent, most strongly shaped authenticity and value judgments, while imperfections had limited impact. Participants showed a clear preference for human-made works, with 72.9% willing to pay more. Notably, effort cues also improved perceptions of AI-generated content, suggesting that process transparency can partially bridge authenticity gaps. These findings extend the effort heuristic to algorithmic creativity and inform the design of transparent human-AI creative systems.

Nazneen Sultana, Mst Rafia Islam, Md. Tanvir Hossain +1

Shahjalal University of Science and Technology, Sociology, Sylhet, Bangladesh · Independent University, Bangladesh, Department of Law, Dhaka, Bangladesh · Shahjalal University of Science and Technology, Industrial Engineering, Sylhet, Bangladesh

Generative AI Evaluation Human Preference Evaluation Computational Creativity

Mar 4, 2026

Generative AI for Managerial Decision-Making under Ambiguity and Sycophancy

Generative artificial intelligence (GenAI) is increasingly being integrated into complex business workflows, fundamentally shifting the boundaries of managerial decision-making. However, the reliability of its strategic advice in ambiguous business contexts remains a critical knowledge gap. To address this gap, this study compares multiple GenAI models in their ability to detect ambiguity, examines whether a systematic ambiguity-resolution process improves response quality, and investigates their susceptibility to sycophantic behavior when confronted with flawed managerial directives. Using a novel four-dimensional business ambiguity taxonomy, we conducted a human-in-the-loop experiment across strategic, tactical, and operational scenarios. The resulting decisions were assessed through a human-validated automated evaluation framework based on agreement, actionability, justification quality, and constraint adherence. The results show that our approach not only distinguishes different types of ambiguity, but also reveals how ambiguity resolution systematically changes model behavior. In particular, resolving ambiguities improved decision quality across all managerial levels, with the strongest gains observed in constraint adherence. The analysis further showed that sycophantic behavior is not uniform across models: some models challenged flawed assumptions, whereas others tended to comply with them. This study contributes to the bounded rationality literature by positioning GenAI as a cognitive scaffold that can detect and resolve ambiguities managers might overlook, while demonstrating that its artificial limitations require human oversight to ensure its reliability as a strategic partner.

Sule Ozturk Birim, Fabrizio Marozzo, Yigit Kazancoglu

Manisa Celal Bayar University, Salihli, Manisa, Turkey · University of Calabria, Rende, Italy · Yasar University, Izmir, Turkey

AI-Assisted Decision Making Generative AI Evaluation LLM Sycophancy Human Oversight of AI

Mar 3, 2026

Baseline Performance of AI Tools in Classifying Cognitive Demand of Mathematical Tasks

Teachers face increasing demands on their time, particularly in adapting mathematics curricula to meet individual student needs while maintaining cognitive rigor. This study evaluates whether AI tools can accurately classify the cognitive demand of mathematical tasks, which is important for creating or adapting tasks that support student learning. We tested eleven AI tools: six general-purpose (ChatGPT, Claude, DeepSeek, Gemini, Grok, Perplexity) and five education-specific (Brisk, Coteach AI, Khanmigo, Magic School, School

.

AI), on their ability to categorize mathematics tasks across four levels of cognitive demand using a research-based framework. The goal was to approximate the performance teachers will achieve with straightforward prompts. On average, AI tools accurately classified cognitive demand in only 63% of cases. Education-specific tools were not more accurate than general-purpose tools, and no tool exceeded 83% accuracy. All tools struggled with tasks at the extremes of cognitive demand (Memorization and Doing Mathematics), exhibiting a systematic bias toward middle-category levels (Procedures with/without Connections). The tools often gave plausible-sounding explanations likely to be persuasive to novice teachers. Error analysis of AI tools' misclassification of the broad level of cognitive demand (high vs. low) revealed that tools consistently overweighted surface textual features over underlying cognitive processes. Further, AI tools showed weaknesses in reasoning about factors that make tasks higher vs. lower cognitive demand. Errors stemmed not from ignoring relevant dimensions, but from incorrectly reasoning about multiple task aspects. These findings carry implications for AI integration into teacher planning workflows and highlight the need for improved prompt engineering and tool development for educational applications.

Danielle S. Fox, Brenda L. Robles, Elizabeth DiPietro Brovey +1

Learning Research and Development Center, University of Pittsburgh · Institute for Learning, University of Pittsburgh

AI in Education Generative AI Evaluation Educational Assessment

Mar 3, 2026

Using Learning Progressions to Guide AI Feedback for Science Learning

Generative artificial intelligence (AI) offers scalable support for formative feedback, yet most AI-generated feedback relies on task-specific rubrics authored by domain experts. While effective, rubric authoring is time-consuming and limits scalability across instructional contexts. Learning progressions (LP) provide a theoretically grounded representation of students' developing understanding and may offer an alternative solution. This study examines whether an LP-driven rubric generation pipeline can produce AI-generated feedback comparable in quality to feedback guided by expert-authored task rubrics. We analyzed AI-generated feedback for written scientific explanations produced by 207 middle school students in a chemistry task. Two pipelines were compared: (a) feedback guided by a human expert-designed, task-specific rubric, and (b) feedback guided by a task-specific rubric automatically derived from a learning progression prior to grading and feedback generation. Two human coders evaluated feedback quality using a multi-dimensional rubric assessing Clarity, Accuracy, Relevance, Engagement and Motivation, and Reflectiveness (10 sub-dimensions). Inter-rater reliability was high, with percent agreement ranging from 89% to 100% and Cohen's kappa values for estimable dimensions (kappa = .66 to .88). Paired t-tests revealed no statistically significant differences between the two pipelines for Clarity (t1 = 0.00, p1 = 1.000; t2 = 0.84, p2 = .399), Relevance (t1 = 0.28, p1 = .782; t2 = -0.58, p2 = .565), Engagement and Motivation (t1 = 0.50, p1 = .618; t2 = -0.58, p2 = .565), or Reflectiveness (t = -0.45, p = .656). These findings suggest that the LP-driven rubric pipeline can serve as an alternative solution.

Xin Xia, Nejla Yuruk, Yun Wang +1

University of Georgia · University of Georgia, Athens GA 30605, USA · Gazi University +1

Educational Feedback Generation Generative AI in Education Rubric-Guided Generation Educational Assessment Generative AI Evaluation

Mar 3, 2026

UniG2U-Bench: Do Unified Models Advance Multimodal Understanding?

Unified multimodal models have recently demonstrated strong generative capabilities, yet whether and when generation improves understanding remains unclear. Existing benchmarks lack a systematic exploration of the specific tasks where generation facilitates understanding. To this end, we introduce UniG2U-Bench, a comprehensive benchmark categorizing generation-to-understanding (G2U) evaluation into 7 regimes and 30 subtasks, requiring varying degrees of implicit or explicit visual transformations. Extensive evaluation of over 30 models reveals three core findings: 1) Unified models generally underperform their base Vision-Language Models (VLMs), and Generate-then-Answer (GtA) inference typically degrades performance relative to direct inference. 2) Consistent enhancements emerge in spatial intelligence, visual illusions, or multi-round reasoning subtasks, where enhanced spatial and shape perception, as well as multi-step intermediate image states, prove beneficial. 3) Tasks with similar reasoning structures and models sharing architectures exhibit correlated behaviors, suggesting that generation-understanding coupling induces class-consistent inductive biases over tasks, pretraining data, and model architectures. These findings highlight the necessity for more diverse training data and novel paradigms to fully unlock the potential of unified multimodal modeling.

Zimo Wen, Boxiu Li, Wanbo Zhang +11

Microsoft Research Asia · Shanghai Jiao Tong University · Fudan University +2

Unified Multimodal Models Multimodal Model Evaluation Multimodal Understanding Generative AI Evaluation Multimodal Reasoning

Mar 2, 2026

Rich Insights from Cheap Signals: Efficient Evaluations via Tensor Factorization

Moving beyond evaluations that collapse performance across heterogeneous prompts toward fine-grained evaluation at the prompt level, or within relatively homogeneous subsets, is necessary to diagnose generative models' strengths and weaknesses. Such fine-grained evaluations, however, suffer from a data bottleneck: human gold-standard labels are too costly at this scale, while automated ratings are often misaligned with human judgment. To resolve this challenge, we propose a novel statistical model based on tensor factorization that merges cheap autorater data with a limited set of human gold-standard labels. Specifically, our approach uses autorater scores to pretrain latent representations of prompts and generative models, and then aligns those pretrained representations to human preferences using a small calibration set. This sample-efficient methodology is robust to autorater quality, more accurately predicts human preferences on a per-prompt basis than standard baselines, and provides tight confidence intervals for key statistical parameters of interest. We also showcase the practical utility of our method by constructing granular leaderboards based on prompt qualities and by estimating model performance solely from autorater scores, eliminating the need for additional human annotations.

Felipe Maia Polo, Aida Nematzadeh, Virginia Aglietti +2

University of Michigan · Google DeepMind

Generative AI Evaluation Automated Evaluation Human Preference Modeling

Mar 1, 2026

A Unified Framework to Quantify Cultural Intelligence of AI

As generative AI technologies are increasingly being launched across the globe, assessing their competence to operate in different cultural contexts is exigently becoming a priority. While recent years have seen numerous and much-needed efforts on cultural benchmarking, these efforts have largely focused on specific aspects of culture and evaluation. While these efforts contribute to our understanding of cultural competence, a unified and systematic evaluation approach is needed for us as a field to comprehensively assess diverse cultural dimensions at scale. Drawing on measurement theory, we present a principled framework to aggregate multifaceted indicators of cultural capabilities into a unified assessment of cultural intelligence. We start by developing a working definition of culture that includes identifying core domains of culture. We then introduce a broad-purpose, systematic, and extensible framework for assessing cultural intelligence of AI systems. Drawing on theoretical framing from psychometric measurement validity theory, we decouple the background concept (i.e., cultural intelligence) from its operationalization via measurement. We conceptualize cultural intelligence as a suite of core capabilities spanning diverse domains, which we then operationalize through a set of indicators designed for reliable measurement. Finally, we identify the considerations, challenges, and research pathways to meaningfully measure these indicators, specifically focusing on data collection, probing strategies, and evaluation metrics.

Sunipa Dev, Vinodkumar Prabhakaran, Rutledge Chin Feman +16

Google Research.

Cross-Cultural Language Model Evaluation Generative AI Evaluation Cultural Knowledge in Language Models

Feb 27, 2026

When Metrics Disagree: Automatic Similarity vs. LLM-as-a-Judge for Clinical Dialogue Evaluation

As Large Language Models (LLMs) are increasingly integrated into healthcare to address complex inquiries, ensuring their reliability remains a critical challenge. Recent studies have highlighted that generic LLMs often struggle in clinical contexts, occasionally producing misleading guidance. To mitigate these risks, this research focuses on the domain-specific adaptation of Llama-2-7B using the Low-Rank Adaptation (LoRA) technique. By injecting trainable low-rank matrices into the Transformer layers, we efficiently adapted the model using authentic patient-physician transcripts while preserving the foundational knowledge of the base model. Our objective was to enhance precision and contextual relevance in responding to medical queries by capturing the specialized nuances of clinical discourse. Due to the resource-intensive nature of large-scale human validation, the model's performance was evaluated through a dual-track framework: Track A utilized traditional lexical similarity metrics (e.g., BLEU, ROUGE), while Track B employed an "LLM-as-a-Judge" paradigm using GPT-4 for semantic assessment. Our results demonstrate that while the LoRA-enhanced model achieved significant improvements across all quantitative lexical dimensions, a profound disagreement surfaced in the GPT-4 evaluation, which marginally favored the baseline model's conversational flow. This metric divergence underscores a pivotal finding: traditional automated scores may not fully reflect clinical utility. Consequently, we propose that while automated metrics and LLM judges serve as valuable developmental proxies, rigorous validation by human medical experts remains an indispensable requirement for the safe deployment of LLMs in healthcare settings.

Bian Sun, Zhenjian Wang, Orvill de la Torre +1

Heinz College, Carnegie Mellon University · School of Computer Science, Carnegie Mellon University

Clinical Language Model Evaluation Generative AI Evaluation Domain Adaptation for LLMs Medical QA LLM-as-a-Judge

Feb 18, 2026

GICDM: Mitigating Hubness for Reliable Distance-Based Generative Model Evaluation

Generative model evaluation commonly relies on high-dimensional embedding spaces to compute distances between samples. We show that dataset representations in these spaces are affected by the hubness phenomenon, which distorts nearest-neighbor relationships and biases distance-based metrics. Building on the classical Iterative Contextual Dissimilarity Measure (ICDM), we introduce Generative ICDM (GICDM), a method to correct neighborhood estimation for both real and generated data. We introduce a multi-scale extension to improve empirical behavior. Extensive experiments on synthetic and real benchmarks demonstrate that GICDM resolves hubness-induced failures, restores reliable metric behavior, and improves alignment with human assessment.

Nicolas Salvy, Hugues Talbot, Bertrand Thirion

Inria, Palaiseau, France · CentraleSupélec, Gif-sur-Yvette, France · Université Paris-Saclay, Gif-sur-Yvette, France +1

Generative AI Evaluation k-Nearest Neighbors

Feb 16, 2026

Exposing Diversity Bias in Deep Generative Models: Statistical Origins and Correction of Diversity Error

Deep generative models have achieved great success in producing high-quality samples, making them a central tool across machine learning applications. Beyond sample quality, an important yet less systematically studied question is whether trained generative models faithfully capture the diversity of the underlying data distribution. In this work, we address this question by directly comparing the diversity of samples generated by state-of-the-art models with that of test samples drawn from the target data distribution, using recently proposed reference-free entropy-based diversity scores, Vendi and RKE. Across multiple benchmark datasets, we find that test data consistently attains substantially higher Vendi and RKE diversity scores than the generated samples, suggesting a systematic downward diversity bias in modern generative models. To understand the origin of this bias, we analyze the finite-sample behavior of entropy-based diversity scores and show that their expected values increase with sample size, implying that diversity estimated from finite training sets could inherently underestimate the diversity of the true distribution. As a result, optimizing the generators to minimize divergence to empirical data distributions would induce a loss of diversity. Finally, we discuss potential diversity-aware regularization and guidance strategies based on Vendi and RKE as principled directions for mitigating this bias, and provide empirical evidence suggesting their potential to improve the results.

Farzan Farnia, Mohammad Jalali, Azim Ospanov

Department of Computer Science and Engineering, The Chinese University of Hong Kong.

Generative Modeling Generative AI Evaluation

Feb 2, 2026

The Verification Crisis: Expert Perceptions of GenAI Disinformation and the Case for Reproducible Provenance

The growth of Generative Artificial Intelligence (GenAI) has shifted disinformation production from manual fabrication to automated, large-scale manipulation. This article presents findings from the first wave of a longitudinal expert perception survey (N=21) involving AI researchers, policymakers, and disinformation specialists. It examines the perceived severity of multimodal threats -- text, image, audio, and video -- and evaluates current mitigation strategies. Results indicate that while deepfake video presents immediate "shock" value, large-scale text generation poses a systemic risk of "epistemic fragmentation" and "synthetic consensus," particularly in the political domain. The survey reveals skepticism about technical detection tools, with experts favoring provenance standards and regulatory frameworks despite implementation barriers. GenAI disinformation research requires reproducible methods. The current challenge is measurement: without standardized benchmarks and reproducibility checklists, tracking or countering synthetic media remains difficult. We propose treating information integrity as an infrastructure with rigor in data provenance and methodological reproducibility.

Alexander Loth, Martin Kappes, Marc-Oliver Pahl

Frankfurt University of Applied Sciences Frankfurt am Main, Germany · IMT Atlantique, UMR IRISA, Chaire Cyber CNI Rennes, France

Generative AI Evaluation Scientific Reproducibility

Jan 29, 2026

Industrialized Deception: The Collateral Effects of LLM-Generated Misinformation on Digital Ecosystems

Generative AI and misinformation research have evolved since our 2024 survey. This paper presents an updated perspective, transitioning from literature review to practical countermeasures. We report on changes in the threat landscape, including improved AI-generated content through Large Language Models (LLMs) and multimodal systems - addressing the adverse impacts of AI on information quality. This work focuses on our practical contributions: JudgeGPT, a platform to evaluate human perception of AI-generated news, and RogueGPT, a controlled stimulus generation engine for research. Together, these tools form an experimental pipeline for studying how humans perceive and detect AI-generated misinformation. Our findings show that detection capabilities have improved, but competition between generation and detection continues. We discuss mitigation strategies such as LLM-based detection, inoculation approaches, and the dual-use nature of generative AI.

Alexander Loth, Martin Kappes, Marc-Oliver Pahl

Frankfurt University of Applied Sciences Frankfurt am Main, Germany · IMT Atlantique, UMR IRISA, Chaire Cyber CNI Rennes, France

Fake News Detection Generative AI Evaluation Misinformation Diffusion

Jan 26, 2026

LLAMA LIMA: A Living Meta-Analysis on the Effects of Generative AI on Learning Mathematics

The capabilities of generative AI in mathematics education are rapidly evolving, posing significant challenges for research to keep pace. Research syntheses remain scarce and risk being outdated by the time of publication. We present a Living Meta-Analysis (LIMA) on the effects of generative AI-based interventions for learning mathematics. We update the literature base every two months and apply a cumulative Bayesian multilevel meta-regression model. This paper reports results from the fifth version, including 34 studies, 7 of which were newly included since the fourth version. Results indicate an increasingly positive effect (g = 0.57, CrI [0.36, 0.79]) and no evidence in favor of a publication bias. Moderator analyses indicate that effects vary by integration mode and intervention duration: Larger effects were observed when generative AI supplemented teacher instruction and in longer interventions. Moderate evidence for stronger effects was observed in studies involving younger learners and more recent technology. For other moderators, evidence is still inconclusive.

Anselm Strohmaier, Samira Bödefeld, Oliver Straser +1

University of Education Freiburg, Institute of Mathematics Education

Generative AI in Education Generative AI Evaluation Mathematics Education

Jan 9, 2026

The Illusion of AI Expertise Under Uncertainty: A Probabilistic Paradigm to Interpret Results Under Ground Truth Variation

Benchmarking the capabilities of AI systems, including Large Language Models (LLMs) and Vision Models, typically ignores the impact of uncertainty in the underlying ground truth answers from experts. This ambiguity is not just limited to human preferences, but is also consequential even in safety critical domains such as medicine where uncertainty is pervasive. In this paper, we introduce a probabilistic paradigm to theoretically explain how high certainty in ground truth answers is almost always necessary for even an expert to achieve high scores. In contrast, in datasets with high variation in ground truth answers, the expert performance drops substantially resulting in almost no difference between a random labeller and an expert. This characteristic also manifests when comparing models, where uncertainty obfuscates differences between poor and high performing models. Therefore, ignoring uncertainty in ground truth evaluation data can result in the misleading conclusion that a non-expert has similar performance to that of an expert. Given ground truth answer variability and commonly used F1 and accuracy metrics, we bring forth a simple and explainable probabilistic coin toss model to estimate expected accuracyand expected F1 and compare the estimated scores with those that experts (human or model) achieve across multiple domains -- including radiology and models such as GPT-5 and Gemini-3. The results lead to the recommendation that stratification by the probability of the ground truth answer becomes critical when expert performance is relatively low. Under stratified evaluation, performance comparison becomes more reliable in high certainty bins, mitigating the effect of the key confounding factor --- uncertainty.

Aparna Elangovan, Lei Xu, Mahsa Elyasi +8

Independent Researcher · Amazon · Gazi University +6

LLM Evaluation Generative AI Evaluation Uncertainty Quantification

Jan 7, 2026

Quantifying the Effect of Test Set Contamination on Generative Evaluations

As frontier AI systems are pretrained on web-scale data, test set contamination has become a critical concern for accurately assessing their capabilities. While research has thoroughly investigated the impact of test set contamination on discriminative evaluations like multiple-choice question-answering, comparatively little research has studied the impact of test set contamination on generative evaluations. In this work, we quantitatively assess the effect of test set contamination on generative evaluations through the language model lifecycle. We pretrain language models on mixtures of web data and the MATH benchmark, sweeping model sizes and number of test set replicas contaminating the pretraining corpus; performance improves with contamination and model size. Using scaling laws, we make a surprising discovery: including even a single test set replica enables models to achieve lower loss than the irreducible error of training on the uncontaminated corpus. We then study further training: overtraining with fresh data reduces the effects of contamination, whereas supervised finetuning on the training set can either increase or decrease performance on test data, depending on the amount of pretraining contamination. Finally, at inference, we identify factors that modulate memorization: high sampling temperatures mitigate contamination effects, and longer solutions are exponentially more difficult to memorize than shorter ones, presenting a contrast with discriminative evaluations, where solutions are only a few tokens in length. By characterizing how generation and memorization interact, we highlight a new layer of complexity for trustworthy evaluation of AI systems.

Rylan Schaeffer, Joshua Kazdan, Baber Abbasi +8

Stanford Computer Science · Stanford Statistics · EleutherAI +3

Language Model Generation Evaluation Benchmark Contamination LLM Memory Memorization in Generative Models Data Contamination in Language Models LLM Training+1

Dec 11, 2025

Developing an LLM-Based Feedback System Grounded in Evidence-Centered Design to Support Physics Problem Solving

Generative AI offers new opportunities for individualized and adaptive learning, e.g., through large language model (LLM)-based feedback systems. While LLMs can produce factually correct feedback for relatively straightforward conceptual tasks, delivering high-quality feedback for tasks that require advanced domain expertise, such as physics problem solving, remains a substantial challenge. This study presents the design and implementation of an LLM-based feedback system for physics problem solving grounded in evidence-centered design and reports a first evaluation within the German Physics Olympiad. Participants rated the usefulness and correctness of the generated feedback for each implemented problem. The collected ratings indicate that the feedback was generally perceived as useful and highly correct. However, an in-depth analysis revealed that the feedback contained errors in 20% of cases; errors that often went unnoticed by the students. We discuss the risks associated with uncritical reliance on LLM-based feedback and outline potential directions for generating more adaptive and reliable LLM-based feedback in the future.

Holger Maus, Fabian Kieser, Stefan Petersen +2

Department of Physics Education, Leibniz Institute for Science and Mathematics Education Kiel, Germany · Department of Physics Education Research, Free University of Berlin Berlin, Germany · Department of Physics and Physics Education Research, Ludwigsburg University of Education Ludwigsburg, Germany

Educational Feedback Generation Educational Technology Generative AI in Education Generative AI Evaluation

Nov 26, 2025

Generalized Design Choices for Deepfake Detectors

The effectiveness of deepfake detection methods often depends less on their core design and more on implementation details such as data preprocessing, augmentation strategies, and optimization techniques. These factors make it difficult to fairly compare detectors and to understand which factors truly contribute to their performance. To address this, we systematically investigate how different design choices influence the accuracy and generalization capabilities of deepfake detection models, focusing on aspects related to training, inference, and incremental updates. By isolating the impact of individual factors, we aim to establish robust, architecture-agnostic best practices for the design and development of future deepfake detection systems. Our experiments identify a set of design choices that consistently improve deepfake detection and enable state-of-the-art performance on the AI-GenBench benchmark.

Lorenzo Pellegrini, Serafino Pandolfini, Davide Maltoni +3

Department of Computer Science and Engineering (DISI) University of Bologna, Italy · IdentifAI, Italy

Deep Learning Optimization Deepfake Detection Generative AI Evaluation

Nov 12, 2025

Generative AI Purpose-built for Social and Mental Health: A Real-World Pilot

Generative AI chatbots built for mental health could extend access to care, but evidence from real-world use is limited. We report a single-arm, naturalistic pilot of a foundation model trained for mental health, among 299 US adults with at least moderate depressive or anxiety symptoms who were followed for up to 12 months. Depression and anxiety symptoms fell by 10 weeks (Cohen's d 0.93 and 0.79), with loneliness, behavioral activation, and social interaction improving. Clinicians confirmed that automated safeguards were escalated appropriately. Participants fell into non-responding (57.2%), improving (37.1%) and rapidly improving (5.7%) trajectories. Early working alliance and greater engagement were associated with better outcomes. The AI deployed ten identified intervention families whose delivery varied with baseline anxiety and depression, with the overall ratio of clinical to non-clinical content increasing according to severity without survey information access. These findings support the feasibility of purpose-built AI for mental health.

Thomas D. Hull, Lizhe Zhang, Caitlin A. Stamatis +2

Slingshot AI · CREATIV Lab, Department of Psychiatry and Behavioral Sciences, University of Washington · Department of Psychiatry, NYU School of Medicine +1

Healthcare Mental Health Mental Health Conversational Agents Conversational Agents Generative AI Evaluation