To effectively enable people to understand new topics, explanations should be tailored to their backgrounds and abilities. Prompting alone has been shown to be insufficient for creating such explanations, and other computational methods are missing so far. Therefore, this paper investigates whether LLMs can be steered to generate explanations that are tailored to a specific group of people. To this end, we propose an approach that first identifies group-specific attributes in terms of explanatory style and knowledge of a specific target group. Building on activation engineering, it then computes attribute-based steering vectors and adds them to the internal activations of an LLM during inference to enable a fine-grained steering. In our experiments, we assess the steering effectiveness in terms of specificity and factuality of the generated explanations. Additionally, we evaluate the explanations in a study with human experts from different target groups. Compared to prompting and state-of-the-art steering baselines, our approach tailors the explanations significantly better to the target group while maintaining the best specificity-factuality balance.
Mingqi Gao, Anthony Sicilia, Weiyan Shics.CL stat.ML
Across various non-verifiable tasks, human evaluation is reliable but expensive, while automatic metrics are more scalable but often biased. Building on prediction-powered inference (PPI), we propose prediction-powered evaluation, a framework that combines limited human judgments with large-scale automatic scores to obtain data-efficient system comparisons that are provably unbiased. We develop parametric and non-parametric procedures, analyze the efficiency trade-off between paired and unpaired designs, and validate the framework on six WMT datasets. We further introduce the Prediction-Powered Saving Ratio (PPSR), a meta-metric that measures how much human annotation an automatic metric can save when used within prediction-powered evaluation. PPSR directly targets metric utility for prediction-powered evaluation and yields more discriminative and stable metric rankings than existing system-level meta-metrics. Overall, our new paradigm reframes automatic metrics as tools for reducing human annotation cost rather than replacing human judgment, and applies broadly to non-verifiable tasks.
Alessandro Tutone, Giorgio Franceschelli, Mirco Musolesics.CL cs.AI cs.CY
Large Language Models (LLMs) are increasingly capable of generating text that challenges human performance in domains requiring creativity, yet evaluating creativity in LLM-generated content remains a significant challenge. Here, we investigate whether current automatic evaluation methods can reliably capture human judgments of creativity. We collect human evaluations of human- and AI-generated short stories from the WritingPrompts dataset across 11 dimensions of creativity, and compare these judgments with automated objective metrics and LLM-as-a-Judge evaluations. Our experiments reveal substantial misalignment between automatic evaluations and human assessments. In particular, LLM-based judges exhibit a systematic preference for AI-generated stories, consistently favoring their stylistic characteristics over the unpredictability and other qualities of human-authored texts. Furthermore, correlation analyses show that widely used automatic metrics exhibit near-zero alignment with human judgments across both human- and AI-generated stories, suggesting that they fail to capture important dimensions of creativity. These findings highlight fundamental limitations in current approaches to the automatic evaluation of creative text and underscore the difficulty of reducing the multidimensional and subjective nature of creativity to computational metrics.
Readable AI output can leave an evaluability gap: even when the source is shown, an overall-quality judgment may not reflect what an output preserves. We investigated how source-text condition and output rendering relate to perceived translation quality, and how output and system appraisals relate to trust and stated disclosure willingness in a plain-text interface. A focal 2 * 2 comparison (N=306) using TransLingo examined simple generated narratives and complex literary-philosophical prose alongside LLM-generated readability-oriented outputs and researcher-revised fidelity-oriented outputs. A descriptive stimulus audit indicated greater source retention in fidelity-oriented outputs in both source-text conditions. Factorial analyses showed a significant rendering-by-source-text-condition interaction in perceived quality. Participants rated fidelity-oriented outputs higher than readability-oriented outputs for the simple narratives, whereas no reliable rendering difference emerged for the complex prose. A corresponding source-condition-dependent pattern was observed for perceived intelligence, agency-oriented anthropomorphic attribution, and task-performance trust. A separate theory-ordered appraisal-structure SEM characterized concurrent associations among perceived quality, perceived intelligence, agency-oriented anthropomorphic attribution, task-performance trust, and stated disclosure willingness across six domains, with task-performance trust as the proximal correlate of stated willingness. The observed rating pattern distinguishes source access from source evaluability: for the complex stimuli, displaying the source did not ensure that one overall-quality rating reflected differences in retained content. It also separates support for evaluating translation output from data-handling support for decisions about what personal text to entrust to a system.
The majority of work on summarization evaluation focuses on general summary quality (e.g., ROUGE, BERTScore) or specific desired properties (e.g., readability, factuality). However, these metrics fail to measure the utility of a summary to an individual user. For example, a biomedical researcher learning about the latest vaccine research will have different informational needs from a family doctor. Query-focused summarization captures part of this need, but in practice, users rarely state everything relevant in a query: a single short query is likely inadequate to distinguish the needs of a researcher from those of a physician. By contrast, a reader's background or persona (their role and expertise) is comparatively stable across queries and recovers much of this missing context, which makes it a practical signal for assessing whether a summary satisfies that reader's needs. In this work, we assess how sensitive popular summarization metrics are to both informational and persona differences, and find that many popular metrics, including strong LLM-as-judge metrics, fail basic perturbation tests of informational content. We additionally conduct an expert human evaluation, measuring summary preferences based on information satisfaction given a specific person's background and use case. We find that both traditional and LLM-based metrics are insufficient measures of information satisfaction and agree poorly with human judgment.
This preprint presents the results of the fourth GENEA Challenge, a large-scale human evaluation of five speech-driven gesture-generation systems trained by participating teams on the Seamless Interaction dataset of dyadic conversations. As in the 2023 GENEA Challenge, we used a disentangled evaluation methodology to assess motion quality and speech alignment without confounding between the two, and performed a dyadic mismatching study to isolate the effect of listening and reacting to the interlocutor. We additionally introduce a new semantic gesture-generation task and a text-mismatching evaluation methodology using the Grounded Gestures subset of the data. In total, we ran four large-scale user studies, collecting over 23,000 votes from 869 test-takers. In the motion-realism study, the dataset's filtered segments had substantially higher motion quality than all challenge submissions (68-95% pairwise winrate). In the speech-alignment study, the motion-capture segments provided a conceptual ceiling at 62% alignment score, with the top submission significantly behind at 32% and the rest only slightly above the 0% expected of an input-independent system. In the dyadic study, motion capture again set the ceiling at 65% appropriateness score, but no submission scored substantially above chance, indicating that the systems could not yet respond to the interlocutor. Finally, the semantic mismatching evaluation found highly expressive gestures in the dataset (test-takers identified the matching transcript 79% of the time), yet almost all submissions failed to generate semantically expressive motion, with the best achieving only an 8% appropriateness score. The collected votes and outputs will be made publicly available at https://genea-workshop.github.io/2026/challenge/ to facilitate reproducibility and further research.
Focus particles such as "even" and "only" are central to formal semantic theories that posit structured representations over sets of alternatives. "Even" highlights unexpected or extreme alternatives, while "only" enforces exclusivity. If such scalar representations are robust and generalizable, they should give rise to consistent judgments across contexts and systems. In this work, we test whether humans and large language models (LLMs) construct stable scalar representations from sentences containing these particles. Using a dataset of approximately 100 items, participants and models were asked to make scalar judgments. Preliminary results suggest that similar outputs across humans and LLMs may arise from different underlying mechanisms.
A trained reinforcement learning policy does not determine the complete behavior that users encounter: deployment code still schedules, admits, suppresses, or replaces its proposed actions. We contribute \emph{runtime action interference} (RAI), an AI control mechanism that preserves policy parameters while regulating action pacing and filtering configured action patterns after inference. RAI releases a proposed action only when its cooldown condition is satisfied and its content detector does not flag the action; otherwise, it dispatches a no-op. The detector covers specified toxic behaviors, including worker-unit harassment, while the cooldown controls action rate. We implement RAI in a replication of AlphaStar actor.py and make the implementation and reproducibility materials available through an open source code repository. We deployed RAI in a \textit{StarCraft~II} human participant study that compared two presentations of the same opponent with high capability and rate limited actions; we withheld its capability claim in one presentation and disclosed it in the other. On response scales from 1 to 5, we observed pooled fairness, trust, and toxicity means of 3.90, 3.50, and 2.00 under claim withholding, compared with 2.62, 4.31, and 2.85 under disclosure. Disclosure corresponded with lower perceived fairness and higher perceived toxicity across every expertise group, whereas trust increased among novices and experts but decreased among intermediate participants. Our human evaluation therefore shows that perceptions of an opponent controlled through RAI can vary substantially with the capability information presented to users, even when the configured control remains constant. We conclude that human-computer evaluations must separate control within the execution stack from capability disclosure and assess fairness, trust, and toxicity as distinct dimensions of human experience.
Frontier LLMs are increasingly put to use on open-ended complex questions, different in nature from the ones they are typically evaluated on. We dedicate more than 4,000 human expert hours to evaluate a selection of six frontier LLMs on a member of this class of problems: EuroExec, our introduced human expert-based benchmark composed of 413 open-ended long-form European executive tasks authored by 47 vetted domain experts, each question drawn from experience in a real case. Every response is manually evaluated through a multi-attribute rubric, an item-specific checklist of requirements, and a preference rank ordering, extracting an aggregate metric "Solve Rate". The strongest model solves only 56.9% of tasks, while expert-written reference answers judged blindly are solved at near-ceiling levels and are preferred over every model response in 74% of direct rankings, placing frontier generative systems well below the professional standard of work they are already used for. We see that the best way to extract this kind of conclusion is by employing human evaluators, carefully checking their consistency through rigorous statistical analysis, and observe that automatic measurements also fall short when evaluating on this case of real-world open-ended problems with a subjective ground truth.
Vilém Zouhar, Julia Kreutzer, Alon Lavie +4cs.CL cs.LG
While human evaluation is the gold standard in many NLP tasks, it suffers from prohibitive costs and poor scalability. When identifying top-performing models, typical evaluation protocols waste effort by exhaustively evaluating all models on the entire benchmark, a safe but inefficient approach. In this work, we formalize multi-model human evaluation as a best-arm identification problem in a multi-armed bandit setup with correlated arms, where pulling an arm corresponds to human-evaluating a model. By sampling adaptively based on the intermediate model rankings obtained on the samples so far, we can focus the annotation budget on the most competitive models. We prove the optimality of the proposed algorithms and show that it improves discrimination between top-performing models. This makes evaluations faster, cheaper and more aligned with large-scale competition evaluation goals.
Rubric-based AI systems for thesis assessment use criterion weights to assign different levels of importance to evaluation criteria. These weights are typically defined through expert judgment, although little empirical evidence exists regarding how thesis supervisors actually prioritize evaluation criteria. Consequently, this study investigates supervisor-derived criterion weights in thesis assessment and evaluates their impact on AI-based assessment. We surveyed 84 thesis supervisors across four academic disciplines and collected weighting data for 35 thesis assessment criteria. Comparison with the default criterion weights of the AI assessment system RubiSCoT [1] revealed substantial divergences between supervisor-derived and default criterion weights. To evaluate the practical implications of these differences, the supervisor-derived weights were integrated into multiple calibration configurations and evaluated on a corpus of 80 German-language theses. The best-performing configuration reduced the mean relative deviation between AI-generated and supervisor-assigned evaluations from 11.18% to 10.85%, although the improvement was not statistically significant. Human supervisors showed substantially stronger agreement with each other, exhibiting a mean inter-supervisor relative deviation of 4.44%. The findings indicate that criterion-weight calibration alone does not substantially improve alignment between AI-generated and human assessments.
Span-guided rewriting aims to preserve meaning by localizing edits to annotated harmful spans, but the same constraint can leave harmful intent insufficiently mitigated. We present a controlled exploratory comparison of span-guided and unguided detoxification on a mixed-source English evaluation set comprising manually curated inputs and HateXplain test items. We conduct a dense blinded human evaluation under a fixed single-generator setting. Human preferences reveal a trade-off rather than a uniformly superior rewriting strategy. Span-guided outputs are favored when localized editing preserves the original stance and avoids unnecessary modification, whereas unguided outputs are favored when broader rewriting achieves more complete mitigation. This contrast varies substantially across the study-defined strata: the two strategies are competitive in the strong stratum, while unguided rewriting is clearly preferred in the mild stratum. Rationale annotations trace this difference to complementary failure risks: residual harm after localized editing and over-modification after broader rewriting. We treat automatic evaluation as a diagnostic rather than a substitute for human judgment. Toxicity-similarity scalarizations, a multi-generator analysis, and two general-purpose LLM judges reproduce parts of the aggregate tendency but do not yield an analogous stratified contrast. These setting-specific findings do not establish a severity-based routing rule. Instead, they motivate evaluation protocols that assess mitigation sufficiency and meaning preservation separately and report both residual harm and over-modification alongside aggregate scores.
Vilém Zouhar, Roman Grundkiewicz, Sara Rajaee +6cs.CL cs.HC
Current human evaluation of machine translation typically assesses single outputs in isolation, a paradigm that suffers from high annotator noise and cost. We introduce Contrastive Error Span Annotation (cESA), a protocol that presents multiple translations of the source input (text, video, audio, image). In cESA, the annotator sees multiple translations of the same document, marks major and minor error spans, and then assigns a score from 0% to 100% on absolute scale. By allowing annotators to access the shared context across multiple outputs, cESA facilitates more consistent and efficient judgments. We validate cESA using a large-scale human evaluation of English->Japanese translations of 12 models, demonstrating reductions in annotation time and noise compared to standard pointwise evaluation. Unlike existing contrastive ranking methods, cESA yields absolute quality judgments that enable simple, interpretable non-parametric model rankings without the need for post-hoc corrections.
Emotional image editing requires more than applying affective filters or modifying predefined visual factors: an effective edit must identify what a particular image can afford for a target emotion. Existing affective image manipulation methods, including recent agentic variants, largely operate within bounded strategy spaces based on predefined factor taxonomies, knowledge libraries, or conventional editing templates, and therefore often miss image-specific, context-grounded strategies. We introduce EmoScope, a multi-agent framework that reframes the task from "how should I edit?" to "what can I edit?" EmoScope first discovers an image-specific editable space through emotion-conditioned affordance reasoning, then uses a semantic hierarchy of anchors, variables, and context to balance content consistency and emotional expressiveness before executing and verifying the edit. Because its plans are expressed as image-specific affordances rather than retrieved templates, EmoScope also exposes the editing strategy as an interactive surface for user refinement at the plan level. In a large-scale human evaluation covering all eight Mikels emotion categories, with 4,693 valid responses across 1,824 pairwise questions, participants preferred EmoScope over two competitive baselines by 88.1% on average. Attribution analysis further shows that EmoScope selects target-emotion-adaptive strategies rather than applying a uniform template. The same affordance-level plan also supports lightweight user refinement in an interactive pilot. Finally, we show that classifier-based metrics exhibit emotion-conditional blind spots toward non-stereotypical, context-grounded edits, and present a relative content-emotion preference-affinity landscape showing that EmoScope's advantage varies systematically across image-emotion combinations.
In the field of Artificial Intelligence, an agent is a system which is able to autonomously make decisions in order to reach a desired goal. As these systems grow more prevalent in our day-to-day lives, there has been an increased need to add explainability features which can provide an account for an agent's behavior. We therefore propose a framework that outlines how to produce comprehensible explanations for policy-aware agents, or agents which have rule-enforcing policies incorporated in their decision-making framework. This framework is designed using insights from the social sciences on how to produce good explanations. It is implemented in the Answer Set Programming language while using Python to assist with information extraction and natural-language translation. Because these agents incur penalties when violating policies, we are able to leverage these penalties to detect undesirable events in scenarios that are counterfactual to the agents' original actions. This lends itself to creating contrastive explanations (e.g., "the agent performed this action because, had it not, undesirable event X would have occurred."), which formulate the core component for our explainability framework. The framework is evaluated using a survey wherein human participants provide feedback on our program-generated explanations.
Two competing perspectives on fluid intelligence (gf) measures propose that performance is primarily constrained either by working memory capacity or by the ability to induce novel relations. The first perspective is currently dominant in measurement, as evident from the use of a limited set of recurring rules, whereas the second perspective is reflected in many definitions but rarely present in measurement. The ARC-AGI benchmark predominantly requires rule induction and was proposed as a measure of gf for both humans and artificial systems. However, its psychometric properties have not yet been examined in human samples. We therefore investigated the psychometric characteristics and nomological network of ARC-AGI in a first study with 100 participants. A compilation of ARC-AGI items showed good psychometric properties and correlated substantially with figural fluid intelligence as measured by a figural reasoning test (\r{ho} = .63). Associations with figural originality were weak. These findings provide initial support for the validity of ARC-AGI as a measure of human fluid intelligence. Future research should include more rule induction tasks as well as additional multivariate covariates. This study is unusual by studying a task in humans that was initially designed for machines. We suggest systematically embedding AI benchmarks into the nomological network of human cognitive abilities to enable more systematic evaluation and interdisciplinary cooperation.
Large language models (LLMs) are used worldwide, yet disproportionately reflect Western values, limiting their ability to represent diverse value systems. We introduce PLURAL, a large-scale, value-focused preference dataset grounded in the Integrated Values Survey (IVS), a nationally representative survey spanning 92 countries. Using a two-stage generation pipeline, we transform survey responses into synthetic preference triplets that preserve normative value signals while producing realistic scenarios. We release an initial version of PLURAL containing ~500,000 preference triplets representing people in 20 diverse countries. We evaluate PLURAL in three ways: (i) dataset-level validation showing that it preserves both cross-country value differences and within-country diversity from the original survey; (ii) automated evaluation showing that training on PLURAL improves alignment with target countries' cultural profiles, reducing mean absolute error by up to 27.7% relative to strong baselines; and (iii) blind human evaluation with 176 evaluators in India, Brazil, and Japan, who judge PLURAL-aligned responses as more representative of their national values. Together, these results show that PLURAL contains learnable signal for value steering, offering a scalable resource for pluralistic alignment. Dataset: https://huggingface.co/datasets/agdhruv/plural-alignment
Yves Ferstler, Adam Podoxin, Ty Brassington +3cs.CL
AI translation of literary works is increasingly common. While the content may be rendered adequately, we do not know enough about how readers experience it in terms of immersiveness and literary effect, aspects poorly captured by automatic machine translation metrics or human evaluation targeting fluency and adequacy. We ask 15 avid readers to compare recently published human translations (HT) to machine translations (MT) generated with an agentic large language model (LLM)-based pipeline, for 15 recent novels in French, Polish, and Japanese and translated into English. Readers evaluated approximately 8K-word excerpts in two conditions: immersive reading of the whole excerpt (30 comparisons) and close reading of 386 aligned HT-MT chunk pairs (772 comparisons), with two readers per book and in alternating order of presentation. Overall, readers find MT "fine", but prefer HT (slightly at excerpt-level 19/30, more clearly at chunk-level 522/772) for its ease, clarity, and immersive nature. Readers' highlights show that MT's quality varies more within one book than HT's does. Crucially, readers cannot reliably tell the two apart (17/30 guess correctly) and tend to prefer the version they believe to be human. Automatic metrics, including LLM-as-a-judge approaches, fail to recover reader preferences and favor MT. We release LAIT (Literary AI Translation), a reader-centered evaluation dataset with 1K reader comments, 2K judgments and preference ratings, and 7.2K span-level annotations, along with our evaluation protocol and supporting interface.
Ümit Mert Çağlar, Alptekin Temizeleess.IV cs.AI cs.CV cs.LG
Volume and quality of datasets are crucial for deep learning model training, yet they are often constrained by availability and data acquisition costs. Synthetic data augmentation can extend existing datasets with realistic images, and the quality of these images is generally assessed through fidelity metrics such as FID, KID, IS, LPIPS and SSIM that measure structural or distributional similarity. However, such metrics, including the widely used FID, focus on visual fidelity without reflecting downstream utility, and can diverge from human perception under perturbations that are imperceptible to human observers. In this work, we systematically evaluate Earth observation datasets alongside synthetic counterparts generated by deep generative models, comparing automatic metrics against human perception and downstream tasks. Our results reveal a stark misalignment: semantics-preserving perturbations such as rotation drastically alter metric scores while leaving human recognition unaffected, and synthetic samples that score poorly on automatic metrics achieve comparable or higher perceived realism, and can improve downstream performance when combined with real data. By benchmarking semantic segmentation models trained on mixed real-synthetic datasets, we demonstrate that quality metrics rooted in ImageNet-pretrained feature spaces are unreliable indicators for geospatial data. Our findings underscore that automatic quality evaluation of synthetic datasets should be grounded in downstream task performance and human evaluation.
Background. Large language models and AI agents are increasingly used to support biomedical research, but native model outputs may omit key analytical steps, misuse methods, or overstate conclusions. We evaluated whether autonomous access to a medical research skill package was associated with higher-quality AI-generated transcriptomic research-analysis outputs compared with native AI without skills. Methods. We conducted an exploratory multi-model human evaluation using a non-small cell lung cancer immunotherapy biomarker task. Six model backbones were tested. The evaluation included 21 anonymized outputs: 9 native-AI outputs and 12 skill-augmented outputs generated through an AI agent implementation represented by OpenClaw. Four non-expert biomedical reviewers and two blinded experts evaluated each output, with two ratings from each reviewer type. The primary outcome was expert-rated overall quality. Results. Skill-augmented outputs showed directionally higher expert overall quality than native-AI outputs (mean 5.50 vs 5.11; difference=0.39; bootstrap 95\% CI, -0.04 to 0.90; Welch p=0.156). Non-expert reviewer quality showed the same direction (mean 4.72 vs 4.47; difference=0.26; bootstrap 95\% CI, -0.25 to 0.80; Welch p=0.373). Expert agreement was limited (single-rating ICC=-0.15), and model-specific effects were descriptive and heterogeneous. Conclusions. Autonomous skill access showed a directional quality signal in this exploratory sample, but the signal was smaller than expert-rating noise and should not be interpreted as confirmatory evidence. The findings primarily motivate larger evaluations of skill-augmented AI agents with stronger reliability controls, platform replication, and biological-validity assessment.
Human evaluation plays a critical role in assessing the quality of generated text. However, the reliability and reproducibility of these evaluations depend on transparent and well-documented protocols -- details that are frequently missing in current practice. In this work, we conduct a large-scale analysis of human evaluation protocols for evaluating long-form generation tasks in *CL conference publications from 2023--2025, including a full manual review of 284 papers and LLM-assisted analysis for another 1.8k+ papers. We define a set of 20 reportable criteria related to reproducibility of human evaluation studies, and apply these criteria to systematically examine reporting norms and practices within the community. We find widespread under-reporting of important aspects of human evaluation study design, leading to ambiguity about what was measured and how, who contributed judgments, and how judgments should be interpreted. Based on these findings, we outline actionable recommendations to support more transparent and reproducible reporting in future research. Our analysis code and annotated dataset can be found at: https://github.com/larchlab/Illusions-of-the-Gold-Standard
Lechen Zhang, Jiarui Liu, Tal Augustcs.CL cs.AI cs.HC
Despite growing interest, most evaluations of large language models' (LLMs') personalization abilities have relied on synthetic data. It remains unclear how well current personalization systems work for real users. In this paper, we study the gap in LLM personalization performance when using synthetic versus human data. We collect human conversations (550 conversations) and judgments across three stages of personalization: extracting user attributes from conversations (5,949 judgments), pairing relevant attributes with new prompts (11,919), and incorporating relevant attributes into a personalized response (1,101). Incorporating human data reveals system limitations at each stage. Models struggle to extract attributes from human conversations, disagree with human judgments on relevant attributes, and generate personalized responses that humans judge no better than generic responses (though that LLM judges widely rate as better). We introduce two lightweight training-based interventions that shift automated personalization evaluation closer to human data in our first two stages. However, in our third stage we find that learned reward models achieve only modest correlation with human ratings, suggesting that human-aligned personalization quality judgments are difficult to model directly. Our collected data provides a foundation for studying how models should extract, select, and incorporate user information in ways that humans find useful.
Human evaluation remains the primary standard for assessing modern AI systems, yet annotator disagreement, bias, and variability make system rankings fragile under standard majority vote aggregation. Majority vote discards annotator reliability and item-level ambiguity, often yielding unstable comparisons across annotator subsets. We introduce STABLEVAL, a disagreement-aware evaluation framework that models latent item correctness and annotator-specific confusion patterns to produce posterior expected item credit and calibrated agent-level scores. Unlike label-denoising approaches such as Dawid-Skene, STABLEVAL is explicitly designed for stable and uncertainty-aware system evaluation rather than hard label recovery. We formalize ranking stability as a first-class evaluation objective and analyze how aggregation methods preserve or distort underlying annotator behavior. Across controlled synthetic experiments and multiple real-world human-annotated benchmarks, majority vote exhibits increasing score error and ranking instability under annotator heterogeneity and adversarial noise, while STABLEVAL yields more stable and statistically grounded system rankings. These results demonstrate that modeling disagreement is essential for robust and reproducible AI evaluation.