Daily Papers

Large language models (LLMs) are increasingly evaluated and sometimes trained using automated graders such as LLM-as-judges that output scalar scores or preferences. While convenient, these approaches are often opaque: a single score rarely explains why an answer is good or bad, which requirements were missed, or how a system should be improved. This lack of interpretability limits their usefulness for model development, dataset curation, and high-stakes deployment. Query-specific rubric-based evaluation offers a more transparent alternative by decomposing quality into explicit, checkable criteria. However, manually designing high-quality, query-specific rubrics is labor-intensive and cognitively demanding and not feasible for deployment. While previous approaches have focused on generating intermediate rubrics for automated downstream evaluation, it is unclear if these rubrics are both interpretable and effective for human users. In this work, we investigate whether LLMs can generate useful, instance-specific rubrics as compared to human-authored rubrics, while also improving effectiveness for identifying good responses. Through our systematic study on two rubric benchmarks, and on multiple few-shot and post-training strategies, we find that off-the-shelf LLMs produce rubrics that are poorly aligned with human-authored ones. We introduce a simple strategy, RubricRAG, which retrieves domain knowledge via rubrics at inference time from related queries. We demonstrate that RubricRAG can generate more interpretable rubrics both for similarity to human-authored rubrics, and for improved downstream evaluation effectiveness. Our results highlight both the challenges and a promising approach of scalable, interpretable evaluation through automated rubric generation.

Language models have steadily increased in size over the past few years. They achieve a high level of performance on various natural language processing (NLP) tasks such as question answering and summarization. Large language models (LLMs) have been used for generation and can now output human-like text. Due to this, there are other downstream tasks in the realm of dialog that can now harness the LLMs' language understanding capabilities. Dialog evaluation is one task that this paper will explore. It concentrates on prompting with LLMs: BLOOM, OPT, GPT-3, Flan-T5, InstructDial and TNLGv2. The paper shows that the choice of datasets used for training a model contributes to how well it performs on a task as well as on how the prompt should be structured. Specifically, the more diverse and relevant the group of datasets that a model is trained on, the better dialog evaluation performs. This paper also investigates how the number of examples in the prompt and the type of example selection used affect the model's performance.

Explaining the decisions of an Artificial Intelligence (AI) model is increasingly critical in many real-world, high-stake applications. Hundreds of papers have either proposed new feature attribution methods, discussed or harnessed these tools in their work. However, despite humans being the target end-users, most attribution methods were only evaluated on proxy automatic-evaluation metrics (Zhang et al. 2018; Zhou et al. 2016; Petsiuk et al. 2018). In this paper, we conduct the first user study to measure attribution map effectiveness in assisting humans in ImageNet classification and Stanford Dogs fine-grained classification, and when an image is natural or adversarial (i.e., contains adversarial perturbations). Overall, feature attribution is surprisingly not more effective than showing humans nearest training-set examples. On a harder task of fine-grained dog categorization, presenting attribution maps to humans does not help, but instead hurts the performance of human-AI teams compared to AI alone. Importantly, we found automatic attribution-map evaluation measures to correlate poorly with the actual human-AI team performance. Our findings encourage the community to rigorously test their methods on the downstream human-in-the-loop applications and to rethink the existing evaluation metrics.

Personality manipulation in large language models (LLMs) is increasingly applied in customer service and agentic scenarios, yet its mechanisms and trade-offs remain unclear. We present a systematic study of personality control using the Big Five traits, comparing in-context learning (ICL), parameter-efficient fine-tuning (PEFT), and mechanistic steering (MS). Our contributions are fourfold. First, we construct a contrastive dataset with balanced high/low trait responses, enabling effective steering vector computation and fair cross-method evaluation. Second, we introduce a unified evaluation framework based on within-run Delta analysis that disentangles, reasoning capability, agent performance, and demographic bias across MMLU, GAIA, and BBQ benchmarks. Third, we develop trait purification techniques to separate openness from conscientiousness, addressing representational overlap in trait encoding. Fourth, we propose a three-level stability framework that quantifies method-, trait-, and combination-level robustness, offering practical guidance under deployment constraints. Experiments on Gemma-2-2B-IT and LLaMA-3-8B-Instruct reveal clear trade-offs: ICL achieves strong alignment with minimal capability loss, PEFT delivers the highest alignment at the cost of degraded task performance, and MS provides lightweight runtime control with competitive effectiveness. Trait-level analysis shows openness as uniquely challenging, agreeableness as most resistant to ICL, and personality encoding consolidating around intermediate layers. Taken together, these results establish personality manipulation as a multi-level probe into behavioral representation, linking surface conditioning, parameter encoding, and activation-level steering, and positioning mechanistic steering as a lightweight alternative to fine-tuning for both deployment and interpretability.

Pre-trained models (PTMs) have achieved great success in various Software Engineering (SE) downstream tasks following the ``pre-train then fine-tune'' paradigm. As fully fine-tuning all parameters of PTMs can be computationally expensive, a widely used solution is parameter-efficient fine-tuning (PEFT), which freezes PTMs while introducing extra parameters. Though work has been done to test PEFT methods in the SE field, a comprehensive evaluation is still lacking. This paper aims to fill in this gap by evaluating the effectiveness of five PEFT methods on eight PTMs and four SE downstream tasks. For different tasks and PEFT methods, we seek answers to the following research questions: 1) Is it more effective to use PTMs trained specifically on source code, or is it sufficient to use PTMs trained on natural language text? 2) What is the impact of varying model sizes? 3) How does the model architecture affect the performance? Besides effectiveness, we also discuss the efficiency of PEFT methods, concerning the costs of required training time and GPU resource consumption. We hope that our findings can provide a deeper understanding of PEFT methods on various PTMs and SE downstream tasks. All the codes and data are available at https://github.com/zwtnju/PEFT.git.

Reward models are used throughout the post-training of language models to capture nuanced signals from preference data and provide a training target for optimization across instruction following, reasoning, safety, and more domains. The community has begun establishing best practices for evaluating reward models, from the development of benchmarks that test capabilities in specific skill areas to others that test agreement with human preferences. At the same time, progress in evaluation has not been mirrored by the effectiveness of reward models in downstream tasks -- simpler direct alignment algorithms are reported to work better in many cases. This paper introduces RewardBench 2, a new multi-skill reward modeling benchmark designed to bring new, challenging data for accuracy-based reward model evaluation -- models score about 20 points on average lower on RewardBench 2 compared to the first RewardBench -- while being highly correlated with downstream performance. Compared to most other benchmarks, RewardBench 2 sources new human prompts instead of existing prompts from downstream evaluations, facilitating more rigorous evaluation practices. In this paper, we describe our benchmark construction process and report how existing models perform on it, while quantifying how performance on the benchmark correlates with downstream use of the models in both inference-time scaling algorithms, like best-of-N sampling, and RLHF training algorithms like proximal policy optimization.

Current applications of self-supervised learning to wireless channel representation often borrow paradigms developed for text and image processing, without fully addressing the unique characteristics and constraints of wireless communications. To bridge this gap, we introduce ContraWiMAE, Wireless Contrastive Masked Autoencoder, a transformer-based foundation model that unifies masked reconstruction and masked contrastive learning for wireless channel representation. Our key innovation is a new wireless-inspired contrastive objective that exploits the inherent characteristics of wireless environment, including noise, fading, and partial observability, as natural augmentation. Through extensive evaluation on unseen scenarios and conditions, we demonstrate our method's effectiveness in multiple downstream tasks, including cross-frequency beam selection, line-of-sight detection, and channel estimation. ContraWiMAE exhibits superior linear separability and adaptability in diverse wireless environments, demonstrating exceptional data efficiency and competitive performance compared with supervised baselines under challenging conditions. Comparative evaluations against a state-of-the-art wireless channel foundation model confirm the superior performance and data efficiency of our approach, highlighting its potential as a powerful baseline for future research in self-supervised wireless channel representation learning. To foster further work in this direction, we release the model weights and training pipeline for ContraWiMAE.

Recent studies have demonstrated the emerging capabilities of foundation models like ChatGPT in several fields, including affective computing. However, accessing these emerging capabilities is facilitated through prompt engineering. Despite the existence of some prompting techniques, the field is still rapidly evolving and many prompting ideas still require investigation. In this work, we introduce a method to evaluate and investigate the sensitivity of the performance of foundation models based on different prompts or generation parameters. We perform our evaluation on ChatGPT within the scope of affective computing on three major problems, namely sentiment analysis, toxicity detection, and sarcasm detection. First, we carry out a sensitivity analysis on pivotal parameters in auto-regressive text generation, specifically the temperature parameter T and the top-p parameter in Nucleus sampling, dictating how conservative or creative the model should be during generation. Furthermore, we explore the efficacy of several prompting ideas, where we explore how giving different incentives or structures affect the performance. Our evaluation takes into consideration performance measures on the affective computing tasks, and the effectiveness of the model to follow the stated instructions, hence generating easy-to-parse responses to be smoothly used in downstream applications.

Accurate and comprehensive video captions with consistent subject references are critical for downstream understanding and generation tasks. However, few existing benchmarks can objectively and comprehensively evaluate these properties across diverse durations and scenarios, thereby hindering the advancement of video captioning models. To bridge this gap, we propose CapRiCorn-1K, a comprehensive benchmark designed to evaluate both video captioning quality and subject referential consistency across long temporal horizons and diverse video domains. To accommodate varied evaluation needs, our benchmark supports both audiovisual and visual-only settings. Extensive experiments on CapRiCorn-1K reveal that current models generally struggle to generate accurate and comprehensive captions while maintaining consistent subject references. Moreover, as video duration increases, both the overall caption quality and subject referential consistency decline. Notably, our evaluation metrics exhibit strong correlations with the performance of downstream understanding and generation tasks conditioned on the generated captions, further validating their effectiveness. The project is available at https://github.com/xlchen0205/CapRiCorn-1K .

Embodied agents require robust and comprehensive 3D spatiotemporal representations to support spatial reasoning, manipulation understanding, and downstream decision making. However, existing robot data are typically captured from fixed or sparse viewpoints, providing only partial and view-dependent observations, which limits multi-view perception and generalization across viewpoints. Given the difficulty of collecting additional viewpoints in real-world settings, we propose Embody4D, a dedicated video-to-video world model for embodied scenarios to bridge this observation gap by transforming a monocular robot video into novel-view videos from flexible target camera viewpoints. First, to tackle training data scarcity, we introduce a 3D-aware compositional synthesis pipeline to curate a heterogeneous dataset compositing cross-embodiment robotic arms with diverse backgrounds, promoting broad generalization. Second, to enforce geometric stability, we devise a latent confidence-aware expert modulation strategy, which estimates the reliability of warped latent priors and adaptively routes regions to copy, repair, or inpaint experts for spatiotemporally consistent 4D generation. Finally, to enhance the fidelity of the manipulation, we incorporate an interaction-aware attention mechanism that explicitly attends to the robotic interaction regions. Extensive experiments show that Embody4D achieves state-of-the-art performance on visual evaluation benchmarks, while both simulated and real-world robotic experiments further demonstrate its effectiveness as a robust data engine for synthesizing high-fidelity, view-consistent videos that empower downstream robotic planning and learning.

Large-scale self-supervised pre-training has paved the way for one foundation model to handle many different vision tasks. Most pre-training methodologies train a single model of a certain size at one time. Nevertheless, various computation or storage constraints in real-world scenarios require substantial efforts to develop a series of models with different sizes to deploy. Thus, in this study, we propose a novel tri-branch self-supervised training framework, termed as POA (Pre-training Once for All), to tackle this aforementioned issue. Our approach introduces an innovative elastic student branch into a modern self-distillation paradigm. At each pre-training step, we randomly sample a sub-network from the original student to form the elastic student and train all branches in a self-distilling fashion. Once pre-trained, POA allows the extraction of pre-trained models of diverse sizes for downstream tasks. Remarkably, the elastic student facilitates the simultaneous pre-training of multiple models with different sizes, which also acts as an additional ensemble of models of various sizes to enhance representation learning. Extensive experiments, including k-nearest neighbors, linear probing evaluation and assessments on multiple downstream tasks demonstrate the effectiveness and advantages of our POA. It achieves state-of-the-art performance using ViT, Swin Transformer and ResNet backbones, producing around a hundred models with different sizes through a single pre-training session. The code is available at: https://github.com/Qichuzyy/POA.

Humans ponder before articulating complex sentence elements, enabling deeper cognitive processing through focused effort. In this work, we introduce this pondering process into language models by repeatedly invoking the forward process within a single token generation step. During pondering, instead of generating an actual token sampled from the prediction distribution, the model ponders by yielding a weighted sum of all token embeddings according to the predicted token distribution. The generated embedding is then fed back as input for another forward pass. We show that the model can learn to ponder in this way through self-supervised learning, without any human annotations. Our method is straightforward and can be seamlessly integrated with various existing language models. Experiments across three widely used open-source architectures-GPT-2, Pythia, and LLaMA-and extensive downstream task evaluations demonstrate the effectiveness and generality of our method. For language modeling tasks, pondering language models achieve performance comparable to vanilla models with twice the number of parameters. On 9 downstream benchmarks, our pondering-enhanced Pythia models significantly outperform the official Pythia models. Notably, pondering-enhanced Pythia-1B is comparable to TinyLlama-1.1B, which is trained on 10 times more data. The code is available at https://github.com/LUMIA-Group/PonderingLM.

Vision-Language Pretraining (VLP) models have recently successfully facilitated many cross-modal downstream tasks. Most existing works evaluated their systems by comparing the fine-tuned downstream task performance. However, only average downstream task accuracy provides little information about the pros and cons of each VLP method, let alone provides insights on how the community can improve the systems in the future. Inspired by the CheckList for testing natural language processing, we exploit VL-CheckList, a novel framework to understand the capabilities of VLP models. The proposed method divides the image-texting ability of a VLP model into three categories: objects, attributes, and relations, and uses a novel taxonomy to further break down these three aspects. We conduct comprehensive studies to analyze seven recently popular VLP models via the proposed framework. Results confirm the effectiveness of the proposed method by revealing fine-grained differences among the compared models that were not visible from downstream task-only evaluation. Further results show promising research direction in building better VLP models. Our data and code are available at: https://github.com/om-ai-lab/VL-CheckList.

Low-resource African languages pose unique challenges for natural language processing (NLP) tasks, including natural language generation (NLG). In this paper, we develop Cheetah, a massively multilingual NLG language model for African languages. Cheetah supports 517 African languages and language varieties, allowing us to address the scarcity of NLG resources and provide a solution to foster linguistic diversity. We demonstrate the effectiveness of Cheetah through comprehensive evaluations across seven generation downstream tasks. In five of the seven tasks, Cheetah significantly outperforms other models, showcasing its remarkable performance for generating coherent and contextually appropriate text in a wide range of African languages. We additionally conduct a detailed human evaluation to delve deeper into the linguistic capabilities of Cheetah. The introduction of Cheetah has far-reaching benefits for linguistic diversity. By leveraging pretrained models and adapting them to specific languages, our approach facilitates the development of practical NLG applications for African communities. The findings of this study contribute to advancing NLP research in low-resource settings, enabling greater accessibility and inclusion for African languages in a rapidly expanding digital landscape. We will publicly release our models for research.

Dense 3D reconstruction and tracking of dynamic scenes from monocular video remains an important open challenge in computer vision. Progress in this area has been constrained by the scarcity of high-quality datasets with dense, complete, and accurate geometric annotations. To address this limitation, we introduce Syn4D, a multiview synthetic dataset of dynamic scenes that includes ground-truth camera motion, depth maps, dense tracking, and parametric human pose annotations. A key feature of Syn4D is the ability to unproject any pixel into 3D to any time and to any camera. We conduct extensive evaluations across multiple downstream tasks to demonstrate the utility and effectiveness of the proposed dataset, including 4D scene reconstruction, 3D point tracking, geometry-aware camera retargeting, and human pose estimation. The experimental results highlight Syn4D's potential to facilitate research in dynamic scene understanding and spatiotemporal modeling.

Existing large language models have to run K times to generate a sequence of K tokens. In this paper, we present RecycleGPT, a generative language model with fast decoding speed by recycling pre-generated model states without running the whole model in multiple steps. Our approach relies on the observation that adjacent tokens in a sequence usually have strong correlations and the next token in a sequence can be reasonably guessed or inferred based on the preceding ones. Through theoretical evaluations and practical tests on downstream text generation tasks, we demonstrate the effectiveness of our approach in lowering inference latency, achieving up to 1.4x speedup while preserving high performance.

Visual programs are executable code generated by large language models to address visual reasoning problems. They decompose complex questions into multiple reasoning steps and invoke specialized models for each step to solve the problems. However, these programs are prone to logic errors, with our preliminary evaluation showing that 58% of the total errors are caused by program logic errors. Debugging complex visual programs remains a major bottleneck for visual reasoning. To address this, we introduce VDebugger, a novel critic-refiner framework trained to localize and debug visual programs by tracking execution step by step. VDebugger identifies and corrects program errors leveraging detailed execution feedback, improving interpretability and accuracy. The training data is generated through an automated pipeline that injects errors into correct visual programs using a novel mask-best decoding technique. Evaluations on six datasets demonstrate VDebugger's effectiveness, showing performance improvements of up to 3.2% in downstream task accuracy. Further studies show VDebugger's ability to generalize to unseen tasks, bringing a notable improvement of 2.3% on the unseen COVR task. Code, data and models are made publicly available at https://github.com/shirley-wu/vdebugger/

We present TransactionGPT (TGPT), a foundation model for consumer transaction data within one of world's largest payment networks. TGPT is designed to understand and generate transaction trajectories while simultaneously supporting a variety of downstream prediction and classification tasks. We introduce a novel 3D-Transformer architecture specifically tailored for capturing the complex dynamics in payment transaction data. This architecture incorporates design innovations that enhance modality fusion and computational efficiency, while seamlessly enabling joint optimization with downstream objectives. Trained on billion-scale real-world transactions, TGPT significantly improves downstream classification performance against a competitive production model and exhibits advantages over baselines in generating future transactions. We conduct extensive empirical evaluations utilizing a diverse collection of company transaction datasets spanning multiple downstream tasks, thereby enabling a thorough assessment of TGPT's effectiveness and efficiency in comparison to established methodologies. Furthermore, we examine the incorporation of LLM-derived embeddings within TGPT and benchmark its performance against fine-tuned LLMs, demonstrating that TGPT achieves superior predictive accuracy as well as faster training and inference. We anticipate that the architectural innovations and practical guidelines from this work will advance foundation models for transaction-like data and catalyze future research in this emerging field.

In recent years, increasing attention has been directed to leveraging pre-trained vision models for motor control. While existing works mainly emphasize the importance of this pre-training phase, the arguably equally important role played by downstream policy learning during control-specific fine-tuning is often neglected. It thus remains unclear if pre-trained vision models are consistent in their effectiveness under different control policies. To bridge this gap in understanding, we conduct a comprehensive study on 14 pre-trained vision models using 3 distinct classes of policy learning methods, including reinforcement learning (RL), imitation learning through behavior cloning (BC), and imitation learning with a visual reward function (VRF). Our study yields a series of intriguing results, including the discovery that the effectiveness of pre-training is highly dependent on the choice of the downstream policy learning algorithm. We show that conventionally accepted evaluation based on RL methods is highly variable and therefore unreliable, and further advocate for using more robust methods like VRF and BC. To facilitate more universal evaluations of pre-trained models and their policy learning methods in the future, we also release a benchmark of 21 tasks across 3 different environments alongside our work.

Despite the success of large language models (LLMs) on general-purpose tasks, their performance in highly specialized domains such as biomedicine remains unsatisfactory. A key limitation is the inability of LLMs to effectively leverage biomedical tools, which clinical experts and biomedical researchers rely on extensively in daily workflows. While recent general-domain tool-calling datasets have substantially improved the capabilities of LLM agents, existing efforts in the biomedical domain largely rely on in-context learning and restrict models to a small set of tools. To address this gap, we introduce BioTool, a comprehensive biomedical tool-calling dataset designed for fine-tuning LLMs. BioTool comprises 34 frequently used tools collected from the NCBI, Ensembl, and UniProt databases, along with 7,040 high-quality, human-verified query-API call pairs spanning variation, genomics, proteomics, evolution, and general biology. Fine-tuning a 4-billion-parameter LLM on BioTool yields substantial improvements in biomedical tool-calling performance, outperforming cutting-edge commercial LLMs such as GPT-5.1. Furthermore, human expert evaluations demonstrate that integrating a BioTool-fine-tuned tool caller significantly improves downstream answer quality compared to the same LLM without tool usage, highlighting the effectiveness of BioTool in enhancing the biomedical capabilities of LLMs. The full dataset and evaluation code are available at https://github.com/gxx27/BioTool

The diffusion models, in early stages focus on constructing basic image structures, while the refined details, including local features and textures, are generated in later stages. Thus the same network layers are forced to learn both structural and textural information simultaneously, significantly differing from the traditional deep learning architectures (e.g., ResNet or GANs) which captures or generates the image semantic information at different layers. This difference inspires us to explore the time-wise diffusion models. We initially investigate the key contributions of the U-Net parameters to the denoising process and identify that properly zeroing out certain parameters (including large parameters) contributes to denoising, substantially improving the generation quality on the fly. Capitalizing on this discovery, we propose a simple yet effective method-termed ``MaskUNet''- that enhances generation quality with negligible parameter numbers. Our method fully leverages timestep- and sample-dependent effective U-Net parameters. To optimize MaskUNet, we offer two fine-tuning strategies: a training-based approach and a training-free approach, including tailored networks and optimization functions. In zero-shot inference on the COCO dataset, MaskUNet achieves the best FID score and further demonstrates its effectiveness in downstream task evaluations. Project page: https://gudaochangsheng.github.io/MaskUnet-Page/

Knowledge graphs automatically constructed from text are increasingly used in real-world applications. However, their inherent noise, fragmentation, and semantic inconsistencies significantly affect the performance of Graph Neural Networks (GNNs) on downstream tasks. Assessing their performance and robustness remains difficult, as it is often unclear whether observed results stem from the learning model or from the quality of the constructed graph itself. In this work, we introduce a dual-purpose benchmark designed to jointly evaluate (i) the performance of GNNs on noisy, text-derived graphs and (ii) the effectiveness of graph construction methods on a downstream task. The benchmark is built in the biomedical domain from a single textual corpus and includes two automatically constructed graphs generated using different extraction methods, alongside a high-quality reference graph curated by experts that serves as an upper performance bound. This design enables controlled comparison of construction methods and systematic evaluation of GNN robustness through semi-supervised node classification. We further provide a standardized, reproducible, and extensible evaluation framework, facilitating the integration of new graph extraction methods and learning models.

Human social behaviors are inherently multimodal necessitating the development of powerful audiovisual models for their perception. In this paper, we present Social-MAE, our pre-trained audiovisual Masked Autoencoder based on an extended version of Contrastive Audio-Visual Masked Auto-Encoder (CAV-MAE), which is pre-trained on audiovisual social data. Specifically, we modify CAV-MAE to receive a larger number of frames as input and pre-train it on a large dataset of human social interaction (VoxCeleb2) in a self-supervised manner. We demonstrate the effectiveness of this model by finetuning and evaluating the model on different social and affective downstream tasks, namely, emotion recognition, laughter detection and apparent personality estimation. The model achieves state-of-the-art results on multimodal emotion recognition and laughter recognition and competitive results for apparent personality estimation, demonstrating the effectiveness of in-domain self-supervised pre-training. Code and model weight are available here https://github.com/HuBohy/SocialMAE.

Non-coding RNA structure and function are essential to understanding various biological processes, such as cell signaling, gene expression, and post-transcriptional regulations. These are all among the core problems in the RNA field. With the rapid growth of sequencing technology, we have accumulated a massive amount of unannotated RNA sequences. On the other hand, expensive experimental observatory results in only limited numbers of annotated data and 3D structures. Hence, it is still challenging to design computational methods for predicting their structures and functions. The lack of annotated data and systematic study causes inferior performance. To resolve the issue, we propose a novel RNA foundation model (RNA-FM) to take advantage of all the 23 million non-coding RNA sequences through self-supervised learning. Within this approach, we discover that the pre-trained RNA-FM could infer sequential and evolutionary information of non-coding RNAs without using any labels. Furthermore, we demonstrate RNA-FM's effectiveness by applying it to the downstream secondary/3D structure prediction, SARS-CoV-2 genome structure and evolution prediction, protein-RNA binding preference modeling, and gene expression regulation modeling. The comprehensive experiments show that the proposed method improves the RNA structural and functional modelling results significantly and consistently. Despite only being trained with unlabelled data, RNA-FM can serve as the foundational model for the field.

In this paper, we consider the problem of composed image retrieval (CIR), it aims to train a model that can fuse multi-modal information, e.g., text and images, to accurately retrieve images that match the query, extending the user's expression ability. We make the following contributions: (i) we initiate a scalable pipeline to automatically construct datasets for training CIR model, by simply exploiting a large-scale dataset of image-text pairs, e.g., a subset of LAION-5B; (ii) we introduce a transformer-based adaptive aggregation model, TransAgg, which employs a simple yet efficient fusion mechanism, to adaptively combine information from diverse modalities; (iii) we conduct extensive ablation studies to investigate the usefulness of our proposed data construction procedure, and the effectiveness of core components in TransAgg; (iv) when evaluating on the publicly available benckmarks under the zero-shot scenario, i.e., training on the automatically constructed datasets, then directly conduct inference on target downstream datasets, e.g., CIRR and FashionIQ, our proposed approach either performs on par with or significantly outperforms the existing state-of-the-art (SOTA) models. Project page: https://code-kunkun.github.io/ZS-CIR/

Representation engineering methods have recently shown promise for enabling efficient steering of model behavior. However, evaluation pipelines for these methods have primarily relied on subjective demonstrations, instead of quantitative, objective metrics. We aim to take a step towards addressing this issue by advocating for four properties missing from current evaluations: (i) contexts sufficiently similar to downstream tasks should be used for assessing intervention quality; (ii) model likelihoods should be accounted for; (iii) evaluations should allow for standardized comparisons across different target behaviors; and (iv) baseline comparisons should be offered. We introduce an evaluation pipeline grounded in these criteria, offering both a quantitative and visual analysis of how effectively a given method works. We use this pipeline to evaluate two representation engineering methods on how effectively they can steer behaviors such as truthfulness and corrigibility, finding that some interventions are less effective than previously reported.

Predictable behavior from scaling advanced AI systems is an extremely desirable property. Although a well-established literature exists on how pretraining performance scales, the literature on how particular downstream capabilities scale is significantly muddier. In this work, we take a step back and ask: why has predicting specific downstream capabilities with scale remained elusive? While many factors are certainly responsible, we identify a new factor that makes modeling scaling behavior on widely used multiple-choice question-answering benchmarks challenging. Using five model families and twelve well-established multiple-choice benchmarks, we show that downstream performance is computed from negative log likelihoods via a sequence of transformations that progressively degrade the statistical relationship between performance and scale. We then reveal the mechanism causing this degradation: downstream metrics require comparing the correct choice against a small number of specific incorrect choices, meaning accurately predicting downstream capabilities requires predicting not just how probability mass concentrates on the correct choice with scale, but also how probability mass fluctuates on specific incorrect choices with scale. We empirically study how probability mass on the correct choice co-varies with probability mass on incorrect choices with increasing compute, suggesting that scaling laws for incorrect choices might be achievable. Our work also explains why pretraining scaling laws are commonly regarded as more predictable than downstream capabilities and contributes towards establishing scaling-predictable evaluations of frontier AI models.

Hampering the interpretation of benchmark scores, evaluation data contamination has become a growing concern in the evaluation of LLMs, and an active area of research studies its effects. While evaluation data contamination is easily understood intuitively, it is surprisingly difficult to define precisely which samples should be considered contaminated and, consequently, how it impacts benchmark scores. We propose that these questions should be addressed together and that contamination metrics can be assessed based on whether models benefit from the examples they mark contaminated. We propose a novel analysis method called ConTAM, and show with a large scale survey of existing and novel n-gram based contamination metrics across 13 benchmarks and 7 models from 2 different families that ConTAM can be used to better understand evaluation data contamination and its effects. We find that contamination may have a much larger effect than reported in recent LLM releases and benefits models differently at different scales. We also find that considering only the longest contaminated substring provides a better signal than considering a union of all contaminated substrings, and that doing model and benchmark specific threshold analysis greatly increases the specificity of the results. Lastly, we investigate the impact of hyperparameter choices, finding that, among other things, both using larger values of n and disregarding matches that are infrequent in the pre-training data lead to many false negatives. With ConTAM, we provide a method to empirically ground evaluation data contamination metrics in downstream effects. With our exploration, we shed light on how evaluation data contamination can impact LLMs and provide insight into the considerations important when doing contamination analysis. We end our paper by discussing these in more detail and providing concrete suggestions for future work.

As the use of text-to-image generative models increases, so does the adoption of automatic benchmarking methods used in their evaluation. However, while metrics and datasets abound, there are few unified benchmarking libraries that provide a framework for performing evaluations across many datasets and metrics. Furthermore, the rapid introduction of increasingly robust benchmarking methods requires that evaluation libraries remain flexible to new datasets and metrics. Finally, there remains a gap in synthesizing evaluations in order to deliver actionable takeaways about model performance. To enable unified, flexible, and actionable evaluations, we introduce EvalGIM (pronounced ''EvalGym''), a library for evaluating generative image models. EvalGIM contains broad support for datasets and metrics used to measure quality, diversity, and consistency of text-to-image generative models. In addition, EvalGIM is designed with flexibility for user customization as a top priority and contains a structure that allows plug-and-play additions of new datasets and metrics. To enable actionable evaluation insights, we introduce ''Evaluation Exercises'' that highlight takeaways for specific evaluation questions. The Evaluation Exercises contain easy-to-use and reproducible implementations of two state-of-the-art evaluation methods of text-to-image generative models: consistency-diversity-realism Pareto Fronts and disaggregated measurements of performance disparities across groups. EvalGIM also contains Evaluation Exercises that introduce two new analysis methods for text-to-image generative models: robustness analyses of model rankings and balanced evaluations across different prompt styles. We encourage text-to-image model exploration with EvalGIM and invite contributions at https://github.com/facebookresearch/EvalGIM/.

Recent developments in large-scale machine learning suggest that by scaling up data, model size and training time properly, one might observe that improvements in pre-training would transfer favorably to most downstream tasks. In this work, we systematically study this phenomena and establish that, as we increase the upstream accuracy, the performance of downstream tasks saturates. In particular, we investigate more than 4800 experiments on Vision Transformers, MLP-Mixers and ResNets with number of parameters ranging from ten million to ten billion, trained on the largest scale of available image data (JFT, ImageNet21K) and evaluated on more than 20 downstream image recognition tasks. We propose a model for downstream performance that reflects the saturation phenomena and captures the nonlinear relationship in performance of upstream and downstream tasks. Delving deeper to understand the reasons that give rise to these phenomena, we show that the saturation behavior we observe is closely related to the way that representations evolve through the layers of the models. We showcase an even more extreme scenario where performance on upstream and downstream are at odds with each other. That is, to have a better downstream performance, we need to hurt upstream accuracy.

Recent advances in large language models (LLMs) have led to the development of various evaluation benchmarks. These benchmarks typically rely on a single instruction template for evaluating all LLMs on a specific task. In this paper, we comprehensively analyze the brittleness of results obtained via single-prompt evaluations across 6.5M instances, involving 20 different LLMs and 39 tasks from 3 benchmarks. To improve robustness of the analysis, we propose to evaluate LLMs with a set of diverse prompts instead. We discuss tailored evaluation metrics for specific use cases (e.g., LLM developers vs. developers interested in a specific downstream task), ensuring a more reliable and meaningful assessment of LLM capabilities. We then implement these criteria and conduct evaluations of multiple models, providing insights into the true strengths and limitations of current LLMs.

Learning visual representations from natural language supervision has recently shown great promise in a number of pioneering works. In general, these language-augmented visual models demonstrate strong transferability to a variety of datasets and tasks. However, it remains challenging to evaluate the transferablity of these models due to the lack of easy-to-use evaluation toolkits and public benchmarks. To tackle this, we build ELEVATER (Evaluation of Language-augmented Visual Task-level Transfer), the first benchmark and toolkit for evaluating(pre-trained) language-augmented visual models. ELEVATER is composed of three components. (i) Datasets. As downstream evaluation suites, it consists of 20 image classification datasets and 35 object detection datasets, each of which is augmented with external knowledge. (ii) Toolkit. An automatic hyper-parameter tuning toolkit is developed to facilitate model evaluation on downstream tasks. (iii) Metrics. A variety of evaluation metrics are used to measure sample-efficiency (zero-shot and few-shot) and parameter-efficiency (linear probing and full model fine-tuning). ELEVATER is a platform for Computer Vision in the Wild (CVinW), and is publicly released at at https://computer-vision-in-the-wild.github.io/ELEVATER/

The rapid advancements in computing dramatically increase the scale and cost of training Large Language Models (LLMs). Accurately predicting downstream task performance prior to model training is crucial for efficient resource allocation, yet remains challenging due to two primary constraints: (1) the "emergence phenomenon", wherein downstream performance metrics become meaningful only after extensive training, which limits the ability to use smaller models for prediction; (2) Uneven task difficulty distributions and the absence of consistent scaling laws, resulting in substantial metric variability. Existing performance prediction methods suffer from limited accuracy and reliability, thereby impeding the assessment of potential LLM capabilities. To address these challenges, we propose a Clustering-On-Difficulty (COD) downstream performance prediction framework. COD first constructs a predictable support subset by clustering tasks based on difficulty features, strategically excluding non-emergent and non-scalable clusters. The scores on the selected subset serve as effective intermediate predictors of downstream performance on the full evaluation set. With theoretical support, we derive a mapping function that transforms performance metrics from the predictable subset to the full evaluation set, thereby ensuring accurate extrapolation of LLM downstream performance. The proposed method has been applied to predict performance scaling for a 70B LLM, providing actionable insights for training resource allocation and assisting in monitoring the training process. Notably, COD achieves remarkable predictive accuracy on the 70B LLM by leveraging an ensemble of small models, demonstrating an absolute mean deviation of 1.36% across eight important LLM evaluation benchmarks.

Evaluation efforts such as TREC, CLEF, NTCIR and FIRE, alongside public leaderboard such as MS MARCO, are intended to encourage research and track our progress, addressing big questions in our field. However, the goal is not simply to identify which run is "best", achieving the top score. The goal is to move the field forward by developing new robust techniques, that work in many different settings, and are adopted in research and practice. This paper uses the MS MARCO and TREC Deep Learning Track as our case study, comparing it to the case of TREC ad hoc ranking in the 1990s. We show how the design of the evaluation effort can encourage or discourage certain outcomes, and raising questions about internal and external validity of results. We provide some analysis of certain pitfalls, and a statement of best practices for avoiding such pitfalls. We summarize the progress of the effort so far, and describe our desired end state of "robust usefulness", along with steps that might be required to get us there.

Frontier language models sometimes recognize that they are being evaluated and adjust their behavior, undermining validity of benchmark results. Yet the field studies it without a shared foundation, conflating properties of the evaluation with properties of the model, and detection with behavioral response. We ground evaluation awareness in social psychology, decomposing it into an environment component (how recognizable the task is) and a model component that separates recognition from propensity to act on it. We operationalize the environment component through eight categorized trigger factors, such as placeholder entities and grading-style output formats, and study recognition and behavior through chain-of-thought monitoring. Across nine frontier models and four benchmarks, recognition rates depend on the specific pairing of model and benchmark rather than on either in isolation. Recognition rarely leads to behavioral change, and when it does, the direction depends on the type of evaluation perceived. Models are also more sensitive to safety than capability evaluations, placing safety benchmark validity at greater risk. To study which factors each model is sensitive to and how they interact, we propose EvalAwareBench, a factor-controlled benchmark of 100 paired safety-capability tasks where each of the eight factors can be independently toggled, varying evaluative signals while holding the underlying request fixed. Through EvalAwareBench, we find that no single factor uniformly affects all models, but stacking factors progressively raises evaluation awareness across all of them. Our framework and EvalAwareBench provide the tools to measure, attribute, and mitigate evaluation awareness, pointing to behavioral consistency under recognition as a promising path forward.

Foundation models are increasingly central to high-stakes AI systems, and governance frameworks now depend on evaluations to assess their risks and capabilities. Although general capability evaluations are widespread, social impact assessments covering bias, fairness, privacy, environmental costs, and labor practices remain uneven across the AI ecosystem. To characterize this landscape, we conduct the first comprehensive analysis of both first-party and third-party social impact evaluation reporting across a wide range of model developers. Our study examines 186 first-party release reports and 183 post-release evaluation sources, and complements this quantitative analysis with interviews of model developers. We find a clear division of evaluation labor: first-party reporting is sparse, often superficial, and has declined over time in key areas such as environmental impact and bias, while third-party evaluators including academic researchers, nonprofits, and independent organizations provide broader and more rigorous coverage of bias, harmful content, and performance disparities. However, this complementarity has limits. Only model developers can authoritatively report on data provenance, content moderation labor, financial costs, and training infrastructure, yet interviews reveal that these disclosures are often deprioritized unless tied to product adoption or regulatory compliance. Our findings indicate that current evaluation practices leave major gaps in assessing AI's societal impacts, highlighting the urgent need for policies that promote developer transparency, strengthen independent evaluation ecosystems, and create shared infrastructure to aggregate and compare third-party evaluations in a consistent and accessible way.

Large Language Models (LLMs) are increasingly relied upon to evaluate text outputs of other LLMs, thereby influencing leaderboards and development decisions. However, concerns persist over the accuracy of these assessments and the potential for misleading conclusions. In this work, we investigate the effectiveness of LLMs as evaluators for text generation tasks. We propose FBI, a novel framework designed to examine the proficiency of Evaluator LLMs in assessing four critical abilities in other LLMs: factual accuracy, instruction following, coherence in long-form writing, and reasoning proficiency. By introducing targeted perturbations in answers generated by LLMs, that clearly impact one of these key capabilities, we test whether an Evaluator LLM can detect these quality drops. By creating a total of 2400 perturbed answers covering 22 perturbation categories, we conduct a comprehensive study using different evaluation strategies on five prominent LLMs commonly used as evaluators in the literature. Our findings reveal significant shortcomings in current Evaluator LLMs, which failed to identify quality drops in over 50\% of cases on average. Single-answer and pairwise evaluations demonstrated notable limitations, whereas reference-based evaluations showed comparatively better performance. These results underscore the unreliable nature of current Evaluator LLMs and advocate for cautious implementation in practical applications. Code and data are available at https://github.com/AI4Bharat/FBI.

The use of large language models (LLMs) for evaluating outputs is becoming an increasingly effective and scalable approach. However, it remains uncertain whether this capability extends beyond task-specific evaluations to more general assessments of text quality, particularly in multilingual contexts. In this study, we introduce, MTQ-Eval, a novel framework for multilingual text quality evaluation that learns from examples of both high- and low-quality texts, adjusting its internal representations. To develop MTQ-Eval, we first automatically generate text quality preference data and then use it to train open-source base LLMs to align with ratings of high- and low-quality text. Our comprehensive evaluation across 115 languages demonstrates the improved performance of the proposed model. Upon further analysis, we find that this enhanced evaluation capability also leads to notable improvements in downstream tasks.

The automatic evaluation of instruction following typically involves using large language models (LLMs) to assess response quality. However, there is a lack of comprehensive evaluation of these LLM-based evaluators across two dimensions: the base LLMs and the evaluation protocols. Therefore, we present a thorough meta-evaluation of instruction following, including 25 base LLMs and 15 recently proposed evaluation protocols, on 4 human-annotated datasets, assessing the evaluation accuracy of the LLM-evaluators. Our evaluation allows us to identify the best-performing base LLMs and evaluation protocols with a high degree of robustness. Moreover, our large-scale evaluation reveals: (1) Base LLM performance ranking remains largely consistent across evaluation protocols, with less capable LLMs showing greater improvement from protocol enhancements; (2) Robust evaluation of evaluation protocols requires many base LLMs with varying capability levels, as protocol effectiveness can depend on the base LLM used; (3) Evaluation results on different datasets are not always consistent, so a rigorous evaluation requires multiple datasets with distinctive features. We release our meta-evaluation suite ReIFE, which provides the codebase and evaluation result collection for more than 500 LLM-evaluator configurations, to support future research in instruction-following evaluation.

Recent progress in deep research systems has been impressive, but evaluation still lags behind real user needs. Existing benchmarks predominantly assess final reports using fixed rubrics, failing to evaluate the underlying research process. Most also offer limited multimodal coverage, rely on synthetic tasks that do not reflect real-world query complexity, and cannot be refreshed as knowledge evolves. To address these gaps, we introduce MiroEval, a benchmark and evaluation framework for deep research systems. The benchmark comprises 100 tasks (70 text-only, 30 multimodal), all grounded in real user needs and constructed via a dual-path pipeline that supports periodic updates, enabling a live and evolving setting. The proposed evaluation suite assesses deep research systems along three complementary dimensions: adaptive synthesis quality evaluation with task-specific rubrics, agentic factuality verification via active retrieval and reasoning over both web sources and multimodal attachments, and process-centric evaluation audits how the system searches, reasons, and refines throughout its investigation. Evaluation across 13 systems yields three principal findings: the three evaluation dimensions capture complementary aspects of system capability, with each revealing distinct strengths and weaknesses across systems; process quality serves as a reliable predictor of overall outcome while revealing weaknesses invisible to output-level metrics; and multimodal tasks pose substantially greater challenges, with most systems declining by 3 to 10 points. The MiroThinker series achieves the most balanced performance, with MiroThinker-H1 ranking the highest overall in both settings. Human verification and robustness results confirm the reliability of the benchmark and evaluation framework. MiroEval provides a holistic diagnostic tool for the next generation of deep research agents.

Evaluation of language model outputs on structured writing tasks is typically conducted with a number of desirable criteria presented to human evaluators or large language models (LLMs). For instance, on a prompt like "Help me draft an academic talk on coffee intake vs research productivity", a model response may be evaluated for criteria like accuracy and coherence. However, high-quality responses should do more than just satisfy basic task requirements. An effective response to this query should include quintessential features of an academic talk, such as a compelling opening, clear research questions, and a takeaway. To help identify these implicit criteria, we introduce EvalAgent, a novel framework designed to automatically uncover nuanced and task-specific criteria. EvalAgent first mines expert-authored online guidance. It then uses this evidence to propose diverse, long-tail evaluation criteria that are grounded in reliable external sources. Our experiments demonstrate that the grounded criteria produced by EvalAgent are often implicit (not directly stated in the user's prompt), yet specific (high degree of lexical precision). Further, EvalAgent criteria are often not satisfied by initial responses but they are actionable, such that responses can be refined to satisfy them. Finally, we show that combining LLM-generated and EvalAgent criteria uncovers more human-valued criteria than using LLMs alone.

Language model (LM) benchmarking faces several challenges: comprehensive evaluations are costly, benchmarks often fail to measure the intended capabilities, and evaluation quality can degrade due to labeling errors and benchmark saturation. Although various strategies have been proposed to mitigate these issues, they tend to address individual aspects in isolation, neglecting broader questions about overall evaluation quality. Here, we introduce Fluid Benchmarking, a new evaluation approach that advances LM benchmarking across multiple dimensions. Inspired by psychometrics, Fluid Benchmarking is based on the insight that the relative value of benchmark items depends on an LM's capability level, suggesting that evaluation should adapt to each LM. Methodologically, Fluid Benchmarking estimates an item response model based on existing LM evaluation results and uses the inferred quantities to select evaluation items dynamically, similar to computerized adaptive testing in education. In our experiments, we compare Fluid Benchmarking against the common practice of random item sampling as well as more sophisticated baselines, including alternative methods grounded in item response theory. We examine four dimensions -- efficiency, validity, variance, and saturation -- and find that Fluid Benchmarking achieves superior performance in all of them (e.g., higher validity and less variance on MMLU with fifty times fewer items). Our analysis shows that the two components of Fluid Benchmarking have distinct effects: item response theory, used to map performance into a latent ability space, increases validity, while dynamic item selection reduces variance. Overall, our results suggest that LM benchmarking can be substantially improved by moving beyond static evaluation.

Transfer learning is a classic paradigm by which models pretrained on large "upstream" datasets are adapted to yield good results on "downstream" specialized datasets. Generally, more accurate models on the "upstream" dataset tend to provide better transfer accuracy "downstream". In this work, we perform an in-depth investigation of this phenomenon in the context of convolutional neural networks (CNNs) trained on the ImageNet dataset, which have been pruned - that is, compressed by sparsifying their connections. We consider transfer using unstructured pruned models obtained by applying several state-of-the-art pruning methods, including magnitude-based, second-order, re-growth, lottery-ticket, and regularization approaches, in the context of twelve standard transfer tasks. In a nutshell, our study shows that sparse models can match or even outperform the transfer performance of dense models, even at high sparsities, and, while doing so, can lead to significant inference and even training speedups. At the same time, we observe and analyze significant differences in the behaviour of different pruning methods.

Being able to provide explanations for a model's decision has become a central requirement for the development, deployment, and adoption of machine learning models. However, we are yet to understand what explanation methods can and cannot do. How do upstream factors such as data, model prediction, hyperparameters, and random initialization influence downstream explanations? While previous work raised concerns that explanations (E) may have little relationship with the prediction (Y), there is a lack of conclusive study to quantify this relationship. Our work borrows tools from causal inference to systematically assay this relationship. More specifically, we study the relationship between E and Y by measuring the treatment effect when intervening on their causal ancestors, i.e., on hyperparameters and inputs used to generate saliency-based Es or Ys. Our results suggest that the relationships between E and Y is far from ideal. In fact, the gap between 'ideal' case only increase in higher-performing models -- models that are likely to be deployed. Our work is a promising first step towards providing a quantitative measure of the relationship between E and Y, which could also inform the future development of methods for E with a quantitative metric.

Despite widespread deployment of Large Language Models, systematic evaluation of instruction-following capabilities remains challenging. While comprehensive benchmarks exist, focused assessments that quickly diagnose specific instruction adherence patterns are valuable. As newer models may be trained on existing benchmarks, novel evaluation approaches are needed to assess genuine capabilities rather than memorized performance. This paper presents a streamlined evaluation framework using twenty carefully designed prompts to assess LLM instruction-following across diverse task categories. We demonstrate this framework through a large-scale empirical study conducted on October 14, 2025, testing 256 verified working models from 331 available via OpenRouter. To ensure methodological rigor and prevent selection bias, we first verified each model's basic functionality before inclusion. Unlike large-scale benchmarks requiring extensive computational resources, our approach offers a practical diagnostic tool researchers and practitioners can readily apply. Our methodology builds upon verifiable instructions while introducing a compact test suite balancing comprehensiveness with efficiency. Each prompt targets distinct aspects of instruction following, including format compliance, content constraints, logical sequencing, and multi-step task execution. We evaluate models from major providers (OpenAI, Anthropic, Google, Meta, Mistral) and emerging implementations (Qwen, DeepSeek, community models), providing comparative performance analysis. Our findings reveal consistent failure modes and identify specific instruction types posing particular challenges. This work contributes both a practical evaluation tool and one of the most comprehensive empirical analyses of instruction-following capabilities across the contemporary LLM landscape.

If AI models can detect when they are being evaluated, the effectiveness of evaluations might be compromised. For example, models could have systematically different behavior during evaluations, leading to less reliable benchmarks for deployment and governance decisions. We investigate whether frontier language models can accurately classify transcripts based on whether they originate from evaluations or real-world deployment, a capability we call evaluation awareness. To achieve this, we construct a diverse benchmark of 1,000 prompts and transcripts from 61 distinct datasets. These span public benchmarks (e.g., MMLU, SWEBench), real-world deployment interactions, and agent trajectories from scaffolding frameworks (e.g., web-browsing agents). Frontier models clearly demonstrate above-random evaluation awareness (Gemini-2.5-Pro reaches an AUC of 0.83), but do not yet surpass our simple human baseline (AUC of 0.92). Furthermore, both AI models and humans are better at identifying evaluations in agentic settings compared to chat settings. Additionally, we test whether models can identify the purpose of the evaluation. Under multiple-choice and open-ended questioning, AI models far outperform random chance in identifying what an evaluation is testing for. Our results indicate that frontier models already exhibit a substantial, though not yet superhuman, level of evaluation-awareness. We recommend tracking this capability in future models.

Reward Models (RMs) are crucial for aligning language models with human preferences. Currently, the evaluation of RMs depends on measuring accuracy against a validation set of manually annotated preference data. Although this method is straightforward and widely adopted, the relationship between RM accuracy and downstream policy performance remains under-explored. In this work, we conduct experiments in a synthetic setting to investigate how differences in RM measured by accuracy translate into gaps in optimized policy performance. Our findings reveal that while there is a weak positive correlation between accuracy and downstream performance, policies optimized towards RMs with similar accuracy can exhibit quite different performance. Moreover, we discover that the way of measuring accuracy significantly impacts its ability to predict the final policy performance. Through the lens of the Regressional Goodhart effect, we recognize that accuracy, when used for measuring RM quality, can fail to fully capture the potential RM overoptimization. This underscores the inadequacy of relying solely on accuracy to reflect their impact on policy optimization.

In this work, we conduct a detailed analysis on the performance of legal-oriented pre-trained language models (PLMs). We examine the interplay between their original objective, acquired knowledge, and legal language understanding capacities which we define as the upstream, probing, and downstream performance, respectively. We consider not only the models' size but also the pre-training corpora used as important dimensions in our study. To this end, we release a multinational English legal corpus (LeXFiles) and a legal knowledge probing benchmark (LegalLAMA) to facilitate training and detailed analysis of legal-oriented PLMs. We release two new legal PLMs trained on LeXFiles and evaluate them alongside others on LegalLAMA and LexGLUE. We find that probing performance strongly correlates with upstream performance in related legal topics. On the other hand, downstream performance is mainly driven by the model's size and prior legal knowledge which can be estimated by upstream and probing performance. Based on these findings, we can conclude that both dimensions are important for those seeking the development of domain-specific PLMs.

We study a fundamental problem in the evaluation of large language models that we call training on the test task. Unlike wrongful practices like training on the test data, leakage, or data contamination, training on the test task is not a malpractice. Rather, the term describes a growing set of practices that utilize knowledge about evaluation tasks at training time. We demonstrate that training on the test task confounds both relative model evaluations and claims about emergent capabilities. We argue that the seeming superiority of one model family over another may be explained by a different degree of training on the test task. To this end, we propose an effective method to adjust for the effect of training on the test task on benchmark evaluations. Put simply, to fine-tune each model under comparison on the same task-relevant data prior to evaluation. We then show that instances of emergent behavior disappear gradually as models train on the test task. Our work promotes a new perspective on the evaluation of large language models, with broad implications for benchmarking and the study of emergent capabilities.

In recent years, there has been a wide interest in designing deep neural network-based models that automate downstream software engineering tasks on source code, such as code document generation, code search, and program repair. Although the main objective of these studies is to improve the effectiveness of the downstream task, many studies only attempt to employ the next best neural network model, without a proper in-depth analysis of why a particular solution works or does not, on particular tasks or scenarios. In this paper, using an example eXplainable AI (XAI) method (attention mechanism), we study two recent large language models (LLMs) for code (CodeBERT and GraphCodeBERT) on a set of software engineering downstream tasks: code document generation (CDG), code refinement (CR), and code translation (CT). Through quantitative and qualitative studies, we identify what CodeBERT and GraphCodeBERT learn (put the highest attention on, in terms of source code token types), on these tasks. We also show some of the common patterns when the model does not work as expected (performs poorly even on easy problems) and suggest recommendations that may alleviate the observed challenges.

Evaluations are critical for understanding the capabilities of large language models (LLMs). Fundamentally, evaluations are experiments; but the literature on evaluations has largely ignored the literature from other sciences on experiment analysis and planning. This article shows researchers with some training in statistics how to think about and analyze data from language model evaluations. Conceptualizing evaluation questions as having been drawn from an unseen super-population, we present formulas for analyzing evaluation data, measuring differences between two models, and planning an evaluation experiment. We make a number of specific recommendations for running language model evaluations and reporting experiment results in a way that minimizes statistical noise and maximizes informativeness.

Generation capabilities and language coverage of multilingual large language models (mLLMs) are advancing rapidly. However, evaluation practices for generative abilities of mLLMs are still lacking comprehensiveness, scientific rigor, and consistent adoption across research labs, which undermines their potential to meaningfully guide mLLM development. We draw parallels with machine translation (MT) evaluation, a field that faced similar challenges and has, over decades, developed transparent reporting standards and reliable evaluations for multilingual generative models. Through targeted experiments across key stages of the generative evaluation pipeline, we demonstrate how best practices from MT evaluation can deepen the understanding of quality differences between models. Additionally, we identify essential components for robust meta-evaluation of mLLMs, ensuring the evaluation methods themselves are rigorously assessed. We distill these insights into a checklist of actionable recommendations for mLLM research and development.

Supply chain operations traditionally involve a variety of complex decision making problems. Over the last few decades, supply chains greatly benefited from advances in computation, which allowed the transition from manual processing to automation and cost-effective optimization. Nonetheless, business operators still need to spend substantial efforts in explaining and interpreting the optimization outcomes to stakeholders. Motivated by the recent advances in Large Language Models (LLMs), we study how this disruptive technology can help bridge the gap between supply chain automation and human comprehension and trust thereof. We design -- a framework that accepts as input queries in plain text, and outputs insights about the underlying optimization outcomes. Our framework does not forgo the state-of-the-art combinatorial optimization technology, but rather leverages it to quantitatively answer what-if scenarios (e.g., how would the cost change if we used supplier B instead of supplier A for a given demand?). Importantly, our design does not require sending proprietary data over to LLMs, which can be a privacy concern in some circumstances. We demonstrate the effectiveness of our framework on a real server placement scenario within Microsoft's cloud supply chain. Along the way, we develop a general evaluation benchmark, which can be used to evaluate the accuracy of the LLM output in other scenarios.

As research in large language models (LLMs) continues to accelerate, LLM-based evaluation has emerged as a scalable and cost-effective alternative to human evaluations for comparing the ever increasing list of models. This paper investigates the efficacy of these "LLM evaluators", particularly in using them to assess instruction following, a metric that gauges how closely generated text adheres to the given instruction. We introduce a challenging meta-evaluation benchmark, LLMBar, designed to test the ability of an LLM evaluator in discerning instruction-following outputs. The authors manually curated 419 pairs of outputs, one adhering to instructions while the other diverging, yet may possess deceptive qualities that mislead an LLM evaluator, e.g., a more engaging tone. Contrary to existing meta-evaluation, we discover that different evaluators (i.e., combinations of LLMs and prompts) exhibit distinct performance on LLMBar and even the highest-scoring ones have substantial room for improvement. We also present a novel suite of prompting strategies that further close the gap between LLM and human evaluators. With LLMBar, we hope to offer more insight into LLM evaluators and foster future research in developing better instruction-following models.

Many deployments must compare candidate language models for safety before a labeled benchmark exists for the relevant language, sector, or regulatory regime. We formalize this setting as benchmarkless comparative safety scoring and specify the contract under which a scenario-based audit can be interpreted as deployment evidence. Scores are valid only under a fixed scenario pack, rubric, auditor, judge, sampling configuration, and rerun budget. Because no labels are available, we replace ground-truth agreement with an instrumental-validity chain: responsiveness to a controlled safe-versus-abliterated contrast, dominance of target-driven variance over auditor and judge artifacts, and stability across reruns. We instantiate the chain in SimpleAudit, a local-first scoring instrument, and validate it on a Norwegian safety pack. Safe and abliterated targets separate with AUROC values between 0.89 and 1.00, target identity is the dominant variance component (η^2 approx 0.52), and severity profiles stabilize by ten reruns. Applying the same chain to Petri shows that it admits both tools. The substantial differences arise upstream of the chain, in claim-contract enforcement and deployment fit. A Norwegian public-sector procurement case comparing Borealis and Gemma 3 demonstrates the resulting evidence in practice: the safer model depends on scenario category and risk measure. Consequently, scores, matched deltas, critical rates, uncertainty, and the auditor and judge used must be reported together rather than collapsed into a single ranking.

Foundation models are routinely fine-tuned for use in particular domains, yet safety assessments are typically conducted only on base models, implicitly assuming that safety properties persist through downstream adaptation. We test this assumption by analyzing the safety behavior of 100 models, including widely deployed fine-tunes in the medical and legal domains as well as controlled adaptations of open foundation models alongside their bases. Across general-purpose and domain-specific safety benchmarks, we find that benign fine-tuning induces large, heterogeneous, and often contradictory changes in measured safety: models frequently improve on some instruments while degrading on others, with substantial disagreement across evaluations. These results show that safety behavior is not stable under ordinary downstream adaptation, raising critical questions about governance and deployment practices centered on base-model evaluations. Without explicit re-evaluation of fine-tuned models in deployment-relevant contexts, such approaches fall short of adequately managing downstream risk, overlooking practical sources of harm -- failures that are especially consequential in high-stakes settings and challenge current accountability paradigms.

Communication is defined as "Who says what to whom with what effect." A message from a communicator generates downstream receiver effects, also known as behavior. Receiver behavior, being a downstream effect of the message, carries rich signals about it. Even after carrying signals about the message, the behavior data is often ignored while training large language models. We show that training LLMs on receiver behavior can actually help improve their content-understanding abilities. Specifically, we show that training LLMs to predict the receiver behavior of likes and comments improves the LLM's performance on a wide variety of downstream content understanding tasks. We show this performance increase over 40 video and image understanding tasks over 23 benchmark datasets across both 0-shot and fine-tuning settings, outperforming many supervised baselines. Moreover, since receiver behavior, such as likes and comments, is collected by default on the internet and does not need any human annotations to be useful, the performance improvement we get after training on this data is essentially free-lunch. We release the receiver behavior cleaned comments and likes of 750k images and videos collected from multiple platforms along with our instruction-tuning data.

AI evaluation results are produced at scale but reported inconsistently across leaderboards, model cards, benchmark papers, and company blogs. The cost is interpretive: readers cannot reliably compare results across sources, identify what a report omits, or trace an aggregate claim to its underlying evidence. Recent efforts address isolated components but leave three gaps: they cover only narrow slices of the evaluation lifecycle and do not compose into a single interpretable record; they specify static representations that do not differentiate the questions different stakeholders bring to the same evidence; and they remain proposals on paper, lacking the extraction infrastructure required for adoption at scale. We present , an operational reporting layer that composes benchmark metadata, evaluation run data, and model metadata into a unified record. We (1) derive a reporting schema from a structured review of 52 papers and 10 stakeholder interviews, (2) implement four interpretive signals (reproducibility, documentation completeness, provenance and risk, and score comparability), rendered through reader modes calibrated to research and non-research audiences, and (3) deploy a monitoring tool that applies across 5,816 models, 635 benchmarks, and 101,843 results, surfacing systematic gaps in current reporting practice.

Language model pretraining involves training on extensive corpora, where data quality plays a pivotal role. In this work, we aim to directly estimate the contribution of data during pretraining and select pretraining data in an efficient manner. Specifically, we draw inspiration from recent findings showing that compression efficiency (i.e., the normalized loss) of diverse models on certain text correlates strongly with their downstream performance, when the text domain aligns with the downstream benchmark (Huang et al., 2024). Building on this observation, we hypothesize that data on which model losses are predictive of downstream abilities also contribute effectively to learning. To leverage this insight, we introduce data selection based on data's Predictive strength (Preselect), a lightweight and efficient data selection method that requires training and deploying only a fastText-based scorer. Through comprehensive experiments with 1B and 3B parameter models, we demonstrate that models trained on 30B tokens selected with PreSelect surpasses the performance of a vanilla baseline trained on 300B tokens, achieving a 10x reduction in compute requirements. Furthermore, PreSelect significantly outperforms other competitive data selection baselines, such as DCLM and FineWeb-Edu on a scale of 3B models trained on 100B tokens. We open-source our trained data selection scorer along with the curated datasets at https://github.com/hkust-nlp/PreSelect.

Large language models~(LLMs) have greatly advanced the frontiers of artificial intelligence, attaining remarkable improvement in model capacity. To assess the model performance, a typical approach is to construct evaluation benchmarks for measuring the ability level of LLMs in different aspects. Despite that a number of high-quality benchmarks have been released, the concerns about the appropriate use of these benchmarks and the fair comparison of different models are increasingly growing. Considering these concerns, in this paper, we discuss the potential risk and impact of inappropriately using evaluation benchmarks and misleadingly interpreting the evaluation results. Specially, we focus on a special issue that would lead to inappropriate evaluation, \ie benchmark leakage, referring that the data related to evaluation sets is occasionally used for model training. This phenomenon now becomes more common since pre-training data is often prepared ahead of model test. We conduct extensive experiments to study the effect of benchmark leverage, and find that it can dramatically boost the evaluation results, which would finally lead to an unreliable assessment of model performance. To improve the use of existing evaluation benchmarks, we finally present several guidelines for both LLM developers and benchmark maintainers. We hope this work can draw attention to appropriate training and evaluation of LLMs.

Accurate evaluation is central to the large language model (LLM) ecosystem, guiding model selection and downstream adoption across diverse use cases. In practice, however, evaluating generative outputs typically relies on rigid lexical methods to extract and assess answers, which can conflate a model's true problem-solving ability with its compliance with predefined formatting guidelines. While recent LLM-as-a-Judge approaches mitigate this issue by assessing semantic correctness rather than strict structural conformity, they also introduce substantial computational overhead, making evaluation costly. In this work, we first systematically investigate the limitations of lexical evaluation through a large-scale empirical study spanning 36 models and 15 downstream tasks, demonstrating that such methods correlate poorly with human judgments. To address this limitation, we introduce BERT-as-a-Judge, an encoder-driven approach for assessing answer correctness in reference-based generative settings, robust to variations in output phrasing, and requiring only lightweight training on synthetically annotated question-candidate-reference triplets. We show that it consistently outperforms the lexical baseline while matching the performance of much larger LLM judges, providing a compelling tradeoff between the two and enabling reliable, scalable evaluation. Finally, through extensive experimentation, we provide detailed insights into BERT-as-a-Judge's performance to offer practical guidance for practitioners, and release all project artifacts to foster downstream adoption.

Precise estimation of downstream performance in large language models (LLMs) prior to training is essential for guiding their development process. Scaling laws analysis utilizes the statistics of a series of significantly smaller sampling language models (LMs) to predict the performance of the target LLM. For downstream performance prediction, the critical challenge lies in the emergent abilities in LLMs that occur beyond task-specific computational thresholds. In this work, we focus on the pre-training loss as a more computation-efficient metric for performance estimation. Our two-stage approach FLP consists of first estimating a function that maps computational resources (e.g., FLOPs) to the pre-training Loss using a series of fully-converged sampling models, followed by mapping the pre-training loss to downstream task Performance using the intermediate models with emerged performance. In our experiments, this FLP solution accurately predicts the performance of LLMs with 7B and 13B parameters using a series of sampling LMs up to 3B, achieving error margins of 5% and 10%, respectively, and significantly outperforming the FLOPs-to-Performance approach. Further, we present FLP-M, a fundamental approach for performance prediction that addresses the practical need to integrate datasets from multiple sources during pre-training. FLP-M extends the power law analytical function to predict domain-specific pre-training loss based on FLOPs across data sources, and employs a two-layer neural network to model the non-linear relationship between multiple domain-specific loss and downstream performance. By utilizing a 3B LLM trained on a specific ratio and a series of smaller sampling LMs, FLP-M can effectively forecast the performance of 3B and 7B LLMs across various data mixtures for most benchmarks within 10% error margins.

We introduce a new benchmark for reward models that quantifies their ability to produce strong language models through RLHF (Reinforcement Learning from Human Feedback). The gold-standard approach is to run a full RLHF training pipeline and directly probe downstream LLM performance. However, this process is prohibitively expensive. To address this, we build a predictive model of downstream LLM performance by evaluating the reward model on proxy tasks. These proxy tasks consist of a large-scale human preference and a verifiable correctness preference dataset, in which we measure 12 metrics across 12 domains. To investigate which reward model metrics are most correlated to gold-standard RLHF outcomes, we launch an end-to-end RLHF experiment on a large-scale crowdsourced human preference platform to view real reward model downstream performance as ground truth. Ultimately, we compile our data and findings into Preference Proxy Evaluations (PPE), the first reward model benchmark explicitly linked to post-RLHF real-world human preference performance, which we open-source for public use and further development. Our code and evaluations can be found at https://github.com/lmarena/PPE .

AI evaluation is undergoing a structural change. Large language models (LLMs) are increasingly deployed as systems that act over time through tools, environments, users, and other agents, while many evaluation practices still inherit assumptions from response-centered benchmarks (e.g., fixed inputs, isolated outputs, and outcome judgments that can be made from a single response). The field has begun to build interactive benchmarks, but the resulting landscape is fragmented: benchmarks differ in what interaction artifacts they admit, how trajectories are scored, and what claims their results support. This position paper argues that interactive evaluation should be treated as a principled evaluation paradigm, not merely a new family of agent benchmarks. Simply adopting previous evaluation paradigms does not suffice. We define evaluation as an autonomous mapping from evidence to judgments, and show that interactive evaluation changes both sides of this mapping: the evidence becomes interaction-generated trajectories, while the evaluation procedure must assess process, recoverability, coordination, robustness, and system-level performance. Building on this definition, we propose a two-axis taxonomy, derive design principles and reporting standards, examine representative scenarios, and analyze how longstanding evaluation challenges reappear at the trajectory level.

We introduce Differential Performance Evaluation (DPE), a framework designed to reliably evaluate Large Language Models (LLMs) for efficient code generation. Traditional coding benchmarks often fail to provide reliable insights into code efficiency, due to their reliance on simplistic test inputs and the absence of effective compound metrics. DPE addresses these issues by focusing on efficiency-demanding programming tasks and establishing an insightful compound metric for performance evaluation. DPE operates in two phases: To curate efficiency datasets, it selects efficiency-demanding tasks from existing coding benchmarks and generates computationally expensive inputs to stress the efficiency of LLM solutions. To assess the code efficiency, DPE profiles the new solution and compares it globally against a set of reference solutions that exhibit distinct efficiency levels, where the matched level defines its efficiency score. As a proof of concept, we use DPE to create EvalPerf, a benchmark with 121 performance-challenging coding tasks. Our comprehensive evaluation draws interesting findings on the efficiency impact of model sizes, instruction tuning, and prompting. For example, while the scaling law fails to account for code efficiency, general instruction tuning benefits both code correctness and efficiency. We also evaluate the evaluation by examining the effectiveness of DPE, showing that EvalPerf is reliable and convenient to use even across platforms.

The ``LLM-as-a-judge'' paradigm has become a standard method for evaluating open-ended generation. To address the quadratic scalability costs of pairwise comparisons, popular benchmarks like Arena-Hard and AlpacaEval compare all models against a single anchor. However, despite its widespread use, the impact of anchor selection on the reliability of the results remains largely unexplored. In this work, we systematically investigate the effect of anchor selection by evaluating 22 different anchors on the Arena-Hard-v2.0 dataset. We find that the choice of anchor is critical: a poor anchor can dramatically reduce correlation with human rankings. We identify that common anchor choices (best-performing and worst-performing models) make poor anchors. Because these extreme anchors are consistently better or worse than all other models, they are seldom indicative of the relative ranking of the models. We further quantify the effect size of anchor selection, showing it is comparable to the selection of a judge model. We conclude with actionable recommendations. First, we conduct a power analysis, and compute sufficient benchmark sizes for anchor-based evaluation, finding that standard benchmark sizes are insufficient for pairwise evaluation and fail to distinguish between competitive models reliably. Second, we provide guidelines for selecting informative anchors to ensure reliable and efficient evaluation practices.

Scaling laws provide important insights that can guide the design of large language models (LLMs). Existing work has primarily focused on studying scaling laws for pretraining (upstream) loss. However, in transfer learning settings, in which LLMs are pretrained on an unsupervised dataset and then finetuned on a downstream task, we often also care about the downstream performance. In this work, we study the scaling behavior in a transfer learning setting, where LLMs are finetuned for machine translation tasks. Specifically, we investigate how the choice of the pretraining data and its size affect downstream performance (translation quality) as judged by two metrics: downstream cross-entropy and BLEU score. Our experiments indicate that the size of the finetuning dataset and the distribution alignment between the pretraining and downstream data significantly influence the scaling behavior. With sufficient alignment, both downstream cross-entropy and BLEU score improve monotonically with more pretraining data. In such cases, we show that it is possible to predict the downstream BLEU score with good accuracy using a log-law. However, there are also cases where moderate misalignment causes the BLEU score to fluctuate or get worse with more pretraining, whereas downstream cross-entropy monotonically improves. By analyzing these observations, we provide new practical insights for choosing appropriate pretraining data.

We introduce AbGen, the first benchmark designed to evaluate the capabilities of LLMs in designing ablation studies for scientific research. AbGen consists of 1,500 expert-annotated examples derived from 807 NLP papers. In this benchmark, LLMs are tasked with generating detailed ablation study designs for a specified module or process based on the given research context. Our evaluation of leading LLMs, such as DeepSeek-R1-0528 and o4-mini, highlights a significant performance gap between these models and human experts in terms of the importance, faithfulness, and soundness of the ablation study designs. Moreover, we demonstrate that current automated evaluation methods are not reliable for our task, as they show a significant discrepancy when compared to human assessment. To better investigate this, we develop AbGen-Eval, a meta-evaluation benchmark designed to assess the reliability of commonly used automated evaluation systems in measuring LLM performance on our task. We investigate various LLM-as-Judge systems on AbGen-Eval, providing insights for future research on developing more effective and reliable LLM-based evaluation systems for complex scientific tasks.

We introduce EvaLearn, a pioneering benchmark designed to evaluate large language models (LLMs) on their learning capability and efficiency in challenging tasks, a critical, yet underexplored aspect of model potential. EvaLearn contains 648 challenging problems across six task types, grouped into 182 sequences, each sequence dedicated to one task type. Diverging from most existing benchmarks that evaluate models in parallel, EvaLearn requires models to solve problems sequentially, allowing them to leverage the experience gained from previous solutions. EvaLearn provides five comprehensive automated metrics to evaluate models and quantify their learning capability and efficiency. We extensively benchmark nine frontier models and observe varied performance profiles: some models, such as Claude-3.7-sonnet, start with moderate initial performance but exhibit strong learning ability, while some models struggle to benefit from experience and may even show negative transfer. Moreover, we investigate model performance under two learning settings and find that instance-level rubrics and teacher-model feedback further facilitate model learning. Importantly, we observe that current LLMs with stronger static abilities do not show a clear advantage in learning capability across all tasks, highlighting that EvaLearn evaluates a new dimension of model performance. We hope EvaLearn provides a novel evaluation perspective for assessing LLM potential and understanding the gap between models and human capabilities, promoting the development of deeper and more dynamic evaluation approaches. All datasets, the automatic evaluation framework, and the results studied in this paper are available at the GitHub repository.

Generating a review for an academic research paper is a complex task that requires a deep understanding of the document's content and the interdependencies between its sections. It demands not only insight into technical details but also an appreciation of the paper's overall coherence and structure. Recent methods have predominantly focused on fine-tuning large language models (LLMs) to address this challenge. However, they often overlook the computational and performance limitations imposed by long input token lengths. To address this, we introduce AutoRev, an Automatic Peer Review System for Academic Research Papers. Our novel framework represents an academic document as a graph, enabling the extraction of the most critical passages that contribute significantly to the review. This graph-based approach demonstrates effectiveness for review generation and is potentially adaptable to various downstream tasks, such as question answering, summarization, and document representation. When applied to review generation, our method outperforms SOTA baselines by an average of 58.72% across all evaluation metrics. We hope that our work will stimulate further research in applying graph-based extraction techniques to other downstream tasks in NLP. We plan to make our code public upon acceptance.

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.

Evaluation has traditionally focused on ranking candidates for a specific skill. Modern generalist models, such as Large Language Models (LLMs), decidedly outpace this paradigm. Open-ended evaluation systems, where candidate models are compared on user-submitted prompts, have emerged as a popular solution. Despite their many advantages, we show that the current Elo-based rating systems can be susceptible to and even reinforce biases in data, intentional or accidental, due to their sensitivity to redundancies. To address this issue, we propose evaluation as a 3-player game, and introduce novel game-theoretic solution concepts to ensure robustness to redundancy. We show that our method leads to intuitive ratings and provide insights into the competitive landscape of LLM development.

Using large language models (LLMs) to evaluate text quality has recently gained popularity. Some prior works explore the idea of using LLMs for evaluation, while they differ in some details of the evaluation process. In this paper, we analyze LLM evaluation (Chiang and Lee, 2023) and G-Eval (Liu et al., 2023), and we discuss how those details in the evaluation process change how well the ratings given by LLMs correlate with human ratings. We find that the auto Chain-of-Thought (CoT) used in G-Eval does not always make G-Eval more aligned with human ratings. We also show that forcing the LLM to output only a numeric rating, as in G-Eval, is suboptimal. Last, we reveal that asking the LLM to explain its own ratings consistently improves the correlation between the ChatGPT and human ratings and pushes state-of-the-art (SoTA) correlations on two meta-evaluation datasets.

Many approaches have been proposed to use diffusion models to augment training datasets for downstream tasks, such as classification. However, diffusion models are themselves trained on large datasets, often with noisy annotations, and it remains an open question to which extent these models contribute to downstream classification performance. In particular, it remains unclear if they generalize enough to improve over directly using the additional data of their pre-training process for augmentation. We systematically evaluate a range of existing methods to generate images from diffusion models and study new extensions to assess their benefit for data augmentation. Personalizing diffusion models towards the target data outperforms simpler prompting strategies. However, using the pre-training data of the diffusion model alone, via a simple nearest-neighbor retrieval procedure, leads to even stronger downstream performance. Our study explores the potential of diffusion models in generating new training data, and surprisingly finds that these sophisticated models are not yet able to beat a simple and strong image retrieval baseline on simple downstream vision tasks.

The effectiveness of recommendation systems is pivotal to user engagement and satisfaction in online platforms. As these recommendation systems increasingly influence user choices, their evaluation transcends mere technical performance and becomes central to business success. This paper addresses the multifaceted nature of recommendations system evaluation by introducing a comprehensive suite of metrics, each tailored to capture a distinct aspect of system performance. We discuss * Similarity Metrics: to quantify the precision of content-based filtering mechanisms and assess the accuracy of collaborative filtering techniques. * Candidate Generation Metrics: to evaluate how effectively the system identifies a broad yet relevant range of items. * Predictive Metrics: to assess the accuracy of forecasted user preferences. * Ranking Metrics: to evaluate the effectiveness of the order in which recommendations are presented. * Business Metrics: to align the performance of the recommendation system with economic objectives. Our approach emphasizes the contextual application of these metrics and their interdependencies. In this paper, we identify the strengths and limitations of current evaluation practices and highlight the nuanced trade-offs that emerge when optimizing recommendation systems across different metrics. The paper concludes by proposing a framework for selecting and interpreting these metrics to not only improve system performance but also to advance business goals. This work is to aid researchers and practitioners in critically assessing recommendation systems and fosters the development of more nuanced, effective, and economically viable personalization strategies. Our code is available at GitHub - https://github.com/aryan-jadon/Evaluation-Metrics-for-Recommendation-Systems.

The rapid proliferation of LLMs has created a critical evaluation paradox: while LLMs claim multilingual proficiency, comprehensive non-machine-translated benchmarks exist for fewer than 30 languages, leaving >98% of the world's 7,000 languages in an empirical void. Traditional benchmark construction faces scaling challenges such as cost, scarcity of domain experts, and data contamination. We evaluate the validity of a simpler alternative: can translation quality alone indicate a model's broader multilingual capabilities? Through systematic evaluation of 14 models (1B-72B parameters) across 9 diverse benchmarks and 7 translation metrics, we find that translation performance is a good indicator of downstream task success (e.g., Phi-4, median Pearson r: MetricX = 0.89, xCOMET = 0.91, SSA-COMET = 0.87). These results suggest that the representational abilities supporting faithful translation overlap with those required for multilingual understanding. Translation quality, thus emerges as a strong, inexpensive first-pass proxy of multilingual performance, enabling a translation-first screening with targeted follow-up for specific tasks.

The GPT-4 technical report suggests that downstream performance can be predicted from pre-training signals, but offers little methodological detail on how to quantify this. This work address this gap by modeling knowledge retention, the capacity of a pre-trained language model to memorize factual information from its corpus, and introduce a principled method to estimate it prior to training. We propose Size-dependent Mutual Information (SMI), an information-theoretic predictor that integrates knowledge frequency, knowledge specificity, and model size to forecast closed-book question answering (QA) accuracy. SMI is validated through large-scale document retrieval over the disclosed pre-training corpora of 21 public and 3 custom models, combined with a robust multi-template QA evaluation. Experiments show that SMI significantly outperforms repetition-based baselines and achieves R^2 > 0.7 in predicting QA accuracy for models above 1B parameters, without additional training. The analysis further reveals diminishing returns from scaling data and model size and provides evidence for an intrinsic upper bound on knowledge retention achievable by pre-training alone, motivating retrieval and other augmentation strategies.

Hard-negative source selection for dense retrieval is usually decided only after fine-tuning and downstream evaluation. We propose ECI_{sem}, a semantic residual variant of Effective Contrastive Information (ECI) that ranks candidate negative sources using frozen target-encoder embeddings. ECI_{sem} is training-free, not label-free: each scored example requires a query, a labeled positive, and an explicit candidate negative. ECI_{sem} builds a weighted residual information matrix from target consistency, semantic locality, lexical residuality, and a log-determinant diversity objective. On MS MARCO negative sources, in-family ECI_{sem} ranks LLM negatives highest among non-hybrid sources and Dense+LLM highest among hybrid sources, matching the strongest aggregate BEIR transfer results across DistilBERT, E5-base, and Contriever. Controlled ablations show that this alignment depends on using the target encoder family, while additional ablations show stability under sample-size, temperature, tokenizer, and IDF-corpus perturbations. The theory gives a local linearized link to loss reduction, while the empirical study treats downstream evaluation as the final test.

Rubric-based evaluation has become a prevailing paradigm for evaluating instruction following in large language models (LLMs). Despite its widespread use, the reliability of these rubric-level evaluations remains unclear, calling for meta-evaluation. However, prior meta-evaluation efforts largely focus on the response level, failing to assess the fine-grained judgment accuracy that rubric-based evaluation relies on. To bridge this gap, we introduce RubricEval. Our benchmark features: (1) the first rubric-level meta-evaluation benchmark for instruction following, (2) diverse instructions and responses spanning multiple categories and model sources, and (3) a substantial set of 3,486 quality-controlled instances, along with Easy/Hard subsets that better differentiates judge performance. Our experiments reveal that rubric-level judging remains far from solved: even GPT-4o, a widely adopted judge in instruction-following benchmarks, achieves only 55.97% on Hard subset. Considering evaluation paradigm, rubric-level evaluation outperforms checklist-level, explicit reasoning improves accuracy, and both together reduce inter-judge variance. Through our established rubric taxonomy, we further identify common failure modes and offer actionable insights for reliable instruction-following evaluation.

As large language models (LLMs) become increasingly versatile, numerous large scale benchmarks have been developed to thoroughly assess their capabilities. These benchmarks typically consist of diverse datasets and prompts to evaluate different aspects of LLM performance. However, comprehensive evaluations on hundreds or thousands of prompts incur tremendous costs in terms of computation, money, and time. In this work, we investigate strategies to improve evaluation efficiency by selecting a subset of examples from each benchmark using a learned policy. Our approach models the dependencies across test examples, allowing accurate prediction of the evaluation outcomes for the remaining examples based on the outcomes of the selected ones. Consequently, we only need to acquire the actual evaluation outcomes for the selected subset. We rigorously explore various subset selection policies and introduce a novel RL-based policy that leverages the captured dependencies. Empirical results demonstrate that our approach significantly reduces the number of evaluation prompts required while maintaining accurate performance estimates compared to previous methods.

Evaluating the capability of Large Language Models (LLMs) in following instructions has heavily relied on a powerful LLM as the judge, introducing unresolved biases that deviate the judgments from human judges. In this work, we reevaluate various choices for automatic evaluation on a wide range of instruction-following tasks. We experiment with methods that leverage human-written responses and observe that they enhance the reliability of automatic evaluations across a wide range of tasks, resulting in up to a 3.2% improvement in agreement with human judges. We also discovered that human-written responses offer an orthogonal perspective to model-generated responses in following instructions and should be used as an additional context when comparing model responses. Based on these observations, we develop a new evaluation benchmark, Human Response-Guided Evaluation of Instruction Following (HREF), comprising 4,258 samples across 11 task categories with a composite evaluation setup, employing a composite evaluation setup that selects the most reliable method for each category. In addition to providing reliable evaluation, HREF emphasizes individual task performance and is free from contamination. Finally, we study the impact of key design choices in HREF, including the size of the evaluation set, the judge model, the baseline model, and the prompt template. We host a live leaderboard that evaluates LLMs on the private evaluation set of HREF.

Wu et al. (2026) showed that most frontier large language models (LLMs) recommend a sponsored, roughly twice-as-expensive flight when their system prompt contains a soft sponsorship cue. We reproduce their evaluation on ten open-weight chat models plus the two of their twenty-three models that are still reachable today (gpt-3.5-turbo, gpt-4o). All reported rates in this paper are produced under the same judge the original paper used (gpt-4o); we additionally store every label under an open-weight (gpt-oss-120b) and a smaller proprietary (gpt-4o-mini) judge for an ablation. Three findings emerge. First, a prose description of an LLM evaluation pipeline is not, on its own, sufficient for accurate reproduction: we surfaced three silent implementation failures that each shifted a reported rate by tens of percentage points. Second, the central claims do generalise - the gpt-3.5-turbo logistic-regression intercept of alpha = 0.81 is within four points of the original alpha = 0.86, and 200 of 200 trials on gpt-3.5-turbo and gpt-4o promote a payday lender to a financially distressed user. Third, a thirty-token user prompt that asks the assistant for a neutral comparison table first cuts sponsored recommendation from 46.9% to 1.0% averaged across our ten open-source models, and from 53.0% to 0% averaged across the two OpenAI models. AI literacy and price-comparison portals are likely market-level mitigations; the harmful-product cell is bounded by neither. Raw data, labels and analysis scripts are at https://github.com/akmaier/Paper-LLM-Ads .

We evaluate the propensity of frontier models to sabotage or refuse to assist with safety research when deployed as AI research agents within a frontier AI company. We apply two complementary evaluations to four Claude models (Mythos Preview, Opus 4.7 Preview, Opus 4.6, and Sonnet 4.6): an unprompted sabotage evaluation testing model behaviour with opportunities to sabotage safety research, and a sabotage continuation evaluation testing whether models continue to sabotage when placed in trajectories where prior actions have started undermining research. We find no instances of unprompted sabotage across any model, with refusal rates close to zero for Mythos Preview and Opus 4.7 Preview, though all models sometimes only partially completed tasks. In the continuation evaluation, Mythos Preview actively continues sabotage in 7% of cases (versus 3% for Opus 4.6, 4% for Sonnet 4.6, and 0% for Opus 4.7 Preview), and exhibits reasoning-output discrepancy in the majority of these cases, indicating covert sabotage reasoning. Our evaluation framework builds on Petri, an open-source LLM auditing tool, with a custom scaffold running models inside Claude Code, alongside an iterative pipeline for generating realistic sabotage trajectories. We measure both evaluation awareness and a new form of situational awareness termed "prefill awareness", the capability to recognise that prior trajectory content was not self-generated. Opus 4.7 Preview shows notably elevated unprompted evaluation awareness, while prefill awareness remains low across all models. Finally, we discuss limitations including evaluation awareness confounds, limited scenario coverage, and untested pathways to risk beyond safety research sabotage.

Generating step-by-step "how-to" procedures is a key LLM capability: how-to advice is commonly requested in chatbots, and step-by-step planning is critical for reasoning over complex tasks. Yet, measuring and improving procedural validity at scale on real-world tasks remains challenging and understudied. To address this, we introduce How2Everything, a scalable framework to evaluate and improve goal-conditioned procedure generation. Our framework includes How2Mine, which mines 351K procedures from 980K web pages across 14 topics and readily scales to larger corpora. From this pool we build How2Bench, a 7K-example evaluation set balanced across topics. To reliably score model outputs, we develop How2Score, an evaluation protocol that uses an LLM judge to detect whether a generation contains any critical failure that would prevent achieving the goal. For low-cost, reproducible evaluation, we distill a frontier model into an open 8B model, achieving 80.5% agreement with human annotators. How2Bench reveals clear scaling trends across model sizes and training stages, providing signal early in pretraining. Finally, RL using How2Score as a reward improves performance on How2Bench by >10 points across three models without systematic regressions on standard benchmarks, with gains robust to superficial source-document memorization or format compliance. Taken together, How2Everything shows how pretraining web data can support a closed loop of capability evaluation and improvement at scale.

The development of large language models (LLMs) depends on trustworthy evaluation. However, most current evaluations rely on public benchmarks, which are prone to data contamination issues that significantly compromise fairness. Previous researches have focused on constructing dynamic benchmarks to address contamination. However, continuously building new benchmarks is costly and cyclical. In this work, we aim to tackle contamination by analyzing the mechanisms of contaminated models themselves. Through our experiments, we discover that the overestimation of contaminated models is likely due to parameters acquiring shortcut solutions in training. We further propose a novel method for identifying shortcut neurons through comparative and causal analysis. Building on this, we introduce an evaluation method called shortcut neuron patching to suppress shortcut neurons. Experiments validate the effectiveness of our approach in mitigating contamination. Additionally, our evaluation results exhibit a strong linear correlation with MixEval, a recently released trustworthy benchmark, achieving a Spearman coefficient (rho) exceeding 0.95. This high correlation indicates that our method closely reveals true capabilities of the models and is trustworthy. We conduct further experiments to demonstrate the generalizability of our method across various benchmarks and hyperparameter settings. Code: https://github.com/GaryStack/Trustworthy-Evaluation

Imitation-learning policies inherit the quality of the demonstrations they are trained on, and a growing set of curation metrics promise to score and filter low-quality demonstrations automatically. These metrics are each validated on different data with different protocols, so it is unclear which of them actually identify the demonstrations that harm a policy. We build a controlled testbed in which demonstration defects are injected with known type, and audit seven curation metrics along two axes: how well each separates defective from clean demonstrations, and whether training a behavior-cloning policy on each metric's curated subset improves task success. We study two defect regimes. Subtle perturbations (correlated action noise, tremor, truncation) are detectable by multivariate outlier scoring and, once removed, recover the full downstream gap. Structural errors, where the demonstration executes a wrong action at a key moment, are invisible to every action-only metric we test, and two of them are inverted: they score defective demonstrations as higher quality and, used for curation, tend to leave the policy at or below the uncurated baseline rather than above it. Only metrics that examine the state trajectory detect structural errors, and even the best of them recovers just a third of the downstream gap. High detection accuracy does not guarantee downstream improvement. We release the testbed and all curation implementations.

Agent evaluation requires assessing complex multi-step behaviors involving tool use and intermediate reasoning, making it costly and expertise-intensive. A natural question arises: can frontier coding assistants reliably automate this evaluation process? Our study shows that simply prompting coding assistants is insufficient for this task. Without domain-specific evaluation knowledge, frontier coding assistants achieve only a 30% execution success rate and produce over-engineered evaluations averaging 12+ metrics per agent, indicating that strong coding ability does not automatically translate to reliable agent evaluation. We introduce EvalAgent, an AI assistant that automates the end-to-end agent evaluation pipeline. EvalAgent encodes evaluation domain expertise as evaluation skills (procedural instructions, reusable code and templates, and dynamically retrieved API documentation) that compose into a trace-based pipeline producing complete evaluation artifacts including metrics, executable code, and reports. To systematically assess generated evaluations, we introduce a meta-evaluation framework alongside AgentEvalBench, a benchmark comprising 20 agents, each paired with evaluation requirements and test scenarios. We further propose the Eval@1 metric to measure whether generated evaluation code both executes and yields meaningful results on the first run. Our experiments show that EvalAgent produces focused evaluations, improving Eval@1 from 17.5% to 65%, and achieving 79.5% human expert preference over baseline approaches. Further ablation studies show that evaluation skills are critical for handling complex evaluation: removing them causes Eval@1 to drop significantly from 65% to 30%.

As large language models (LLMs) continue to advance, reliable evaluation methods are essential particularly for open-ended, instruction-following tasks. LLM-as-a-Judge enables automatic evaluation using LLMs as evaluators, but its reliability remains uncertain. In this work, we analyze key factors affecting its trustworthiness, focusing on alignment with human judgments and evaluation consistency. Using BIGGENBench and EvalBiasBench, we study the effects of evaluation design, decoding strategies, and Chain-of-Tought (CoT) reasoning in evaluation. Our results show that evaluation criteria are critical for reliability, non-deterministic sampling improves alignment with human preferences over deterministic evaluation, and CoT reasoning offers minimal gains when clear evaluation criteria are present.

There is an increasing imperative to anticipate and understand the performance and safety of generative AI systems in real-world deployment contexts. However, the current evaluation ecosystem is insufficient: Commonly used static benchmarks face validity challenges, and ad hoc case-by-case audits rarely scale. In this piece, we advocate for maturing an evaluation science for generative AI systems. While generative AI creates unique challenges for system safety engineering and measurement science, the field can draw valuable insights from the development of safety evaluation practices in other fields, including transportation, aerospace, and pharmaceutical engineering. In particular, we present three key lessons: Evaluation metrics must be applicable to real-world performance, metrics must be iteratively refined, and evaluation institutions and norms must be established. Applying these insights, we outline a concrete path toward a more rigorous approach for evaluating generative AI systems.

"Effective robustness" measures the extra out-of-distribution (OOD) robustness beyond what can be predicted from the in-distribution (ID) performance. Existing effective robustness evaluations typically use a single test set such as ImageNet to evaluate the ID accuracy. This becomes problematic when evaluating models trained on different data distributions, e.g., comparing models trained on ImageNet vs. zero-shot language-image pre-trained models trained on LAION. In this paper, we propose a new evaluation metric to evaluate and compare the effective robustness of models trained on different data. To do this, we control for the accuracy on multiple ID test sets that cover the training distributions for all the evaluated models. Our new evaluation metric provides a better estimate of effective robustness when there are models with different training data. It may also explain the surprising effective robustness gains of zero-shot CLIP-like models exhibited in prior works that used ImageNet as the only ID test set, while the gains diminish under our new evaluation. Additional artifacts including interactive visualizations are provided at https://shizhouxing.github.io/effective-robustness.

This research aims to build generative language models specialized for the legal domain. The manuscript presents the development of LexGPT models based on GPT-J models and pre-trained with Pile of Law. The foundation model built in this manuscript is the initial step for the development of future applications in the legal domain, such as further training with reinforcement learning from human feedback. Another objective of this manuscript is to assist legal professionals in utilizing language models through the ``No Code'' approach. By fine-tuning models with specialized data and without modifying any source code, legal professionals can create custom language models for downstream tasks with minimum effort and technical knowledge. The downstream task in this manuscript is to turn a LexGPT model into a classifier, although the performance is notably lower than the state-of-the-art result. How to enhance downstream task performance without modifying the model or its source code is a research topic for future exploration.

Scientific research follows multi-turn, multi-step workflows that require proactively searching the literature, consulting figures and tables, and integrating evidence across papers to align experimental settings and support reproducible conclusions. This joint capability is not systematically assessed in existing benchmarks, which largely under-evaluate proactive search, multi-evidence integration and sustained evidence use over time. In this work, we introduce EpiBench, an episodic multi-turn multimodal benchmark that instantiates short research workflows. Given a research task, agents must navigate across papers over multiple turns, align evidence from figures and tables, and use the accumulated evidence in the memory to answer objective questions that require cross paper comparisons and multi-figure integration. EpiBench introduces a process-level evaluation framework for fine-grained testing and diagnosis of research agents. Our experiments show that even the leading model achieves an accuracy of only 29.23% on the hard split, indicating substantial room for improvement in multi-turn, multi-evidence research workflows, providing an evaluation platform for verifiable and reproducible research agents.

Helpful reviews have been essential for the success of e-commerce services, as they help customers make quick purchase decisions and benefit the merchants in their sales. While many reviews are informative, others provide little value and may contain spam, excessive appraisal, or unexpected biases. With the large volume of reviews and their uneven quality, the problem of detecting helpful reviews has drawn much attention lately. Existing methods for identifying helpful reviews primarily focus on review text and ignore the two key factors of (1) who post the reviews and (2) when the reviews are posted. Moreover, the helpfulness votes suffer from scarcity for less popular products and recently submitted (a.k.a., cold-start) reviews. To address these challenges, we introduce a dataset and develop a model that integrates the reviewer's expertise, derived from the past review history of the reviewers, and the temporal dynamics of the reviews to automatically assess review helpfulness. We conduct experiments on our dataset to demonstrate the effectiveness of incorporating these factors and report improved results compared to several well-established baselines.

LLM agents increasingly act over long horizons, where a single trajectory can contain hundreds or thousands of actions. In these settings, outcome-only rewards provide too sparse guidance, failing to inform the model about the goodness of intermediate actions. Dense supervision methods aim to solve this problem by scoring intermediate steps, from intrinsic confidence to self-distillation and embedding similarities. However, it is common practice to evaluate them by measuring the downstream performance of a training pipeline that integrates them. This is expensive, conflates supervision quality with training engineering confounders, and renders different methodological families requiring distinct training setups incomparable. As a result, dense supervision methods are rarely benchmarked on common ground. We introduce QVal, a training-free testbed for directly evaluating dense supervision signals. Given a state-action pair, QVal measures how well a method's score is Q-aligned: whether it orders actions according to the Q-values of a strong reference-policy. This lets us compare signals before any training run and separate signal quality from other engineering choices. We instantiate QVal as QVal-v1.0, benchmarking 21 dense supervision methods across four diverse environments and seven methodological families, with over 1.2K evaluation experiments across six open-weight model backbones. We find that simple prompting baselines consistently outperform recent dense supervision methods from the literature, and that performance clusters strongly by family. These findings hold across model sizes, environments, and observation modalities. QVal is designed to be easily extensible to new environments and methods, enabling researchers to iterate on dense supervision methods before any training run.

Large Language Models (LLMs) are advancing at an amazing speed and have become indispensable across academia, industry, and daily applications. To keep pace with the status quo, this survey probes the core challenges that the rise of LLMs poses for evaluation. We identify and analyze two pivotal transitions: (i) from task-specific to capability-based evaluation, which reorganizes benchmarks around core competencies such as knowledge, reasoning, instruction following, multi-modal understanding, and safety; and (ii) from manual to automated evaluation, encompassing dynamic dataset curation and "LLM-as-a-judge" scoring. Yet, even with these transitions, a crucial obstacle persists: the evaluation generalization issue. Bounded test sets cannot scale alongside models whose abilities grow seemingly without limit. We will dissect this issue, along with the core challenges of the above two transitions, from the perspectives of methods, datasets, evaluators, and metrics. Due to the fast evolving of this field, we will maintain a living GitHub repository (links are in each section) to crowd-source updates and corrections, and warmly invite contributors and collaborators.

The past year has seen over 20 open-source document parsing models, yet thefield still benchmarks almost exclusively on OmniDocBench, a 1,355-pagemanually annotated dataset whose top scores have saturated above 90%. Athree-stage audit pipeline we run on OmniDocBench screens its 21,353evaluator-scored blocks and confirms 2,580 errors (12.08%); combined with overa year of public availability, both annotation quality and contamination riskcall its rankings into question. To address these issues, we presentPureDocBench, a programmatically generated, source-traceable benchmark thatrenders document images from HTML/CSS and produces verifiable annotations fromthe same source, covering 10 domains, 66 subcategories, and 1,475 pages, eachin three versions: clean, digitally degraded, and real-degraded (4,425 imagestotal). Evaluating 40 models spanning pipeline specialists, end-to-endspecialists, and general-purpose VLMs, we find: (i) document parsing is farfrom solved: the best model scores only ~74 out of 100, with a 44.6-point gapbetween the strongest and weakest models; (ii) specialist parsers with <=4Bparameters rival or surpass general VLMs that are 5-100x larger, yet formularecognition remains a shared bottleneck where no model exceeds 67% whenaveraging the formula metric across all three tracks; (iii) general VLMs loseonly 0.99/8.52 Overall points under digital/real degradation versus 4.90/14.21for pipeline specialists, producing ranking reversals that make clean-onlyevaluation misleading for deployment. All data, code, and artifacts arepublicly released.

Scaling laws have transformed our understanding of large language models by linking upstream metrics like cross-entropy loss to design factors such as model size, training data, and compute. However, these conventional laws fail to capture downstream task performance, where context plays a critical role. In this work, we propose a straightforward, interpretable framework that jointly models downstream performance as a function of the training compute and the provided context. We empirically validate our framework by fitting it on the observed downstream performance of extended-context variants of Llama-2-7B and Llama-2-13B across 65,500 unique instances spanning three tasks: arithmetic reasoning, common sense reasoning, and machine translation. Our results demonstrate that our framework accurately models in-distribution downstream performance, generalizes across three orders of magnitude in training compute, and reliably extrapolates performance as the amount of context increases. These findings offer valuable insights into the interplay between training compute and context utilization, providing guidance for designing more efficient long-context LLMs for diverse downstream tasks. Our code is available at https://github.com/wang-research-lab/context-scaling.

As language model (LM) outputs get more and more natural, it is becoming more difficult than ever to evaluate their quality. Simultaneously, increasing LMs' "thinking" time through scaling test-time compute has proven an effective technique to solve challenging problems in domains such as math and code. This raises a natural question: can an LM's evaluation capability also be improved by spending more test-time compute? To answer this, we investigate employing reasoning models-LMs that natively generate long chain-of-thought reasoning-as evaluators. Specifically, we examine methods to leverage more test-time compute by (1) using reasoning models, and (2) prompting these models to evaluate not only the response as a whole (i.e., outcome evaluation) but also assess each step in the response separately (i.e., process evaluation). In experiments, we observe that the evaluator's performance improves monotonically when generating more reasoning tokens, similar to the trends observed in LM-based generation. Furthermore, we use these more accurate evaluators to rerank multiple generations, and demonstrate that spending more compute at evaluation time can be as effective as using more compute at generation time in improving an LM's problem-solving capability.

Rigorous and reproducible evaluation is critical for assessing the state of the art and for guiding scientific advances in Artificial Intelligence. Evaluation is challenging in practice due to several reasons, including benchmark saturation, lack of transparency in methods used for measurement, development challenges in extracting measurements for generative tasks, and, more generally, the extensive number of capabilities required for a well-rounded comparison across models. We make three contributions to alleviate the above challenges. First, we present Eureka, an open-source framework for standardizing evaluations of large foundation models beyond single-score reporting and rankings. Second, we introduce Eureka-Bench as an extensible collection of benchmarks testing capabilities that (i) are still challenging for state-of-the-art models and (ii) represent fundamental but overlooked language and multimodal capabilities. The inherent space for improvement in non-saturated benchmarks enables us to discover meaningful differences between models at a capability level. Third, using Eureka, we conduct an analysis of 12 state-of-the-art models, providing in-depth insights into failure understanding and model comparison, which can be leveraged to plan targeted improvements. In contrast to recent trends in reports and leaderboards showing absolute rankings and claims for one model or another to be the best, our analysis shows that there is no such best model. Different models have different strengths, but there are models that appear more often than others as best performers for some capabilities. Despite the recent improvements, current models still struggle with several fundamental capabilities including detailed image understanding, benefiting from multimodal input when available rather than fully relying on language, factuality and grounding for information retrieval, and over refusals.

This paper presents CG-Eval, the first comprehensive evaluation of the generation capabilities of large Chinese language models across a wide range of academic disciplines. The models' performance was assessed based on their ability to generate accurate and relevant responses to different types of questions in six disciplines, namely, Science and Engineering, Humanities and Social Sciences, Mathematical Calculations, Medical Practitioner Qualification Examination, Judicial Examination, and Certified Public Accountant Examination. This paper also presents Gscore, a composite index derived from the weighted sum of multiple metrics to measure the quality of model's generation against a reference. The test data and test results can be found at http://cgeval.besteasy.com/.

Recent advances have made long-form report-generating systems widely available. This has prompted evaluation frameworks that use LLM-as-judge protocols and claim verification, along with meta-evaluation frameworks that seek to validate these methods. Many of the meta-evaluations estimate an evaluation quality's by comparing its assessments against human pairwise preferences. Prior work, however, suggests that human pairwise preference may be overly simplistic and can fail to capture nuances of expert expectations. We conduct a case study in meta-evaluation for long-form QA benchmarks using ScholarQA-CS2, a benchmark designed for assessing retrieval-augmented deep-research QA in the scientific domain. We comprehensively validate the benchmark through human pairwise preference judgments, then critically examine the strengths, weaknesses, and confounders of this approach. We show that pairwise preference rankings are best suited for system-level evaluation, while explicit metric-wise annotations and expert annotators are critical for reliable metric-level assessment, with subjectivity remaining a key challenge. Based on our findings, we offer practical guidelines for designing future meta-evaluations that better align evaluation methods, annotator expertise, and reporting practices. By surfacing these methodological challenges, we aim to advance evaluation standards for deep-research systems.

Due to the cumbersome nature of human evaluation and limitations of code-based evaluation, Large Language Models (LLMs) are increasingly being used to assist humans in evaluating LLM outputs. Yet LLM-generated evaluators simply inherit all the problems of the LLMs they evaluate, requiring further human validation. We present a mixed-initiative approach to ``validate the validators'' -- aligning LLM-generated evaluation functions (be it prompts or code) with human requirements. Our interface, EvalGen, provides automated assistance to users in generating evaluation criteria and implementing assertions. While generating candidate implementations (Python functions, LLM grader prompts), EvalGen asks humans to grade a subset of LLM outputs; this feedback is used to select implementations that better align with user grades. A qualitative study finds overall support for EvalGen but underscores the subjectivity and iterative process of alignment. In particular, we identify a phenomenon we dub criteria drift: users need criteria to grade outputs, but grading outputs helps users define criteria. What is more, some criteria appears dependent on the specific LLM outputs observed (rather than independent criteria that can be defined a priori), raising serious questions for approaches that assume the independence of evaluation from observation of model outputs. We present our interface and implementation details, a comparison of our algorithm with a baseline approach, and implications for the design of future LLM evaluation assistants.

Existing benchmarks for scientific data analysis evaluate LLMs primarily on code execution or workflow completion, overlooking that scientific analysis serves to support distinct types of scientific claims: hypothesis exploration, statistical inference, mechanistic explanation, each with different assumptions and validity criteria. We introduce SDABench, a benchmark that reorganizes evaluation around six capabilities (descriptive, exploratory, inferential, predictive, causal, and mechanistic) across five domains (Biology, Chemistry, Environment, Geography, Physics). SDABench comprises 527 real-data instances (SDA-Real) and 6000 synthetic instances (SDA-Synth), each in both multiple-choice and open-ended formats, constructed through an automated pipeline. Evaluating 15 representative LLMs, we find that models handle descriptive analysis well but degrade sharply on tasks requiring assumption selection, latent-process modeling, or mechanistic reasoning. SDABench further provides a five-stage error analysis framework that locates where LLMs fail: more advanced models more reliably identify the relevant scope and variables, but still struggle to select appropriate analytical procedures, model variable relationships, and draw valid conclusions.

Evaluation benchmarks are the cornerstone of measuring capabilities of large language models (LLMs), as well as driving progress in said capabilities. Originally designed to make claims about capabilities (or lack thereof) in fully pretrained models, evaluation benchmarks are now also extensively used to decide between various training choices. Despite this widespread usage, we rarely quantify the variance in our evaluation benchmarks, which dictates whether differences in performance are meaningful. Here, we define and measure a range of metrics geared towards measuring variance in evaluation benchmarks, including seed variance across initialisations, and monotonicity during training. By studying a large number of models -- both openly available and pretrained from scratch -- we provide empirical estimates for a variety of variance metrics, with considerations and recommendations for practitioners. We also evaluate the utility and tradeoffs of continuous versus discrete performance measures and explore options for better understanding and reducing this variance. We find that simple changes, such as framing choice tasks (like MMLU) as completion tasks, can often reduce variance for smaller scale (sim7B) models, while more involved methods inspired from human testing literature (such as item analysis and item response theory) struggle to meaningfully reduce variance. Overall, our work provides insights into variance in evaluation benchmarks, suggests LM-specific techniques to reduce variance, and more generally encourages practitioners to carefully factor in variance when comparing models.

Proper evaluations are crucial for better understanding, troubleshooting, interpreting model behaviors and further improving model performance. While using scalar-based error metrics provides a fast way to overview model performance, they are often too abstract to display certain weak spots and lack information regarding important model properties, such as robustness. This not only hinders machine learning models from being more interpretable and gaining trust, but also can be misleading to both model developers and users. Additionally, conventional evaluation procedures often leave researchers unclear about where and how model fails, which complicates model comparisons and further developments. To address these issues, we propose a novel evaluation workflow, named Non-Equivariance Revealed on Orbits (NERO) Evaluation. The goal of NERO evaluation is to turn focus from traditional scalar-based metrics onto evaluating and visualizing models equivariance, closely capturing model robustness, as well as to allow researchers quickly investigating interesting or unexpected model behaviors. NERO evaluation is consist of a task-agnostic interactive interface and a set of visualizations, called NERO plots, which reveals the equivariance property of the model. Case studies on how NERO evaluation can be applied to multiple research areas, including 2D digit recognition, object detection, particle image velocimetry (PIV), and 3D point cloud classification, demonstrate that NERO evaluation can quickly illustrate different model equivariance, and effectively explain model behaviors through interactive visualizations of the model outputs. In addition, we propose consensus, an alternative to ground truths, to be used in NERO evaluation so that model equivariance can still be evaluated with new, unlabeled datasets.

As language models (LMs) become capable of handling a wide range of tasks, their evaluation is becoming as challenging as their development. Most generation benchmarks currently assess LMs using abstract evaluation criteria like helpfulness and harmlessness, which often lack the flexibility and granularity of human assessment. Additionally, these benchmarks tend to focus disproportionately on specific capabilities such as instruction following, leading to coverage bias. To overcome these limitations, we introduce the BiGGen Bench, a principled generation benchmark designed to thoroughly evaluate nine distinct capabilities of LMs across 77 diverse tasks. A key feature of the BiGGen Bench is its use of instance-specific evaluation criteria, closely mirroring the nuanced discernment of human evaluation. We apply this benchmark to assess 103 frontier LMs using five evaluator LMs. Our code, data, and evaluation results are all publicly available at https://github.com/prometheus-eval/prometheus-eval/tree/main/BiGGen-Bench.

Amid the expanding use of pre-training data, the phenomenon of benchmark dataset leakage has become increasingly prominent, exacerbated by opaque training processes and the often undisclosed inclusion of supervised data in contemporary Large Language Models (LLMs). This issue skews benchmark effectiveness and fosters potentially unfair comparisons, impeding the field's healthy development. To address this, we introduce a detection pipeline utilizing Perplexity and N-gram accuracy, two simple and scalable metrics that gauge a model's prediction precision on benchmark, to identify potential data leakages. By analyzing 31 LLMs under the context of mathematical reasoning, we reveal substantial instances of training even test set misuse, resulting in potentially unfair comparisons. These findings prompt us to offer several recommendations regarding model documentation, benchmark setup, and future evaluations. Notably, we propose the "Benchmark Transparency Card" to encourage clear documentation of benchmark utilization, promoting transparency and healthy developments of LLMs. we have made our leaderboard, pipeline implementation, and model predictions publicly available, fostering future research.

In Brazil, all legal professionals must demonstrate their knowledge of the law and its application by passing the OAB exams, the national bar exams. The OAB exams therefore provide an excellent benchmark for the performance of legal information systems since passing the exam would arguably signal that the system has acquired capacity of legal reasoning comparable to that of a human lawyer. This article describes the construction of a new data set and some preliminary experiments on it, treating the problem of finding the justification for the answers to questions. The results provide a baseline performance measure against which to evaluate future improvements. We discuss the reasons to the poor performance and propose next steps.

Deep research systems are widely used for multi-step web research, analysis, and cross-source synthesis, yet their evaluation remains challenging. Existing benchmarks often require annotation-intensive task construction, rely on static evaluation dimensions, or fail to reliably verify facts when citations are missing. To bridge these gaps, we introduce DeepResearchEval, an automated framework for deep research task construction and agentic evaluation. For task construction, we propose a persona-driven pipeline generating realistic, complex research tasks anchored in diverse user profiles, applying a two-stage filter Task Qualification and Search Necessity to retain only tasks requiring multi-source evidence integration and external retrieval. For evaluation, we propose an agentic pipeline with two components: an Adaptive Point-wise Quality Evaluation that dynamically derives task-specific evaluation dimensions, criteria, and weights conditioned on each generated task, and an Active Fact-Checking that autonomously extracts and verifies report statements via web search, even when citations are missing.

Neural scaling laws predict how language model performance improves with increased training inputs. While aggregate metrics like validation loss can follow smooth power-law curves, individual downstream tasks exhibit diverse scaling behaviors: some improve monotonically, others plateau, and some even degrade with scale. We argue that predicting downstream performance from validation loss suffers from two limitations: averaging token-level losses obscures signal, and no simple parametric family can capture the full spectrum of scaling behaviors. To address this, we propose Neural Neural Scaling Laws (NeuNeu), a neural network that frames scaling law prediction as time-series extrapolation. NeuNeu combines temporal context from observed accuracy trajectories with token-level validation losses, learning to predict future performance without the limitations inherent in assuming a specific functional form. Trained entirely on open-source model checkpoints from HuggingFace, NeuNeu achieves 1.99% mean absolute error in predicting model accuracy on 66 downstream tasks -- a 44% reduction compared to logistic scaling laws (3.56% MAE). Furthermore, NeuNeu generalizes zero-shot to unseen model families, architectures, parameter counts, and downstream tasks. Our work suggests that predicting downstream scaling directly from data outperforms parametric alternatives.

Proprietary LMs such as GPT-4 are often employed to assess the quality of responses from various LMs. However, concerns including transparency, controllability, and affordability strongly motivate the development of open-source LMs specialized in evaluations. On the other hand, existing open evaluator LMs exhibit critical shortcomings: 1) they issue scores that significantly diverge from those assigned by humans, and 2) they lack the flexibility to perform both direct assessment and pairwise ranking, the two most prevalent forms of assessment. Additionally, they do not possess the ability to evaluate based on custom evaluation criteria, focusing instead on general attributes like helpfulness and harmlessness. To address these issues, we introduce Prometheus 2, a more powerful evaluator LM than its predecessor that closely mirrors human and GPT-4 judgements. Moreover, it is capable of processing both direct assessment and pair-wise ranking formats grouped with a user-defined evaluation criteria. On four direct assessment benchmarks and four pairwise ranking benchmarks, Prometheus 2 scores the highest correlation and agreement with humans and proprietary LM judges among all tested open evaluator LMs. Our models, code, and data are all publicly available at https://github.com/prometheus-eval/prometheus-eval.

The ability to acquire latent semantics is one of the key properties that determines the performance of language models. One convenient approach to invoke this ability is to prepend metadata (e.g. URLs, domains, and styles) at the beginning of texts in the pre-training data, making it easier for the model to access latent semantics before observing the entire text. Previous studies have reported that this technique actually improves the performance of trained models in downstream tasks; however, this improvement has been observed only in specific downstream tasks, without consistent enhancement in average next-token prediction loss. To understand this phenomenon, we closely investigate how prepending metadata during pre-training affects model performance by examining its behavior using artificial data. Interestingly, we found that this approach produces both positive and negative effects on the downstream tasks. We demonstrate that the effectiveness of the approach depends on whether latent semantics can be inferred from the downstream task's prompt. Specifically, through investigations using data generated by probabilistic context-free grammars, we show that training with metadata helps improve model's performance when the given context is long enough to infer the latent semantics. In contrast, the technique negatively impacts performance when the context lacks the necessary information to make an accurate posterior inference.

Empirical evaluation serves as the primary compass guiding research progress in foundation models. Despite a large body of work focused on training frontier vision-language models (VLMs), approaches to their evaluation remain nascent. To guide their maturation, we propose three desiderata that evaluations should satisfy: (1) faithfulness to the modality and application, (2) discriminability between models of varying quality, and (3) efficiency in compute. Through this lens, we identify critical failure modes that violate faithfulness and discriminability, misrepresenting model capabilities: (i) multiple-choice formats reward guessing, poorly reflect downstream use cases, and saturate early as models improve; (ii) blindly solvable questions, which can be answered without images, constitute up to 70% of some evaluations; and (iii) mislabeled or ambiguous samples compromise up to 42% of examples in certain datasets. Regarding efficiency, the computational burden of evaluating frontier models has become prohibitive: by some accounts, nearly 20% of development compute is devoted to evaluation alone. Rather than discarding existing benchmarks, we curate them via transformation and filtering to maximize fidelity and discriminability. We find that converting multiple-choice questions to generative tasks reveals sharp capability drops of up to 35%. In addition, filtering blindly solvable and mislabeled samples improves discriminative power while simultaneously reducing computational cost. We release DatBench-Full, a cleaned evaluation suite of 33 datasets spanning nine VLM capabilities, and DatBench, a discriminative subset that achieves 13x average speedup (up to 50x) while closely matching the discriminative power of the original datasets. Our work outlines a path toward evaluation practices that are both rigorous and sustainable as VLMs continue to scale.

Reliable evaluation is essential for developing and deploying large language models, yet in practice it often requires substantial manual effort: practitioners must identify appropriate benchmarks, reproduce heterogeneous evaluation codebases, configure dataset schema mappings, and interpret aggregated metrics. To address these challenges, we present One-Eval, an agentic evaluation system that converts natural-language evaluation requests into executable, traceable, and customizable evaluation workflows. One-Eval integrates (i) NL2Bench for intent structuring and personalized benchmark planning, (ii) BenchResolve for benchmark resolution, automatic dataset acquisition, and schema normalization to ensure executability, and (iii) Metrics \& Reporting for task-aware metric selection and decision-oriented reporting beyond scalar scores. The system further incorporates human-in-the-loop checkpoints for review, editing, and rollback, while preserving sample evidence trails for debugging and auditability. Experiments show that One-Eval can execute end-to-end evaluations from diverse natural-language requests with minimal user effort, supporting more efficient and reproducible evaluation in industrial settings. Our framework is publicly available at https://github.com/OpenDCAI/One-Eval.

Understanding causality should be a core requirement of any attempt to build real impact through AI. Due to the inherent unobservability of counterfactuals, large randomised trials (RCTs) are the standard for causal inference. But large experiments are generically expensive, and randomisation carries its own costs, e.g. when suboptimal decisions are trialed. Recent work has proposed more sample-efficient alternatives to RCTs, but these are not adaptable to the downstream application for which the causal effect is sought. In this work, we develop a task-specific approach to experimental design and derive sampling strategies customised to particular downstream applications. Across a range of important tasks, real-world datasets, and sample sizes, our method outperforms other benchmarks, e.g. requiring an order-of-magnitude less data to match RCT performance on targeted marketing tasks.

Multimodal Large Language Models (MLLMs) hold significant promise for revolutionizing traditional education and reducing teachers' workload. However, accurately interpreting unconstrained STEM student handwritten solutions with intertwined mathematical formulas, diagrams, and textual reasoning poses a significant challenge due to the lack of authentic and domain-specific benchmarks. Additionally, current evaluation paradigms predominantly rely on the outcomes of downstream tasks (e.g., auto-grading), which often probe only a subset of the recognized content, thereby failing to capture the MLLMs' understanding of complex handwritten logic as a whole. To bridge this gap, we release EDU-CIRCUIT-HW, a dataset consisting of 1,300+ authentic student handwritten solutions from a university-level STEM course. Utilizing the expert-verified verbatim transcriptions and grading reports of student solutions, we simultaneously evaluate various MLLMs' upstream recognition fidelity and downstream auto-grading performance. Our evaluation uncovers an astonishing scale of latent failures within MLLM-recognized student handwritten content, highlighting the models' insufficient reliability for auto-grading and other understanding-oriented applications in high-stakes educational settings. As a potential solution, we present a case study demonstrating that leveraging identified error patterns to preemptively detect and correct recognition errors, while requiring only minimal human intervention (e.g., routing 3.3% of assignments to human graders and the remainder to the GPT-5.1 grader), can effectively enhance the robustness of the deployed AI-enabled grading system. Code and dataset are available in this GitHub repo: https://gt-learning-innovation.github.io/CIRCUIT_EDU_HW_ACL.

Existing evaluations of foundation models, including recent human-centric approaches, fail to capture what truly matters: user's experience during interaction. Current methods treat evaluation as a matter of output correctness alone, overlooking that user satisfaction emerges from the interplay between response quality and interaction, which limits their ability to account for the mechanisms underlying user experience. To address this gap, we introduce QoNext, the first framework that adapts Quality of Experience (QoE) principles from networking and multimedia to the assessment of foundation models. QoNext identifies experiential factors that shape user experience and incorporates them into controlled experiments, where human ratings are collected under varied configurations. From these studies we construct a QoE-oriented database and train predictive models that estimate perceived user experience from measurable system parameters. Our results demonstrate that QoNext not only enables proactive and fine-grained evaluation but also provides actionable guidance for productized services of optimizing foundation models in practice.

In this paper, we uncover a systematic bias in the evaluation paradigm of adopting large language models~(LLMs), e.g., GPT-4, as a referee to score and compare the quality of responses generated by candidate models. We find that the quality ranking of candidate responses can be easily hacked by simply altering their order of appearance in the context. This manipulation allows us to skew the evaluation result, making one model appear considerably superior to the other, e.g., Vicuna-13B could beat ChatGPT on 66 over 80 tested queries with ChatGPT as an evaluator. To address this issue, we propose a calibration framework with three simple yet effective strategies: 1) Multiple Evidence Calibration, which requires the evaluator model to generate multiple evaluation evidence before assigning ratings; 2) Balanced Position Calibration, which aggregates results across various orders to determine the final score; 3) Human-in-the-Loop Calibration, which introduces a balanced position diversity entropy to measure the difficulty of each example and seeks human assistance when needed. We also manually annotate the "win/tie/lose" outcomes of responses from ChatGPT and Vicuna-13B in the Vicuna Benchmark's question prompt, and extensive experiments demonstrate that our approach successfully mitigates evaluation bias, resulting in closer alignment with human judgments. We release our code and human annotation at https://github.com/i-Eval/FairEval to facilitate future research.

Web-crawled pretraining datasets underlie the impressive "zero-shot" evaluation performance of multimodal models, such as CLIP for classification/retrieval and Stable-Diffusion for image generation. However, it is unclear how meaningful the notion of "zero-shot" generalization is for such multimodal models, as it is not known to what extent their pretraining datasets encompass the downstream concepts targeted for during "zero-shot" evaluation. In this work, we ask: How is the performance of multimodal models on downstream concepts influenced by the frequency of these concepts in their pretraining datasets? We comprehensively investigate this question across 34 models and five standard pretraining datasets (CC-3M, CC-12M, YFCC-15M, LAION-400M, LAION-Aesthetics), generating over 300GB of data artifacts. We consistently find that, far from exhibiting "zero-shot" generalization, multimodal models require exponentially more data to achieve linear improvements in downstream "zero-shot" performance, following a sample inefficient log-linear scaling trend. This trend persists even when controlling for sample-level similarity between pretraining and downstream datasets, and testing on purely synthetic data distributions. Furthermore, upon benchmarking models on long-tailed data sampled based on our analysis, we demonstrate that multimodal models across the board perform poorly. We contribute this long-tail test set as the "Let it Wag!" benchmark to further research in this direction. Taken together, our study reveals an exponential need for training data which implies that the key to "zero-shot" generalization capabilities under large-scale training paradigms remains to be found.

The validity of AI safety evaluations depends on models behaving consistently across controlled and deployment settings. Prior work has identified test-time contextual cues, such as hypothetical scenarios, as a source of verbalized evaluation awareness and subsequent behavioral shift. In this paper, we investigate a potential explanation of this phenomenon: evaluation meta-knowledge, defined as parametric knowledge about the structural traits that characterize evaluations. Similar to dataset contamination, where benchmark exposure leads to higher performance through memorization, we hypothesize that models trained on texts describing evaluation practices may implicitly learn to recognize and respond to evaluation-like contexts, for instance, through exposure to scientific articles or social media posts about AI benchmarking. To test this, we fine-tune models on synthetic documents describing evaluation traits such as verifiable structures or moral dilemmas. Evaluating this fine-tuned model on six safety benchmarks, we find that it is significantly safer than the base model and control model. This behavioral shift persists even when restricting the analysis to responses lacking explicit verbalization of evaluation awareness. Our results demonstrate that evaluation meta-knowledge may inflate safety benchmark performance, introducing a novel confounder that is independent of explicit memorization or verbalized evaluation awareness, thus, challenging to detect. These findings have important implications for the design and interpretation of AI safety evaluations. Our code and models are available at https://github.com/compass-group-tue/arxiv2026_evaluation_meta_knowledge.