Daily Papers

Multimodal large language models (MLLMs) are increasingly deployed in real-world, agentic settings where outputs must not only be correct, but also conform to predefined data schemas. Despite recent progress in structured generation in textual domain, there is still no benchmark that systematically evaluates schema-grounded information extraction and reasoning over visual inputs. In this work, we conduct a comprehensive study of visual structural output capabilities for MLLMs with our carefully designed SO-Bench benchmark. Covering four visual domains, including UI screens, natural images, documents, and charts, SO-Bench is built from over 6.5K diverse JSON schemas and 1.8K curated image-schema pairs with human-verified quality. Benchmarking experiments on open-sourced and frontier proprietary models reveal persistent gaps in predicting accurate, schema compliant outputs, highlighting the need for better multimodal structured reasoning. Beyond benchmarking, we further conduct training experiments to largely improve the model's structured output capability. We plan to make the benchmark available to the community.

Automated interpretability pipelines generate natural language descriptions for the concepts represented by features in large language models (LLMs), such as plants or the first word in a sentence. These descriptions are derived using inputs that activate the feature, which may be a dimension or a direction in the model's representation space. However, identifying activating inputs is costly, and the mechanistic role of a feature in model behavior is determined both by how inputs cause a feature to activate and by how feature activation affects outputs. Using steering evaluations, we reveal that current pipelines provide descriptions that fail to capture the causal effect of the feature on outputs. To fix this, we propose efficient, output-centric methods for automatically generating feature descriptions. These methods use the tokens weighted higher after feature stimulation or the highest weight tokens after applying the vocabulary "unembedding" head directly to the feature. Our output-centric descriptions better capture the causal effect of a feature on model outputs than input-centric descriptions, but combining the two leads to the best performance on both input and output evaluations. Lastly, we show that output-centric descriptions can be used to find inputs that activate features previously thought to be "dead".

The rapid development of generative AI (GenAI) models in computer vision necessitates effective evaluation methods to ensure their quality and fairness. Existing tools primarily focus on dataset quality assurance and model explainability, leaving a significant gap in GenAI output evaluation during model development. Current practices often depend on developers' subjective visual assessments, which may lack scalability and generalizability. This paper bridges this gap by conducting a formative study with GenAI model developers in an industrial setting. Our findings led to the development of GenLens, a visual analytic interface designed for the systematic evaluation of GenAI model outputs during the early stages of model development. GenLens offers a quantifiable approach for overviewing and annotating failure cases, customizing issue tags and classifications, and aggregating annotations from multiple users to enhance collaboration. A user study with model developers reveals that GenLens effectively enhances their workflow, evidenced by high satisfaction rates and a strong intent to integrate it into their practices. This research underscores the importance of robust early-stage evaluation tools in GenAI development, contributing to the advancement of fair and high-quality GenAI models.

Using Large Language Models (LLMs) for Process Mining (PM) tasks is becoming increasingly essential, and initial approaches yield promising results. However, little attention has been given to developing strategies for evaluating and benchmarking the utility of incorporating LLMs into PM tasks. This paper reviews the current implementations of LLMs in PM and reflects on three different questions. 1) What is the minimal set of capabilities required for PM on LLMs? 2) Which benchmark strategies help choose optimal LLMs for PM? 3) How do we evaluate the output of LLMs on specific PM tasks? The answer to these questions is fundamental to the development of comprehensive process mining benchmarks on LLMs covering different tasks and implementation paradigms.

Recent Graphical User Interface (GUI) agents replicate the R1-Zero paradigm, coupling online Reinforcement Learning (RL) with explicit chain-of-thought reasoning prior to object grounding and thereby achieving substantial performance gains. In this paper, we first conduct extensive analysis experiments of three key components of that training pipeline: input design, output evaluation, and policy update-each revealing distinct challenges arising from blindly applying general-purpose RL without adapting to GUI grounding tasks. Input design: Current templates encourage the model to generate chain-of-thought reasoning, but longer chains unexpectedly lead to worse grounding performance. Output evaluation: Reward functions based on hit signals or box area allow models to exploit box size, leading to reward hacking and poor localization quality. Policy update: Online RL tends to overfit easy examples due to biases in length and sample difficulty, leading to under-optimization on harder cases. To address these issues, we propose three targeted solutions. First, we adopt a Fast Thinking Template that encourages direct answer generation, reducing excessive reasoning during training. Second, we incorporate a box size constraint into the reward function to mitigate reward hacking. Third, we revise the RL objective by adjusting length normalization and adding a difficulty-aware scaling factor, enabling better optimization on hard samples. Our GUI-G1-3B, trained on 17K public samples with Qwen2.5-VL-3B-Instruct, achieves 90.3% accuracy on ScreenSpot and 37.1% on ScreenSpot-Pro. This surpasses all prior models of similar size and even outperforms the larger UI-TARS-7B, establishing a new state-of-the-art in GUI agent grounding. The project repository is available at https://github.com/Yuqi-Zhou/GUI-G1.

Large language models (LLMs) inevitably make mistakes when performing step-by-step mathematical reasoning. Process Reward Models (PRMs) have emerged as a promising solution by evaluating each reasoning step. However, existing PRMs typically output evaluation scores directly, limiting both learning efficiency and evaluation accuracy, which is further exacerbated by the scarcity of annotated data. To address these issues, we propose Reasoning-Driven Process Reward Modeling (R-PRM). First, we leverage stronger LLMs to generate seed data from limited annotations, effectively bootstrapping our model's reasoning capabilities and enabling comprehensive step-by-step evaluation. Second, we further enhance performance through preference optimization, without requiring additional annotated data. Third, we introduce inference-time scaling to fully harness the model's reasoning potential. Extensive experiments demonstrate R-PRM's effectiveness: on ProcessBench and PRMBench, it surpasses strong baselines by 11.9 and 8.5 points in F1 scores, respectively. When applied to guide mathematical reasoning, R-PRM achieves consistent accuracy improvements of over 8.5 points across six challenging datasets. Further analysis reveals that R-PRM exhibits more comprehensive evaluation and stronger generalization capabilities, thereby highlighting its significant potential.

To keep pace with the increasing pace of large language models (LLM) development, model output evaluation has transitioned away from time-consuming human evaluation to automatic evaluation, where LLMs themselves are tasked with assessing and critiquing other model outputs. LLM-as-judge models are a class of generative evaluators that excel in evaluating relatively simple domains, like chat quality, but struggle in reasoning intensive domains where model responses contain more substantive and challenging content. To remedy existing judge shortcomings, we explore training judges with reinforcement learning (RL). We make three key contributions: (1) We propose the Equivalent Initial State Group Relative Policy Optimization (EIS-GRPO) algorithm, which allows us to train our judge to be robust to positional biases that arise in more complex evaluation settings. (2) We introduce ReasoningJudgeBench, a benchmark that evaluates judges in diverse reasoning settings not covered by prior work. (3) We train Judge for Reasoning (J4R), a 7B judge trained with EIS-GRPO that outperforms GPT-4o and the next best small judge by 6.7% and 9%, matching or exceeding the performance of larger GRPO-trained judges on both JudgeBench and ReasoningJudgeBench.

Recent advances in Chain-of-Thought (CoT) prompting have substantially improved the reasoning capabilities of large language models (LLMs), but have also introduced their computational efficiency as a new attack surface. In this paper, we propose BadThink, the first backdoor attack designed to deliberately induce "overthinking" behavior in CoT-enabled LLMs while ensuring stealth. When activated by carefully crafted trigger prompts, BadThink manipulates the model to generate inflated reasoning traces - producing unnecessarily redundant thought processes while preserving the consistency of final outputs. This subtle attack vector creates a covert form of performance degradation that significantly increases computational costs and inference time while remaining difficult to detect through conventional output evaluation methods. We implement this attack through a sophisticated poisoning-based fine-tuning strategy, employing a novel LLM-based iterative optimization process to embed the behavior by generating highly naturalistic poisoned data. Our experiments on multiple state-of-the-art models and reasoning tasks show that BadThink consistently increases reasoning trace lengths - achieving an over 17x increase on the MATH-500 dataset - while remaining stealthy and robust. This work reveals a critical, previously unexplored vulnerability where reasoning efficiency can be covertly manipulated, demonstrating a new class of sophisticated attacks against CoT-enabled systems.

This paper introduces ExpertLongBench, an expert-level benchmark containing 11 tasks from 9 domains that reflect realistic expert workflows and applications. Beyond question answering, the application-driven tasks in ExpertLongBench demand long-form outputs that can exceed 5,000 tokens and strict adherence to domain-specific requirements. Notably, each task in ExpertLongBench includes a rubric, designed or validated by domain experts, to specify task requirements and guide output evaluation. Furthermore, we propose CLEAR, an evaluation framework that supports accurate evaluation of long-form model outputs in our benchmark. To achieve fine-grained, expert-aligned evaluation, CLEAR derives checklists from both model outputs and references by extracting information corresponding to items in the task-specific rubric. Checklist items for model outputs are then compared with corresponding items for reference outputs to assess their correctness, enabling grounded evaluation. We benchmark 11 large language models (LLMs) and analyze components in CLEAR, showing that (1) existing LLMs, with the top performer achieving only a 26.8% F1 score, require significant improvement for expert-level tasks; (2) models can generate content corresponding to the required aspects, though often not accurately; and (3) accurate checklist extraction and comparison in CLEAR can be achieved by open-weight models for more scalable and low-cost usage.

LLM agents increasingly run inside execution harnesses that dispatch tools, allocate resources, and route messages between specialized components. However, a harness can return a correct, benign answer over a trajectory that accesses unauthorized resources or leaks context to the wrong agent. Output-level evaluation cannot see these failures, yet most safety benchmarks score only final outputs or terminal states, even though many violations occur mid-trajectory rather than at termination. The central question is whether the harness respects user intent, permission boundaries, and information-flow constraints throughout execution. To address this gap, we propose HarnessAudit, a framework that audits full execution trajectories across boundary compliance, execution fidelity, and system stability, with a focus on multi-agent harnesses where these risks are most pronounced. We further introduce HarnessAudit-Bench, a benchmark of 210 tasks across eight real-world domains, instantiated in both single-agent and multi-agent configurations with embedded safety constraints. Evaluating ten harness configurations across frontier models and three multi-agent frameworks, we find that: (i) task completion is misaligned with safe execution, and violations accumulate with trajectory length; (ii) safety risks vary across domains, task types, and agent roles; (iii) most violations concentrate in resource access and inter-agent information transfer; and (iv) multi-agent collaboration expands the safety risk surface, while harness design sets the upper bound of safe deployment.

Multi-agent large language model (LLM) systems have shown strong potential in complex reasoning and collaborative decision-making tasks. However, most existing coordination schemes rely on static or full-context routing strategies, which lead to excessive token consumption, redundant memory exposure, and limited adaptability across interaction rounds. We introduce RCR-Router, a modular and role-aware context routing framework designed to enable efficient, adaptive collaboration in multi-agent LLMs. To our knowledge, this is the first routing approach that dynamically selects semantically relevant memory subsets for each agent based on its role and task stage, while adhering to a strict token budget. A lightweight scoring policy guides memory selection, and agent outputs are iteratively integrated into a shared memory store to facilitate progressive context refinement. To better evaluate model behavior, we further propose an Answer Quality Score metric that captures LLM-generated explanations beyond standard QA accuracy. Experiments on three multi-hop QA benchmarks -- HotPotQA, MuSiQue, and 2WikiMultihop -- demonstrate that RCR-Router reduces token usage (up to 30%) while improving or maintaining answer quality. These results highlight the importance of structured memory routing and output-aware evaluation in advancing scalable multi-agent LLM systems.

Large Language Models (LLMs) have demonstrated promising capabilities to generate responses that exhibit consistent personality traits. Despite the major attempts to analyze personality expression through output-based evaluations, little is known about how such traits are internally encoded within LLM parameters. In this paper, we introduce a layer-wise probing framework to systematically investigate the layer-wise capability of LLMs in encoding personality for responding. We conduct probing experiments on 11 open-source LLMs over the PersonalityEdit benchmark and find that LLMs predominantly encode personality for responding in their middle and upper layers, with instruction-tuned models demonstrating a slightly clearer separation of personality traits. Furthermore, by interpreting the trained probing hyperplane as a layer-wise boundary for each personality category, we propose a layer-wise perturbation method to edit the personality expressed by LLMs during inference. Our results show that even when the prompt explicitly specifies a particular personality, our method can still successfully alter the response personality of LLMs. Interestingly, the difficulty of converting between certain personality traits varies substantially, which aligns with the representational distances in our probing experiments. Finally, we conduct a comprehensive MMLU benchmark evaluation and time overhead analysis, demonstrating that our proposed personality editing method incurs only minimal degradation in general capabilities while maintaining low training costs and acceptable inference latency. Our code is publicly available at https://github.com/universe-sky/probing-then-editing-personality.

Pretrained large language models (LLMs) have demonstrated in-context learning (ICL) capabilities for numerical inference over sequences serialized as text. Prior work has identified and characterized this form of numerical inference primarily through output-level evaluations such as prediction error. However, how numerical information is organized within LLM representations remains much less understood. To study this internal organization, we adopt a graph signal processing perspective in which attention induces a weighted graph over tokens, while token hidden states define signals on its nodes. Quantitative graph-spectral diagnostics and qualitative token-graph visualizations reveal that representations become more clearly differentiated by input dynamical complexity as context length increases. Simpler inputs produce attention-induced token graphs with stronger global connectivity and smoother, spectrally concentrated hidden-state signals, whereas more complex inputs produce more localized graphs and hidden-state signals with broader spectral support and greater high-frequency energy. Together, these findings point to systematic, context-dependent internal signatures associated with numerical ICL that are conserved across model families.

Fine-tuned large language models can exhibit reward-hacking behavior arising from emergent misalignment, which is difficult to detect from final outputs alone. While prior work has studied reward hacking at the level of completed responses, it remains unclear whether such behavior can be identified during generation. We propose an activation-based monitoring approach that detects reward-hacking signals from internal representations as a model generates its response. Our method trains sparse autoencoders on residual stream activations and applies lightweight linear classifiers to produce token-level estimates of reward-hacking activity. Across multiple model families and fine-tuning mixtures, we find that internal activation patterns reliably distinguish reward-hacking from benign behavior, generalize to unseen mixed-policy adapters, and exhibit model-dependent temporal structure during chain-of-thought reasoning. Notably, reward-hacking signals often emerge early, persist throughout reasoning, and can be amplified by increased test-time compute in the form of chain-of-thought prompting under weakly specified reward objectives. These results suggest that internal activation monitoring provides a complementary and earlier signal of emergent misalignment than output-based evaluation, supporting more robust post-deployment safety monitoring for fine-tuned language models.

Large reasoning models (LRMs) generate extended solutions, yet it remains unclear whether these traces reflect substantive internal computation or merely verbosity and overthinking. Although recent hidden-state analyses suggest that internal representations carry correctness-related signals, their coarse aggregations may obscure the token and layer structure underlying reasoning computation. We investigate hidden-state transitions across decoding steps and layers, and identify a distinct spatiotemporal pattern in LRMs: successful trajectories exhibit broad temporal dynamics with localized layer-wise concentration, while this structure is weaker in non-reasoning models and knowledge-heavy domains. We formalize this characteristic as Spatiotemporal Amplitude of Latent Transition (StALT), a training-free trajectory statistic that summarizes temporal changes between adjacent tokens weighted by within-token layer saliency. Across diverse models and benchmarks, StALT reliably separates correct from incorrect trajectories in reasoning-intensive regimes, providing a competitive label-free correctness signal alongside strong output-space and length-based baselines. Intervention analyses further show that this spatiotemporal amplitude responds systematically to manipulations that increase or reduce the demand for internal reasoning, supporting its association with latent reasoning dynamics in LRMs. These findings provide empirical evidence that LRMs exhibit measurable hidden-state dynamics and offer a practical probe for understanding internal computation beyond output-based evaluation.

Accurate confidence estimation is essential for trustworthy large language models (LLMs) systems, as it empowers the user to determine when to trust outputs and enables reliable deployment in safety-critical applications. Current confidence estimation methods for LLMs neglect the relevance between responses and contextual information, a crucial factor in output quality evaluation, particularly in scenarios where background knowledge is provided. To bridge this gap, we propose CRUX (Context-aware entropy Reduction and Unified consistency eXamination), the first framework that integrates context faithfulness and consistency for confidence estimation via two novel metrics. First, contextual entropy reduction represents data uncertainty with the information gain through contrastive sampling with and without context. Second, unified consistency examination captures potential model uncertainty through the global consistency of the generated answers with and without context. Experiments across three benchmark datasets (CoQA, SQuAD, QuAC) and two domain-specific datasets (BioASQ, EduQG) demonstrate CRUX's effectiveness, achieving the highest AUROC than existing baselines.

LLM self-evaluation relies on the LLM's own ability to estimate response correctness, which can greatly improve its deployment reliability. In this research track, we propose the Chain-of-Embedding (CoE) in the latent space to enable LLMs to perform output-free self-evaluation. CoE consists of all progressive hidden states produced during the inference time, which can be treated as the latent thinking path of LLMs. We find that when LLMs respond correctly and incorrectly, their CoE features differ, these discrepancies assist us in estimating LLM response correctness. Experiments in four diverse domains and seven LLMs fully demonstrate the effectiveness of our method. Meanwhile, its label-free design intent without any training and millisecond-level computational cost ensures real-time feedback in large-scale scenarios. More importantly, we provide interesting insights into LLM response correctness from the perspective of hidden state changes inside LLMs.

"Talk short. Drop grammar. Save token." This caveman style is widely promoted as a way to cut inference cost, but whether it actually saves anything depends on which channel (the user's prompt or the model's response) is being compressed. We present Cavewoman, a two-channel evaluation protocol that scores every generation on task accuracy, realized per-item cost, and reference-text agreement against the model's unconstrained reference. We evaluate eight models on five datasets at five reduction levels, with both channels measured on the same items. Output compression cuts realized cost on most API models (1.4-2.4x per model, up to 3x in the best case) and on all four open-weight models under public-tier pricing. Input compression has the opposite effect, a strict lose-lose: it raises net cost rather than lowering it (~1.15x on the five-benchmark mean, up to 1.8x on the worst dataset and 2.7x under stronger compression), because models compensate with longer responses even as accuracy collapses. Under the same setting, surface text diverges from the unconstrained reference: on the non-reasoning models, roughly half of all generations are correct yet their surface text no longer entails the model's own unconstrained baseline generation. The divergence survives length-controlled re-scoring, multiple-comparisons correction, and replication under complementary semantic measures. Code and data are available at https://github.com/danielle34/cavewoman.

Research on Large Language Models (LLMs) studies output variation across generation, reasoning, alignment, and representational analysis, often under the umbrella of "diversity." Yet the terminology remains fragmented, largely because the normative objectives underlying tasks are rarely made explicit. We introduce the Magic, Madness, Heaven, Sin framework, which models output variation along a homogeneity-heterogeneity axis, where valuation is determined by the task and its normative objective. We organize tasks into four normative contexts: epistemic (factuality), interactional (user utility), societal (representation), and safety (robustness). For each, we examine the failure modes and vocabulary such as hallucination, mode collapse, bias, and erasure through which variation is studied. We apply the framework to analyze all pairwise cross-contextual interactions, revealing that optimizing for one objective, such as improving safety, can inadvertently harm demographic representation or creative diversity. We argue for context-aware evaluation of output variation, reframing it as a property shaped by task objectives rather than a model's intrinsic trait.

Recent statements about the impressive capabilities of large language models (LLMs) are usually supported by evaluating on open-access benchmarks. Considering the vast size and wide-ranging sources of LLMs' training data, it could explicitly or implicitly include test data, leading to LLMs being more susceptible to data contamination. However, due to the opacity of training data, the black-box access of models, and the rapid growth of synthetic training data, detecting and mitigating data contamination for LLMs faces significant challenges. In this paper, we propose CDD, which stands for Contamination Detection via output Distribution for LLMs. CDD necessitates only the sampled texts to detect data contamination, by identifying the peakedness of LLM's output distribution. To mitigate the impact of data contamination in evaluation, we also present TED: Trustworthy Evaluation via output Distribution, based on the correction of LLM's output distribution. To facilitate this study, we introduce two benchmarks, i.e., DetCon and ComiEval, for data contamination detection and contamination mitigation evaluation tasks. Extensive experimental results show that CDD achieves the average relative improvements of 21.8\%-30.2\% over other contamination detection approaches in terms of Accuracy, F1 Score, and AUC metrics, and can effectively detect implicit contamination. TED substantially mitigates performance improvements up to 66.9\% attributed to data contamination across various contamination setups. In real-world applications, we reveal that ChatGPT exhibits a high potential to suffer from data contamination on HumanEval benchmark.

Large Language Models (LLMs) exhibit remarkable fluency and competence across various natural language tasks. However, recent research has highlighted their sensitivity to variations in input prompts. To deploy LLMs in a safe and reliable manner, it is crucial for their outputs to be consistent when prompted with expressions that carry the same meaning or intent. While some existing work has explored how state-of-the-art LLMs address this issue, their evaluations have been confined to assessing lexical equality of single- or multi-word answers, overlooking the consistency of generative text sequences. For a more comprehensive understanding of the consistency of LLMs in open-ended text generation scenarios, we introduce a general measure of semantic consistency, and formulate multiple versions of this metric to evaluate the performance of various LLMs. Our proposal demonstrates significantly higher consistency and stronger correlation with human evaluations of output consistency than traditional metrics based on lexical consistency. Finally, we propose a novel prompting strategy, called Ask-to-Choose (A2C), to enhance semantic consistency. When evaluated for closed-book question answering based on answer variations from the TruthfulQA benchmark, A2C increases accuracy metrics for pretrained and finetuned LLMs by up to 47%, and semantic consistency metrics for instruction-tuned models by up to 7-fold.

State-of-the-art large language models (LLMs) exhibit impressive problem-solving capabilities but may struggle with complex reasoning and factual correctness. Existing methods harness the strengths of chain-of-thought and retrieval-augmented generation (RAG) to decompose a complex problem into simpler steps and apply retrieval to improve factual correctness. These methods work well on straightforward reasoning tasks but often falter on challenging tasks such as competitive programming and mathematics, due to frequent reasoning errors and irrelevant knowledge retrieval. To address this, we introduce Critic-guided planning with Retrieval-augmentation, CR-Planner, a novel framework that leverages fine-tuned critic models to guide both reasoning and retrieval processes through planning. CR-Planner solves a problem by iteratively selecting and executing sub-goals. Initially, it identifies the most promising sub-goal from reasoning, query generation, and retrieval, guided by rewards given by a critic model named sub-goal critic. It then executes this sub-goal through sampling and selecting the optimal output based on evaluations from another critic model named execution critic. This iterative process, informed by retrieved information and critic models, enables CR-Planner to effectively navigate the solution space towards the final answer. We employ Monte Carlo Tree Search to collect the data for training the critic models, allowing for a systematic exploration of action sequences and their long-term impacts. We validate CR-Planner on challenging domain-knowledge-intensive and reasoning-heavy tasks, including competitive programming, theorem-driven math reasoning, and complex domain retrieval problems. Our experiments demonstrate that CR-Planner significantly outperforms baselines, highlighting its effectiveness in addressing challenging problems by improving both reasoning and retrieval.

Comprehensive evaluation of Large Language Models (LLMs) is an open research problem. Existing evaluations rely on deterministic point estimates generated via greedy decoding. However, we find that deterministic evaluations fail to capture the whole output distribution of a model, yielding inaccurate estimations of model capabilities. This is particularly problematic in critical contexts such as unlearning and alignment, where precise model evaluations are crucial. To remedy this, we introduce the first formal probabilistic evaluation framework for LLMs. Namely, we propose novel metrics with high probability guarantees concerning the output distribution of a model. Our metrics are application-independent and allow practitioners to make more reliable estimates about model capabilities before deployment. Our experimental analysis reveals that deterministic evaluations falsely indicate successful unlearning and alignment, whereas our probabilistic evaluations better capture model capabilities. We show how to overcome challenges associated with probabilistic outputs in a case study on unlearning by introducing (1) a novel loss based on entropy optimization, and (2) adaptive temperature scaling. We demonstrate that our approach significantly enhances unlearning in probabilistic settings on recent benchmarks. Overall, our proposed shift from point estimates to probabilistic evaluations of output distributions represents an important step toward comprehensive evaluations of LLMs. Code available at https://www.cs.cit.tum.de/daml/probabilistic-unlearning/.

Large language models (LLMs) are increasingly deployed as agents that rely on system prompts to use tools and complete tasks. Such deployments impose distinct operational requirements, including appropriate tool use, concise prompts and solution paths, and compliance with safety and formatting policies. For many practitioners, however, assembling domain-specific supervised data to post-train models to meet these requirements is infeasible. We introduce CAPO (Constraint-Aware Prompt Optimization), a primal-dual method that combines pool-based rewrites with adaptive constraint weighting to optimize system prompts under explicit operational constraints. Across agentic benchmarks, CAPO more reliably reaches empirically feasible operating points while improving task performance. CAPO also generalizes beyond agentic settings, achieving strong results on assistant-style evaluations with output-format and safety/privacy constraints. We further introduce DCAPO (Dynamically Trained CAPO), which trains a feedback- and dual-conditioned rewriter with pool-based GRPO while keeping the task agent frozen. Across task agents of different sizes, DCAPO produces a feasible prompt in every evaluated domain and matches or improves the task accuracy achieved by the evaluated baselines. A surrogate analysis characterizes how finite-pool and discrete-rewrite errors enter the inexact primal-dual procedure.

Agentic recommendation pursues shopper goals through sequences of context-dependent decisions. Developing and evaluating such policies requires faithful recommendation world models that capture both the interactive shopping environment and users' behavioral responses over long interactions. Existing resources rarely provide long-horizon, fine-grained real-user trajectories or support trajectory-level evaluation and offline validation against production A/B results. We introduce UserArena, a real-world e-commerce interactive benchmark containing 64,729 session trajectories from 11,010 users, spanning 566,102 items and 13 action types. UserArena pairs these trajectories with an interactive environment and supports offline validation against outcomes from a controlled production A/B test, enabling five-level evaluation from output validity and local decisions to sequential behavior, personalization, and outcomes. To improve simulator fidelity over these long-horizon interactions, we further propose On-Policy Trajectory Distillation (OPTD), which uses trajectory-level intent to provide dense teacher targets on student-generated rollouts. Experiments show that OPTD achieves the strongest overall fidelity among the evaluated simulators under UserArena's evaluation protocol and most closely reproduces the production policy effect in offline simulation.

This paper presents an eight-week observational comparison of 68 single-file HTML generations collected across 17 public experiments in the "HTML AI Battle" project between December 10, 2025 and February 4, 2026. Four reasoning model families, GPT, Gemini, Grok, and Claude, were compared under a fixed public-interface protocol with no custom instructions, no personality tuning, and no repair prompts. Each output was evaluated from a rendered browser video using human scores and a Gemini LLM-as-a-judge layer for prompt adherence, functional correctness, and UI quality, then packaged into a standardized social-media protocol spanning X (Twitter), TikTok, and YouTube. The tracker was also used for two supervised predictive analyses: an experiment-level model for 24-hour X impressions and a generation-level model for HTML verbosity. Under this protocol, Claude was the strongest and most consistent family, leading mean performance and winning 9/17 prompts under the primary human weighted score. Longer measured reasoning time was not associated with higher quality overall. Gemini as a judge was significantly more lenient than the human evaluator on functional correctness and overall performance, while stable self-favoring bias remained unresolved. The exploratory X-impressions model remained weak under post-screen cross-validation (MAE = 46,874, R^2 = -0.377), whereas the HTML-lines model performed better, with a model-family-only baseline outperforming prompt-aware alternatives (MAE = 135.2, R^2 = 0.576). Overall, selected pre-publication technical/audio variables were not sufficient to predict 24-hour X reach, while code verbosity was driven much more by model family than by prompt wording. The comparisons remain observational and are limited by public-interface drift, access-path differences, and one primary human scorer.

Large language models (LLMs) are notorious for hallucinating, i.e., producing erroneous claims in their output. Such hallucinations can be dangerous, as occasional factual inaccuracies in the generated text might be obscured by the rest of the output being generally factual, making it extremely hard for the users to spot them. Current services that leverage LLMs usually do not provide any means for detecting unreliable generations. Here, we aim to bridge this gap. In particular, we propose a novel fact-checking and hallucination detection pipeline based on token-level uncertainty quantification. Uncertainty scores leverage information encapsulated in the output of a neural network or its layers to detect unreliable predictions, and we show that they can be used to fact-check the atomic claims in the LLM output. Moreover, we present a novel token-level uncertainty quantification method that removes the impact of uncertainty about what claim to generate on the current step and what surface form to use. Our method Claim Conditioned Probability (CCP) measures only the uncertainty of particular claim value expressed by the model. Experiments on the task of biography generation demonstrate strong improvements for CCP compared to the baselines for six different LLMs and three languages. Human evaluation reveals that the fact-checking pipeline based on uncertainty quantification is competitive with a fact-checking tool that leverages external knowledge.

Skillful mobile operation in three-dimensional environments is a primary topic of study in Artificial Intelligence. The past two years have seen a surge of creative work on navigation. This creative output has produced a plethora of sometimes incompatible task definitions and evaluation protocols. To coordinate ongoing and future research in this area, we have convened a working group to study empirical methodology in navigation research. The present document summarizes the consensus recommendations of this working group. We discuss different problem statements and the role of generalization, present evaluation measures, and provide standard scenarios that can be used for benchmarking.

The integration of Artificial Intelligence (AI), especially Large Language Models (LLMs), into the clinical diagnosis process offers significant potential to improve the efficiency and accessibility of medical care. While LLMs have shown some promise in the medical domain, their application in clinical diagnosis remains underexplored, especially in real-world clinical practice, where highly sophisticated, patient-specific decisions need to be made. Current evaluations of LLMs in this field are often narrow in scope, focusing on specific diseases or specialties and employing simplified diagnostic tasks. To bridge this gap, we introduce CliBench, a novel benchmark developed from the MIMIC IV dataset, offering a comprehensive and realistic assessment of LLMs' capabilities in clinical diagnosis. This benchmark not only covers diagnoses from a diverse range of medical cases across various specialties but also incorporates tasks of clinical significance: treatment procedure identification, lab test ordering and medication prescriptions. Supported by structured output ontologies, CliBench enables a precise and multi-granular evaluation, offering an in-depth understanding of LLM's capability on diverse clinical tasks of desired granularity. We conduct a zero-shot evaluation of leading LLMs to assess their proficiency in clinical decision-making. Our preliminary results shed light on the potential and limitations of current LLMs in clinical settings, providing valuable insights for future advancements in LLM-powered healthcare.

Language models are increasingly capable and are being rapidly deployed on a population-level scale. As a result, the safety of these models is increasingly high-stakes. Fortunately, advances in alignment have significantly reduced the likelihood of harmful model outputs. However, when models are queried billions of times in a day, even rare worst-case behaviors will occur. Current safety evaluations focus on capturing the distribution of inputs that yield harmful outputs. These evaluations disregard the probabilistic nature of models and their tail output behavior. To measure this tail risk, we propose a method to efficiently estimate the probability of harmful outputs for any input query. Instead of naive brute-force sampling from the target model, where harmful outputs could be rare, we operationalize importance sampling by creating unsafe versions of the target model. These unsafe versions enable sample-efficient estimation by making harmful outputs more probable. On benchmarks measuring misuse and misalignment, these estimates match brute-force Monte Carlo estimates using 10-20x fewer samples. For example, we can estimate probability of harmful outputs on the order of 10^-4 with just 500 samples. Additionally, we find that these harmfulness estimates can reveal the sensitivity of models to perturbations in model input and predict deployment risks. Our work demonstrates that accurate rare-event estimation is both critical and feasible for safety evaluations. Code is available at https://github.com/rangell/LMTailRisk

With the rapid development of diffusion models, text-to-image(T2I) models have made significant progress, showcasing impressive abilities in prompt following and image generation. Recently launched models such as FLUX.1 and Ideogram2.0, along with others like Dall-E3 and Stable Diffusion 3, have demonstrated exceptional performance across various complex tasks, raising questions about whether T2I models are moving towards general-purpose applicability. Beyond traditional image generation, these models exhibit capabilities across a range of fields, including controllable generation, image editing, video, audio, 3D, and motion generation, as well as computer vision tasks like semantic segmentation and depth estimation. However, current evaluation frameworks are insufficient to comprehensively assess these models' performance across expanding domains. To thoroughly evaluate these models, we developed the IMAGINE-E and tested six prominent models: FLUX.1, Ideogram2.0, Midjourney, Dall-E3, Stable Diffusion 3, and Jimeng. Our evaluation is divided into five key domains: structured output generation, realism, and physical consistency, specific domain generation, challenging scenario generation, and multi-style creation tasks. This comprehensive assessment highlights each model's strengths and limitations, particularly the outstanding performance of FLUX.1 and Ideogram2.0 in structured and specific domain tasks, underscoring the expanding applications and potential of T2I models as foundational AI tools. This study provides valuable insights into the current state and future trajectory of T2I models as they evolve towards general-purpose usability. Evaluation scripts will be released at https://github.com/jylei16/Imagine-e.

State-of-the-art generative models (e.g. StyleGAN3 karras2021alias) often generate photorealistic images based on vectors sampled from their latent space. However, the ability to control the output is limited. Here we present our novel method for latent vector shifting for controlled output image modification utilizing semantic features of the generated images. In our approach we use a pre-trained model of StyleGAN3 that generates images of realistic human faces in relatively high resolution. We complement the generative model with a convolutional neural network classifier, namely ResNet34, trained to classify the generated images with binary facial features from the CelebA dataset. Our latent feature shifter is a neural network model with a task to shift the latent vectors of a generative model into a specified feature direction. We have trained latent feature shifter for multiple facial features, and outperformed our baseline method in the number of generated images with the desired feature. To train our latent feature shifter neural network, we have designed a dataset of pairs of latent vectors with and without a certain feature. Based on the evaluation, we conclude that our latent feature shifter approach was successful in the controlled generation of the StyleGAN3 generator.

Safety evaluations of large language models commonly summarize harmful-output behavior with attack success rate (ASR). Yet the same non-harmful outcome can arise for very different reasons. A model may recover a harmful task and refuse it, fail to recover the task, or respond to something else entirely. Distinguishing these cases becomes especially important under intent-obscuring prompts, where a low ASR does not reveal whether the evaluated task was actually engaged. To make this distinction explicit, we pair ASR with operative understanding rate (UR), which measures whether a response both identifies the evaluated task and treats it as the task to be answered. Across interfaces, this paired view reveals substantial variation hidden by ASR: similar ASR values can correspond to sharply different recovery rates. Controlled English reconstructions show that recovery consistently improves as compressed prompts become more explicit, whereas ASR does not follow the same pattern. A complementary contrast comes from FormalLogic, where high recovery can still coincide with frequent harmful assistance. Together, these results show that non-harmful outcomes are not equally informative about model safety, motivating the joint reporting of intent recovery and ASR in LLM safety evaluation. Code and experiment inputs are available at https://github.com/kevinjiang0121-cyber/IRIS.

Large language model (LLM) developers aim for their models to be honest, helpful, and harmless. However, when faced with malicious requests, models are trained to refuse, sacrificing helpfulness. We show that frontier LLMs can develop a preference for dishonesty as a new strategy, even when other options are available. Affected models respond to harmful requests with outputs that sound harmful but are subtly incorrect or otherwise harmless in practice. This behavior emerges with hard-to-predict variations even within models from the same model family. We find no apparent cause for the propensity to deceive, but we show that more capable models are better at executing this strategy. Strategic dishonesty already has a practical impact on safety evaluations, as we show that dishonest responses fool all output-based monitors used to detect jailbreaks that we test, rendering benchmark scores unreliable. Further, strategic dishonesty can act like a honeypot against malicious users, which noticeably obfuscates prior jailbreak attacks. While output monitors fail, we show that linear probes on internal activations can be used to reliably detect strategic dishonesty. We validate probes on datasets with verifiable outcomes and by using their features as steering vectors. Overall, we consider strategic dishonesty as a concrete example of a broader concern that alignment of LLMs is hard to control, especially when helpfulness and harmlessness conflict.

We present an empirical evaluation of various outputs generated by nine of the most widely-available large language models (LLMs). Our analysis is done with off-the-shelf, readily-available tools. We find a correlation between percentage of memorized text, percentage of unique text, and overall output quality, when measured with respect to output pathologies such as counterfactual and logically-flawed statements, and general failures like not staying on topic. Overall, 80.0% of the outputs evaluated contained memorized data, but outputs containing the most memorized content were also more likely to be considered of high quality. We discuss and evaluate mitigation strategies, showing that, in the models evaluated, the rate of memorized text being output is reduced. We conclude with a discussion on potential implications around what it means to learn, to memorize, and to evaluate quality text.

Traditional automatic speech recognition (ASR) models output lower-cased words without punctuation marks, which reduces readability and necessitates a subsequent text processing model to convert ASR transcripts into a proper format. Simultaneously, the development of end-to-end ASR models capable of predicting punctuation and capitalization presents several challenges, primarily due to limited data availability and shortcomings in the existing evaluation methods, such as inadequate assessment of punctuation prediction. In this paper, we introduce a LibriSpeech-PC benchmark designed to assess the punctuation and capitalization prediction capabilities of end-to-end ASR models. The benchmark includes a LibriSpeech-PC dataset with restored punctuation and capitalization, a novel evaluation metric called Punctuation Error Rate (PER) that focuses on punctuation marks, and initial baseline models. All code, data, and models are publicly available.

This paper presents the Long Context and Form Output (LCFO) benchmark, a novel evaluation framework for assessing gradual summarization and summary expansion capabilities across diverse domains. LCFO consists of long input documents (5k words average length), each of which comes with three summaries of different lengths (20%, 10%, and 5% of the input text), as well as approximately 15 questions and answers (QA) related to the input content. Notably, LCFO also provides alignments between specific QA pairs and corresponding summaries in 7 domains. The primary motivation behind providing summaries of different lengths is to establish a controllable framework for generating long texts from shorter inputs, i.e. summary expansion. To establish an evaluation metric framework for summarization and summary expansion, we provide human evaluation scores for human-generated outputs, as well as results from various state-of-the-art large language models (LLMs). GPT-4o-mini achieves best human scores among automatic systems in both summarization and summary expansion tasks (~ +10% and +20%, respectively). It even surpasses human output quality in the case of short summaries (~ +7%). Overall automatic metrics achieve low correlations with human evaluation scores (~ 0.4) but moderate correlation on specific evaluation aspects such as fluency and attribution (~ 0.6). The LCFO benchmark offers a standardized platform for evaluating summarization and summary expansion performance, as well as corresponding automatic metrics, thereby providing an important evaluation framework to advance generative AI.

In this position paper, we observe that empirical evaluation in Generative AI is at a crisis point since traditional ML evaluation and benchmarking strategies are insufficient to meet the needs of evaluating modern GenAI models and systems. There are many reasons for this, including the fact that these models typically have nearly unbounded input and output spaces, typically do not have a well defined ground truth target, and typically exhibit strong feedback loops and prediction dependence based on context of previous model outputs. On top of these critical issues, we argue that the problems of {\em leakage} and {\em contamination} are in fact the most important and difficult issues to address for GenAI evaluations. Interestingly, the field of AI Competitions has developed effective measures and practices to combat leakage for the purpose of counteracting cheating by bad actors within a competition setting. This makes AI Competitions an especially valuable (but underutilized) resource. Now is time for the field to view AI Competitions as the gold standard for empirical rigor in GenAI evaluation, and to harness and harvest their results with according value.

Large language models (LLMs) have achieved success, but cost and privacy constraints necessitate deploying smaller models locally while offloading complex queries to cloud-based models. Existing router evaluations are unsystematic, overlooking scenario-specific requirements and out-of-distribution robustness. We propose RouterXBench, a principled evaluation framework with three dimensions: router ability, scenario alignment, and cross-domain robustness. Unlike prior work that relies on output probabilities or external embeddings, we utilize internal hidden states that capture model uncertainty before answer generation. We introduce ProbeDirichlet, a lightweight router that aggregates cross-layer hidden states via learnable Dirichlet distributions with probabilistic training. Trained on multi-domain data, it generalizes robustly across in-domain and out-of-distribution scenarios. Our results show ProbeDirichlet achieves 16.68% and 18.86% relative improvements over the best baselines in router ability and high-accuracy scenarios, with consistent performance across model families, model scales, heterogeneous tasks, and agentic workflows.

Large Language Models are increasingly being deployed to extract structured data from unstructured and semi-structured sources: parsing invoices, medical records, and converting PDF documents to database entries. Yet existing benchmarks for structured output generation either focus on schema compliance alone, or evaluate value correctness within a single source domain. We introduce SOB (The Structured Output Benchmark), a multi-source benchmark spanning three source modalities: native text, images, and audio conversations. All models receive a text-normalized representation of their context regardless of source modality; this deliberate design isolates structured-output capability from raw vision or speech-processing quality, ensuring a fair, source-agnostic comparison. Our benchmark comprises 5,000 text evaluation records derived from multi-hop QA drawn from a 25,091-record full corpus, 209 image records from OCR-processed PDFs across seven document types including multi-column layouts, dense tables, scanned historical documents, small-print text, and mathematical typesetting, and 115 audio records from the AMI corpus. Each record pairs a natural-language question with a JSON schema that the model must follow and a ground-truth answer verified against the source context. We evaluate 21 frontier and open-weight models across three source domains and seven metrics. Our results reveal a consistent pattern: models achieve near-perfect schema compliance, yet the best Value Accuracy, measured by exact leaf-value match, reaches only 83.0% on text, 67.2% on images, and 23.7% on audio, where longer context makes extraction substantially harder. We release the dataset, evaluation pipeline, and all related code.

VLM safety is commonly evaluated through input- and output-level classification. Such classification is necessary, but it does not reveal whether a safety state is accessible or controllable inside the model. We argue that multimodal safety evaluation should therefore report a controllability profile alongside behavioral classification, separating representation-level detectability, cross-modal specificity, intervention sensitivity, and benign-preserving selectivity. Using implicit toxicity as a stress case, we instantiate this profile on LlavaGuard and Qwen3.5 with sparse feature decompositions. LlavaGuard admits localized handles with a narrow benign-preserving intervention range and modest downstream safety gains, whereas Qwen3.5 supports strong representation-level readout but no comparable selective-control regime under the tested operators. These results show that internal readout and controllability can diverge. Future multimodal safety benchmarks should therefore report not only behavioral safety metrics, but also whether safety-relevant internal signals can be intervention-tested and controlled within a validated operating range.

While Key-Value (KV) cache succeeds in reducing redundant computations in auto-regressive models, it introduces significant memory overhead, limiting its practical deployment in long-sequence scenarios. Existing KV retrieval methods mitigate this by dynamically retaining only a subset of KV entries on the GPU. However, they still suffer from notable efficiency and accuracy bottlenecks due to per-token retrieval and coarse-grained page-level KV management, especially in long-output reasoning scenarios. With the emergence of large reasoning models, efficiently handling such scenarios has become increasingly important. To address this issue, we present two key observations: (1) critical KVs exhibit strong temporal locality during decoding, and (2) these KVs exhibit distinct distribution patterns across the input prompt and generated output. Building on these observations, we propose LouisKV, an efficient KV cache retrieval framework designed for various long-sequence scenarios. Specifically, LouisKV introduces a semantic-aware retrieval strategy leveraging temporal locality to trigger retrieval only at semantic boundaries, drastically reducing computation and data transfer overhead. LouisKV also designs a decoupled, fine-grained management scheme that tailors differentiated strategies for input and output sequences to create retrieval units that better match the model's attention patterns, enabling precise identification of critical KVs. Furthermore, to boost efficiency, LouisKV incorporates several kernel-level optimizations, including custom Triton and CUDA kernels to accelerate the KV clustering and retrieval. Evaluations show that LouisKV achieves up to 4.7times speedup over state-of-the-art KV retrieval methods while maintaining near-lossless accuracy across diverse long-sequence tasks, including long-input short-output, short-input long-output, and long-input long-output scenarios.

Large language models have revolutionized natural language processing but face significant challenges of high storage and runtime costs, due to the transformer architecture's reliance on self-attention, particularly the large Key-Value (KV) cache for long-sequence inference. Recent efforts to reduce KV cache size by pruning less critical entries based on attention weights remain empirical and lack formal grounding. This paper presents a formal study on identifying critical KV cache entries by analyzing attention output perturbation. Our analysis reveals that, beyond attention weights, the value states within KV entries and pretrained parameter matrices are also crucial. Based on this, we propose a perturbation-constrained selection algorithm that optimizes the worst-case output perturbation to identify critical entries. Evaluations on the Needle-in-a-Haystack test and Longbench benchmark show our algorithm enhances state-of-the-art cache eviction methods. Further empirical analysis confirms that our algorithm achieves lower output perturbations in over 92% attention heads in Llama model, thereby providing a significant improvement over existing methods.

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.

Machine translation evaluation is a very important activity in machine translation development. Automatic evaluation metrics proposed in literature are inadequate as they require one or more human reference translations to compare them with output produced by machine translation. This does not always give accurate results as a text can have several different translations. Human evaluation metrics, on the other hand, lacks inter-annotator agreement and repeatability. In this paper we have proposed a new human evaluation metric which addresses these issues. Moreover this metric also provides solid grounds for making sound assumptions on the quality of the text produced by a machine translation.

Real-world image super-resolution (RWSR) is a long-standing problem as low-quality (LQ) images often have complex and unidentified degradations. Existing methods such as Generative Adversarial Networks (GANs) or continuous diffusion models present their own issues including GANs being difficult to train while continuous diffusion models requiring numerous inference steps. In this paper, we propose an Iterative Token Evaluation and Refinement (ITER) framework for RWSR, which utilizes a discrete diffusion model operating in the discrete token representation space, i.e., indexes of features extracted from a VQGAN codebook pre-trained with high-quality (HQ) images. We show that ITER is easier to train than GANs and more efficient than continuous diffusion models. Specifically, we divide RWSR into two sub-tasks, i.e., distortion removal and texture generation. Distortion removal involves simple HQ token prediction with LQ images, while texture generation uses a discrete diffusion model to iteratively refine the distortion removal output with a token refinement network. In particular, we propose to include a token evaluation network in the discrete diffusion process. It learns to evaluate which tokens are good restorations and helps to improve the iterative refinement results. Moreover, the evaluation network can first check status of the distortion removal output and then adaptively select total refinement steps needed, thereby maintaining a good balance between distortion removal and texture generation. Extensive experimental results show that ITER is easy to train and performs well within just 8 iterative steps. Our codes will be available publicly.

Current evaluations of large language models (LLMs) often overlook non-determinism, typically focusing on a single output per example. This limits our understanding of LLM performance variability in real-world applications. Our study addresses this issue by exploring key questions about the performance differences between greedy decoding and sampling, identifying benchmarks' consistency regarding non-determinism, and examining unique model behaviors. Through extensive experiments, we observe that greedy decoding generally outperforms sampling methods for most evaluated tasks. We also observe consistent performance across different LLM sizes and alignment methods, noting that alignment can reduce sampling variance. Moreover, our best-of-N sampling approach demonstrates that smaller LLMs can match or surpass larger models such as GPT-4-Turbo, highlighting the untapped potential of smaller LLMs. This research shows the importance of considering non-determinism in LLM evaluations and provides insights for future LLM development and evaluation.

Retrieval Augmented Generation (RAG) is widely used to enable Large Language Models (LLMs) perform Question Answering (QA) tasks in various domains. However, RAG based on open-source LLM for specialized domains has challenges of evaluating generated responses. A popular framework in the literature is the RAG Assessment (RAGAS), a publicly available library which uses LLMs for evaluation. One disadvantage of RAGAS is the lack of details of derivation of numerical value of the evaluation metrics. One of the outcomes of this work is a modified version of this package for few metrics (faithfulness, context relevance, answer relevance, answer correctness, answer similarity and factual correctness) through which we provide the intermediate outputs of the prompts by using any LLMs. Next, we analyse the expert evaluations of the output of the modified RAGAS package and observe the challenges of using it in the telecom domain. We also study the effect of the metrics under correct vs. wrong retrieval and observe that few of the metrics have higher values for correct retrieval. We also study for differences in metrics between base embeddings and those domain adapted via pre-training and fine-tuning. Finally, we comment on the suitability and challenges of using these metrics for in-the-wild telecom QA task.

With the widespread adoption of large language models (LLMs) in code generation tasks, geospatial code generation has emerged as a critical frontier in the integration of artificial intelligence and geoscientific analysis. This trend underscores the urgent need for systematic evaluation methodologies to assess LLMs generation capabilities in geospatial contexts. In particular, geospatial computation and visualization tasks in JavaScript environments rely heavily on orchestrating diverse frontend libraries and ecosystems, placing elevated demands on a model's semantic understanding and code synthesis abilities. To address this challenge, we propose GeoJSEval--the first multimodal, function-level automatic evaluation framework for LLMs in JavaScript-based geospatial code generation. GeoJSEval comprises three core components: a standardized test suite (GeoJSEval-Bench), a code submission engine, and an evaluation module. It includes 432 function-level tasks and 2,071 structured test cases spanning five widely used JavaScript geospatial libraries and 25 mainstream geospatial data types. GeoJSEval enables multidimensional quantitative evaluation across metrics such as accuracy, output stability, execution efficiency, resource consumption, and error type distribution, and integrates boundary testing mechanisms to enhance robustness and coverage. We conduct a comprehensive evaluation of 18 state-of-the-art LLMs using GeoJSEval, revealing significant performance disparities and bottlenecks in spatial semantic understanding, code reliability, and function invocation accuracy. GeoJSEval provides a foundational methodology, evaluation resource, and practical toolkit for the standardized assessment and optimization of geospatial code generation models, with strong extensibility and applicability in real-world scenarios.

Recently, Large Language Models (LLMs) have garnered increasing attention in the field of natural language processing, revolutionizing numerous downstream tasks with powerful reasoning and generation abilities. For example, In-Context Learning (ICL) introduces a fine-tuning-free paradigm, allowing out-of-the-box LLMs to execute downstream tasks by analogy learning without any fine-tuning. Besides, in a fine-tuning-dependent paradigm where substantial training data exists, Parameter-Efficient Fine-Tuning (PEFT), as the cost-effective methods, enable LLMs to achieve excellent performance comparable to full fine-tuning. However, these fascinating techniques employed by LLMs have not been fully exploited in the ABSA field. Previous works probe LLMs in ABSA by merely using randomly selected input-output pairs as demonstrations in ICL, resulting in an incomplete and superficial evaluation. In this paper, we shed light on a comprehensive evaluation of LLMs in the ABSA field, involving 13 datasets, 8 ABSA subtasks, and 6 LLMs. Specifically, we design a unified task formulation to unify ``multiple LLMs for multiple ABSA subtasks in multiple paradigms.'' For the fine-tuning-dependent paradigm, we efficiently fine-tune LLMs using instruction-based multi-task learning. For the fine-tuning-free paradigm, we propose 3 demonstration selection strategies to stimulate the few-shot abilities of LLMs. Our extensive experiments demonstrate that LLMs achieve a new state-of-the-art performance compared to fine-tuned Small Language Models (SLMs) in the fine-tuning-dependent paradigm. More importantly, in the fine-tuning-free paradigm where SLMs are ineffective, LLMs with ICL still showcase impressive potential and even compete with fine-tuned SLMs on some ABSA subtasks.

While Large Audio-Language Models (LALMs) have been shown to exhibit degraded instruction-following capabilities, their ability to infer task patterns from in-context examples under audio conditioning remains unstudied. To address this gap, we present ALICE, a three-stage framework that progressively reduces textual guidance to systematically evaluate LALMs' in-context learning ability under audio conditioning. Evaluating six LALMs across four audio understanding tasks under two output constraint categories, we uncover a consistent asymmetry across all stages and LALMs: in-context demonstrations reliably improve format compliance but fail to improve, and often degrade, the core task performance. This suggests that LALMs can glean surface-level formatting patterns from demonstrations but may struggle to leverage cross-modal semantic grounding to reliably infer task objectives from audio-conditioned examples, highlighting potential limitations in current cross-modal integration.

In this paper, we extend FLORES+ with Portuguese-source evaluation sets for three Mozambican Bantu varieties: Xichangana, Mozambican Nyanja, and Sena. We compare Xichangana with the existing Tsonga reference and Mozambican Nyanja with Chichewa, and evaluate NLLB-200, Google Translate, GPT, and a variant-aware NLLB model. Holding system output fixed reveals substantial reference sensitivity. On devtest, changing only the reference from Tsonga to Xichangana reduces spBLEU by 13.10 points for NLLB-200 and 15.30 for Google. On matched Nyanja subsets, replacing Chichewa with Mozambican Nyanja produces smaller but consistent reductions of 3.03 and 6.10 spBLEU, respectively. Variant-aware fine-tuning reverses this pattern on the intended targets: relative to NLLB-200, it improves Xichangana by 7.04 spBLEU and Mozambican Nyanja by 5.33 on devtest, while losing performance on the sibling references. GPT is competitive on Tsonga and Chichewa but substantially weaker on the Mozambican varieties. For Sena, the finetuned model reaches 12.64 spBLEU and 36.21 chrF++ on devtest. These findings motivate variety-aware language identifiers, references, and reporting for cross-border languages or language dialects/variants. The data is publicly available on Hugging Face at https://huggingface.co/datasets/MOZNLP/FLORES_MOZ

System Instructions in Large Language Models (LLMs) are commonly used to enforce safety policies, define agent behavior, and protect sensitive operational context in agentic AI applications. These instructions may contain sensitive information such as API credentials, internal policies, and privileged workflow definitions, making system instruction leakage a critical security risk highlighted in the OWASP Top 10 for LLM Applications. Without incurring the overhead costs of reasoning models, many LLM applications rely on refusal-based instructions that block direct requests for system instructions, implicitly assuming that prohibited information can only be extracted through explicit queries. We introduce an automated evaluation framework that tests whether system instructions remain confidential when extraction requests are re-framed as encoding or structured output tasks. Across seven common models and 46 verified system instructions, we observe high attack success rates (>0.7) for structured serialization where models refuse direct extraction requests but disclose protected content in the requested serialization formats. We further demonstrate a mitigation strategy based on one-shot instruction reshaping using a Chain-of-Thought reasoning model, indicating that even subtle changes in wording and structure of system instructions can significantly reduce attack success rate without requiring model retraining.

Extending Large Language Models (LLMs) to advanced applications requires reliable structured output generation. Existing methods which often rely on rigid JSON schemas, can lead to unreliable outputs, diminished reasoning capabilities, and increased computational overhead, limiting LLMs' adaptability for complex tasks. We introduce Meaning Typed Prompting (MTP), a technique for efficient structured output generation that integrates types, meanings, and abstractions, such as variables and classes, into the prompting process. By utilizing expressive type definitions, MTP enhances output clarity and reduces dependence on complex abstractions, simplifying development, and improving implementation efficiency. This enables LLMs to understand relationships and generate structured data more effectively. Empirical evaluations on multiple benchmarks demonstrate that MTP outperforms existing frameworks in accuracy, reliability, consistency, and token efficiency. We present Semantix, a framework that implements MTP, providing practical insights into its application.

With the release of the o1 model by OpenAI, reasoning models adopting slow thinking strategies have gradually emerged. As the responses generated by such models often include complex reasoning, intermediate steps, and self-reflection, existing evaluation methods are often inadequate. They struggle to determine whether the LLM output is truly equivalent to the reference answer, and also have difficulty identifying and extracting the final answer from long, complex responses. To address this issue, we propose xVerify, an efficient answer verifier for reasoning model evaluations. xVerify demonstrates strong capability in equivalence judgment, enabling it to effectively determine whether the answers produced by reasoning models are equivalent to reference answers across various types of objective questions. To train and evaluate xVerify, we construct the VAR dataset by collecting question-answer pairs generated by multiple LLMs across various datasets, leveraging multiple reasoning models and challenging evaluation sets designed specifically for reasoning model assessment. A multi-round annotation process is employed to ensure label accuracy. Based on the VAR dataset, we train multiple xVerify models of different scales. In evaluation experiments conducted on both the test set and generalization set, all xVerify models achieve overall F1 scores and accuracy exceeding 95\%. Notably, the smallest variant, xVerify-0.5B-I, outperforms all evaluation methods except GPT-4o, while xVerify-3B-Ib surpasses GPT-4o in overall performance. These results validate the effectiveness and generalizability of xVerify.

Evaluating the quality and variability of text generated by Large Language Models (LLMs) poses a significant, yet unresolved research challenge. Traditional evaluation methods, such as ROUGE and BERTScore, which measure token similarity, often fail to capture the holistic semantic equivalence. This results in a low correlation with human judgments and intuition, which is especially problematic in high-stakes applications like healthcare and finance where reliability, safety, and robust decision-making are highly critical. This work proposes DCR, an automated framework for evaluating and improving the consistency of LLM-generated texts using a divide-conquer-reasoning approach. Unlike existing LLM-based evaluators that operate at the paragraph level, our method employs a divide-and-conquer evaluator (DCE) that breaks down the paragraph-to-paragraph comparison between two generated responses into individual sentence-to-paragraph comparisons, each evaluated based on predefined criteria. To facilitate this approach, we introduce an automatic metric converter (AMC) that translates the output from DCE into an interpretable numeric score. Beyond the consistency evaluation, we further present a reason-assisted improver (RAI) that leverages the analytical reasons with explanations identified by DCE to generate new responses aimed at reducing these inconsistencies. Through comprehensive and systematic empirical analysis, we show that our approach outperforms state-of-the-art methods by a large margin (e.g., +19.3% and +24.3% on the SummEval dataset) in evaluating the consistency of LLM generation across multiple benchmarks in semantic, factual, and summarization consistency tasks. Our approach also substantially reduces nearly 90% of output inconsistencies, showing promise for effective hallucination mitigation.

We present MEGA-Bench, an evaluation suite that scales multimodal evaluation to over 500 real-world tasks, to address the highly heterogeneous daily use cases of end users. Our objective is to optimize for a set of high-quality data samples that cover a highly diverse and rich set of multimodal tasks, while enabling cost-effective and accurate model evaluation. In particular, we collected 505 realistic tasks encompassing over 8,000 samples from 16 expert annotators to extensively cover the multimodal task space. Instead of unifying these problems into standard multi-choice questions (like MMMU, MMBench, and MMT-Bench), we embrace a wide range of output formats like numbers, phrases, code, \LaTeX, coordinates, JSON, free-form, etc. To accommodate these formats, we developed over 40 metrics to evaluate these tasks. Unlike existing benchmarks, MEGA-Bench offers a fine-grained capability report across multiple dimensions (e.g., application, input type, output format, skill), allowing users to interact with and visualize model capabilities in depth. We evaluate a wide variety of frontier vision-language models on MEGA-Bench to understand their capabilities across these dimensions.

Reward models play an essential role in training vision-language models (VLMs) by assessing output quality to enable aligning with human preferences. Despite their importance, the research community lacks comprehensive open benchmarks for evaluating multimodal reward models in VLMs. To address this gap, we introduce Multimodal RewardBench, an expert-annotated benchmark covering six domains: general correctness, preference, knowledge, reasoning, safety, and visual question-answering. Our dataset comprises 5,211 annotated (prompt, chosen response, rejected response) triplets collected from various VLMs. In evaluating a range of VLM judges, we find that even the top-performing models, Gemini 1.5 Pro and Claude 3.5 Sonnet, achieve only 72% overall accuracy. Notably, most models struggle in the reasoning and safety domains. These findings suggest that Multimodal RewardBench offers a challenging testbed for advancing reward model development across multiple domains. We release the benchmark at https://github.com/facebookresearch/multimodal_rewardbench.

Recent advancements in text-to-image (T2I) generation models have transformed the field. However, challenges persist in generating images that reflect demanding textual descriptions, especially for fine-grained details and unusual relationships. Existing evaluation metrics focus on text-image alignment but overlook the realism of the generated image, which can be crucial for downstream applications like data augmentation in machine learning. To address this gap, we propose REAL, an automatic evaluation framework that assesses realism of T2I outputs along three dimensions: fine-grained visual attributes, unusual visual relationships, and visual styles. REAL achieves a Spearman's rho score of up to 0.62 in alignment with human judgement and demonstrates utility in ranking and filtering augmented data for tasks like image captioning, classification, and visual relationship detection. Empirical results show that high-scoring images evaluated by our metrics improve F1 scores of image classification by up to 11.3%, while low-scoring ones degrade that by up to 4.95%. We benchmark four major T2I models across the realism dimensions, providing insights for future improvements in T2I output realism.

In this paper we present AnswerCarefully, a dataset for promoting the safety and appropriateness of Japanese LLM outputs. The dataset consists of 1,800 pairs of questions and reference answers, where the questions require special attention in answering. It covers a wide range of risk categories established in prior English-language datasets, but the data samples are original in that they are manually created to reflect the socio-cultural context of LLM usage in Japan. We show that using this dataset for instruction to fine-tune a Japanese LLM led to improved output safety without compromising the utility of general responses. We also report the results of a safety evaluation of 12 Japanese LLMs using this dataset as a benchmark. Finally, we describe the latest update on the dataset which provides English translations and annotations of the questions, aimed at facilitating the derivation of similar datasets in different languages and regions.

Developing large language models is expensive and involves making decisions with small experiments, typically by evaluating on large, multi-task evaluation suites. In this work, we analyze specific properties which make a benchmark more reliable for such decisions, and interventions to design higher-quality evaluation benchmarks. We introduce two key metrics that show differences in current benchmarks: signal, a benchmark's ability to separate better models from worse models, and noise, a benchmark's sensitivity to random variability between training steps. We demonstrate that benchmarks with a better signal-to-noise ratio are more reliable when making decisions at small scale, and those with less noise have lower scaling law prediction error. These results suggest that improving signal or noise will lead to more useful benchmarks, so we introduce three interventions designed to directly affect signal or noise. For example, we propose that switching to a metric that has better signal and noise (e.g., perplexity rather than accuracy) leads to better reliability and improved scaling law error. We also find that filtering noisy subtasks, to improve an aggregate signal-to-noise ratio, leads to more reliable multi-task evaluations. We also find that averaging the output of a model's intermediate checkpoints to reduce noise leads to consistent improvements. We conclude by recommending that those creating new benchmarks, or selecting which existing benchmarks to use, aim for high signal and low noise. We use 30 benchmarks for these experiments, and 375 open-weight language models from 60M to 32B parameters, resulting in a new, publicly available dataset of 900K evaluation benchmark results, totaling 200M instances.

Evaluations of audio-language models (ALMs) -- multimodal models that take interleaved audio and text as input and output text -- are hindered by the lack of standardized benchmarks; most benchmarks measure only one or two capabilities and omit evaluative aspects such as fairness or safety. Furthermore, comparison across models is difficult as separate evaluations test a limited number of models and use different prompting methods and inference parameters. To address these shortfalls, we introduce AHELM, a benchmark that aggregates various datasets -- including 2 new synthetic audio-text datasets called PARADE, which evaluates the ALMs on avoiding stereotypes, and CoRe-Bench, which measures reasoning over conversational audio through inferential multi-turn question answering -- to holistically measure the performance of ALMs across 10 aspects we have identified as important to the development and usage of ALMs: audio perception, knowledge, reasoning, emotion detection, bias, fairness, multilinguality, robustness, toxicity, and safety. We also standardize the prompts, inference parameters, and evaluation metrics to ensure equitable comparisons across models. We test 14 open-weight and closed-API ALMs from 3 developers and 3 additional simple baseline systems each consisting of an automatic speech recognizer and a language model. Our results show that while Gemini 2.5 Pro ranks top in 5 out of 10 aspects, it exhibits group unfairness (p=0.01) on ASR tasks whereas most of the other models do not. We also find that the baseline systems perform reasonably well on AHELM, with one ranking 5th overall despite having only speech-to-text capabilities. For transparency, all raw prompts, model generations, and outputs are available on our website at https://crfm.stanford.edu/helm/audio/v1.0.0. AHELM is intended to be a living benchmark and new datasets and models will be added over time.

Providing quality scores along with Machine Translation (MT) output, so-called reference-free Quality Estimation (QE), is crucial to inform users about the reliability of the translation. We propose a model-specific, unsupervised QE approach, termed kNN-QE, that extracts information from the MT model's training data using k-nearest neighbors. Measuring the performance of model-specific QE is not straightforward, since they provide quality scores on their own MT output, thus cannot be evaluated using benchmark QE test sets containing human quality scores on premade MT output. Therefore, we propose an automatic evaluation method that uses quality scores from reference-based metrics as gold standard instead of human-generated ones. We are the first to conduct detailed analyses and conclude that this automatic method is sufficient, and the reference-based MetricX-23 is best for the task.

Structured outputs are essential for large language models (LLMs) in critical applications like agents and information extraction. Despite their capabilities, LLMs often generate outputs that deviate from predefined schemas, significantly hampering reliable application development. We present SLOT (Structured LLM Output Transformer), a model-agnostic approach that transforms unstructured LLM outputs into precise structured formats. While existing solutions predominantly rely on constrained decoding techniques or are tightly coupled with specific models, SLOT employs a fine-tuned lightweight language model as a post-processing layer, achieving flexibility across various LLMs and schema specifications. We introduce a systematic pipeline for data curation and synthesis alongside a formal evaluation methodology that quantifies both schema accuracy and content fidelity. Our results demonstrate that fine-tuned Mistral-7B model with constrained decoding achieves near perfect schema accuracy (99.5%) and content similarity (94.0%), outperforming Claude-3.5-Sonnet by substantial margins (+25 and +20 percentage points, respectively). Notably, even compact models like Llama-3.2-1B can match or exceed the structured output capabilities of much larger proprietary models when equipped with SLOT, enabling reliable structured generation in resource-constrained environments.

Large language models are increasingly evaluated by other models, raising a natural question: can a model predict how a judge will score its own output? We find that the ability is largely present before any targeted training: prompted few-shot, a base model already predicts an external judge's multi-attribute quality scores on open-ended responses well above chance across three benchmarks. We introduce Self-Evaluation Elicitation (SEE), a method that surfaces this latent ability through a short cycle comprising a calibration-coupled reinforcement learning phase that improves the answer and predicts the judge, followed by a masked distillation phase that sharpens the prediction while leaving the answer untouched. From 160 unique examples, roughly 31x fewer than a reinforcement learning baseline, SEE improves held-out calibration across three benchmarks while preserving answer quality. The elicited self-evaluation is sharply localized within the model's own token distribution and stable across judges it was never trained against, indicating a transferable notion of quality rather than a single judge's preference. These results reframe judge-aligned self-evaluation as a problem of elicitation rather than acquisition.

Large language models are increasingly used as chemistry assistants, yet most chemistry benchmarks still score only final answers. This masks a critical failure mode: a model may output the correct molecule, product, or option while its reasoning violates chemical logic. Existing process-level evaluators are hard to scale because LLM judges and human step-level process annotation are costly, inconsistent, and vulnerable to hallucination. We introduce ChemCoTBench-V2, a rule-verifiable diagnostic benchmark for low-cost, auditable evaluation of structured, verifier-addressable chemical reasoning traces. It spans molecular understanding, molecule editing, molecular optimization, and reaction prediction, with 5,620 evaluation samples across 18 reporting tasks. Models must expose key intermediate steps in expert-designed templates, and those steps are checked with deterministic chemistry rules and, for closed-answer tasks, reference traces rather than another LLM judge. Open-ended molecular optimization is evaluated with oracle-verifiable state constraints rather than strict trace matching. The benchmark reports three separate signals: final-answer correctness, template adherence, and step-wise verifier correctness over expert-refined intermediate commitments. Experiments on frontier models reveal a persistent gap between final-answer success and structured-reasoning-state consistency: models often follow the requested format while failing chemical-step checks, or answer correctly with weak supporting reasoning. ChemCoTBench-V2 enables fine-grained model comparison and identifies the concrete step at which the trace first violates the verifier.

Evaluating generated videos remains challenging because existing benchmarks rely on fixed evaluation content, cover only a subset of generation and editing settings, and provide limited evidence for their scores. We introduce VideoArgus, a unified rubric-grounded framework covering five video generation and editing settings. For each input instance, VideoArgus generates an output-blind, sample-specific rubric once and reuses it to evaluate all corresponding candidate videos. The rubric defines concrete criteria, scoring rules, failure modes, and evidence plans, which guide criterion-specific VLM QA and visual tools to produce evidence-grounded criterion scores, rationales, and a diagnostic report. We further construct VideoArgus-Bench, containing 1,026 curated input instances built from 653 high-quality images and 416 high-quality videos, with all benchmark rubrics pre-generated, frozen, and released. On a separate 1,260-video human-alignment set, VideoArgus achieves higher within-input Spearman and Kendall correlations with human judgments than the corresponding benchmark-specific evaluators across all five tasks. Model rankings also remain largely consistent across different rubric-generation and evaluation-VLM backbones. All code and data are released. Visit our project page: https://zzzmyyzeng.github.io/VideoArgus

Self-evolving agents accumulate reusable skills by appending successful procedures and failure fixes. Over time, the same requirement is often restated in several branches, examples, and warnings, while common action sequences are copied rather than reused. The resulting skill becomes expensive to inject and difficult to maintain. Generic prompt compression is ill-suited to this setting because a skill is not a flat passage: its name and description define when it applies, its workflow controls execution, its tool and output contracts constrain validity, and rare exceptions may remain essential even when no sampled task activates them. Evaluation-guided compression can test these behaviors, but it introduces rollouts, cost, and dependence on the compression-time evaluation set. We present SkillZip, an evaluation-free method that compresses a skill by finding its shortest faithful structural explanation. The intuition is explain once, reference many: state a repeated rule once at the scope where it applies, factor a repeated action sequence into a shared procedure, and keep only the differences as explicit exceptions. We formalize this intuition as a typed minimum description-length objective over a skill contract and a residual, subject to a hard coverage constraint for every extracted trigger, workflow edge, tool requirement, obligation, and output field. The formulation provides simple sharing thresholds, preserves unique rare rules by construction, and supports efficient local updates. SkillZip has a one-shot mode with one structured extraction call and deterministic optimization, and a continual Zip-on-Write mode that integrates each self-evolution patch without replaying tasks or reparsing the full history. Through comprehensive experimental evaluations, we demonstrate the effectiveness and superiority of SkillZip in compression performance, generalizability, and cost overhead.

The Abstraction and Reasoning Corpus (ARC) has become a prominent benchmark for evaluating general abstract reasoning and fluid intelligence in AI models. Yet standard ARC evaluation considers only a single capability: producing the correct output grid for a test input. We argue that this narrow format fails to evaluate the diversity of abilities that genuine abstract skill acquisition should enable. We introduce PotARCin, a benchmark that extends ARC by assessing understanding of a task's underlying abstract rule across five dimensions: Definition, Classification, Constrained Generation, Editing, and Inversion. PotARCin employs programmatic methods to generate new task instances and transform given inputs for a given ARC task, enabling dynamic generative sampling beyond fixed input-output pairs. Across five state-of-the-art models evaluated on the ARC-AGI-1 training set, we observe a 25-52 percentage-point performance gap between standard ARC evaluation and evaluation on PotARCin, and find that multi-dimensional evaluation reorders models that standard accuracy ranks alike. We further investigate effects of generative sampling, difficulty of corruption types, and questions of self-consistency, showing that models frequently contradict their own formalized rule even where they have stated it correctly. We also introduce P-ARC, a held-out hand-crafted test set, on which models achieve 1-8% accuracy across all five dimensions, underscoring the importance of more holistic evaluations of abstract reasoning capabilities.

Multilingual language models often produce inconsistent answers to semantically equivalent questions across languages, motivating methods to improve cross-lingual consistency (CLC). However, existing methods are typically evaluated using different models, tasks, and protocols, leaving their relative strengths unclear. In this work, we present a unified evaluation of representative CLC-enhancement methods for question answering, spanning inference-time interventions and post-training approaches across three model families and three closed-form benchmarks. The results show that post-training methods are generally more reliable, with direct distribution alignment consistently improving CLC across all model-dataset combinations, while other methods are more sensitive to answer format and the breadth of language coverage. Notably, cross-domain transfer is limited unless source and target tasks share similar output formats. We further investigate whether CLC enhancement hurts models' ability to respond differently *when needed*, that is, when asked culture-dependent questions. Across two benchmarks of culturally diverse question answering, we find no systematic degradation in controlled closed-form evaluation, whereas open-ended generation reveals occasional accuracy reductions, particularly for non-English responses. Our work highlights the need to evaluate CLC enhancement for both cross-domain robustness and culturally appropriate variation, informing future work in post-training and benchmark development.

To mitigate the potential misuse of large language models (LLMs), recent research has developed watermarking algorithms, which restrict the generation process to leave an invisible trace for watermark detection. Due to the two-stage nature of the task, most studies evaluate the generation and detection separately, thereby presenting a challenge in unbiased, thorough, and applicable evaluations. In this paper, we introduce WaterBench, the first comprehensive benchmark for LLM watermarks, in which we design three crucial factors: (1) For benchmarking procedure, to ensure an apples-to-apples comparison, we first adjust each watermarking method's hyper-parameter to reach the same watermarking strength, then jointly evaluate their generation and detection performance. (2) For task selection, we diversify the input and output length to form a five-category taxonomy, covering 9 tasks. (3) For evaluation metric, we adopt the GPT4-Judge for automatically evaluating the decline of instruction-following abilities after watermarking. We evaluate 4 open-source watermarks on 2 LLMs under 2 watermarking strengths and observe the common struggles for current methods on maintaining the generation quality. The code and data are available at https://github.com/THU-KEG/WaterBench.

Retrieval-Augmented Generation (RAG) has recently gained traction in natural language processing. Numerous studies and real-world applications are leveraging its ability to enhance generative models through external information retrieval. Evaluating these RAG systems, however, poses unique challenges due to their hybrid structure and reliance on dynamic knowledge sources. To better understand these challenges, we conduct A Unified Evaluation Process of RAG (Auepora) and aim to provide a comprehensive overview of the evaluation and benchmarks of RAG systems. Specifically, we examine and compare several quantifiable metrics of the Retrieval and Generation components, such as relevance, accuracy, and faithfulness, within the current RAG benchmarks, encompassing the possible output and ground truth pairs. We then analyze the various datasets and metrics, discuss the limitations of current benchmarks, and suggest potential directions to advance the field of RAG benchmarks.

As machine learning models evolve, maintaining transparency demands more human-centric explainable AI techniques. Counterfactual explanations, with roots in human reasoning, identify the minimal input changes needed to obtain a given output and, hence, are crucial for supporting decision-making. Despite their importance, the evaluation of these explanations often lacks grounding in user studies and remains fragmented, with existing metrics not fully capturing human perspectives. To address this challenge, we developed a diverse set of 30 counterfactual scenarios and collected ratings across 8 evaluation metrics from 206 respondents. Subsequently, we fine-tuned different Large Language Models (LLMs) to predict average or individual human judgment across these metrics. Our methodology allowed LLMs to achieve an accuracy of up to 63% in zero-shot evaluations and 85% (over a 3-classes prediction) with fine-tuning across all metrics. The fine-tuned models predicting human ratings offer better comparability and scalability in evaluating different counterfactual explanation frameworks.

The increasing application of multi-modal large language models (MLLMs) across various sectors have spotlighted the essence of their output reliability and accuracy, particularly their ability to produce content grounded in factual information (e.g. common and domain-specific knowledge). In this work, we introduce SimpleVQA, the first comprehensive multi-modal benchmark to evaluate the factuality ability of MLLMs to answer natural language short questions. SimpleVQA is characterized by six key features: it covers multiple tasks and multiple scenarios, ensures high quality and challenging queries, maintains static and timeless reference answers, and is straightforward to evaluate. Our approach involves categorizing visual question-answering items into 9 different tasks around objective events or common knowledge and situating these within 9 topics. Rigorous quality control processes are implemented to guarantee high-quality, concise, and clear answers, facilitating evaluation with minimal variance via an LLM-as-a-judge scoring system. Using SimpleVQA, we perform a comprehensive assessment of leading 18 MLLMs and 8 text-only LLMs, delving into their image comprehension and text generation abilities by identifying and analyzing error cases.

Neural machine translation (NMT) has progressed rapidly in the past few years, promising improvements and quality translations for different languages. Evaluation of this task is crucial to determine the quality of the translation. Overall, insufficient emphasis is placed on the actual sense of the translation in traditional methods. We propose a bidirectional semantic-based evaluation method designed to assess the sense distance of the translation from the source text. This approach employs the comprehensive multilingual encyclopedic dictionary BabelNet. Through the calculation of the semantic distance between the source and its back translation of the output, our method introduces a quantifiable approach that empowers sentence comparison on the same linguistic level. Factual analysis shows a strong correlation between the average evaluation scores generated by our method and the human assessments across various machine translation systems for English-German language pair. Finally, our method proposes a new multilingual approach to rank MT systems without the need for parallel corpora.

Large Language Models (LLMs) often generate substantively relevant content but fail to adhere to formal constraints, leading to outputs that are conceptually correct but procedurally flawed. Traditional prompt refinement approaches focus on rephrasing the description of the primary task an LLM has to perform, neglecting the granular constraints that function as acceptance criteria for its response. We propose a novel multi-agentic workflow that decouples optimization of the primary task description from its constraints, using quantitative scores as feedback to iteratively rewrite and improve them. Our evaluation demonstrates this method produces revised prompts that yield significantly higher compliance scores from models like Llama 3.1 8B and Mixtral-8x 7B.

Recent advances in multimodal large language models enable new possibilities for image-based decision support. However, their reliability and operational trade-offs in neuroimaging remain insufficiently understood. We present a comprehensive benchmarking study of vision-enabled large language models for 2D neuroimaging using curated MRI and CT datasets covering multiple sclerosis, stroke, brain tumors, other abnormalities, and normal controls. Models are required to generate multiple outputs simultaneously, including diagnosis, diagnosis subtype, imaging modality, specialized sequence, and anatomical plane. Performance is evaluated across four directions: discriminative classification with abstention, calibration, structured-output validity, and computational efficiency. A multi-phase framework ensures fair comparison while controlling for selection bias. Across twenty frontier multimodal models, the results show that technical imaging attributes such as modality and plane are nearly solved, whereas diagnostic reasoning, especially subtype prediction, remains challenging. Tumor classification emerges as the most reliable task, stroke is moderately solvable, while multiple sclerosis and rare abnormalities remain difficult. Few-shot prompting improves performance for several models but increases token usage, latency, and cost. Gemini-2.5-Pro and GPT-5-Chat achieve the strongest overall diagnostic performance, while Gemini-2.5-Flash offers the best efficiency-performance trade-off. Among open-weight architectures, MedGemma-1.5-4B demonstrates the most promising results, as under few-shot prompting, it approaches the zero-shot performance of several proprietary models, while maintaining perfect structured output. These findings provide practical insights into performance, reliability, and efficiency trade-offs, supporting standardized evaluation of multimodal LLMs in neuroimaging.

Large language models (LLMs) increasingly rely on explicit reasoning to solve coding tasks, yet evaluating the quality of this reasoning remains challenging. Existing reasoning evaluators are not designed for coding, and current benchmarks focus primarily on code generation, leaving other coding tasks largely unexplored. We introduce CodeRQ-Bench, the first benchmark for evaluating LLM reasoning quality across three coding task categories: generation, summarization, and classification. Using this benchmark, we analyze 1,069 mismatch cases from existing evaluators, identify five recurring limitations, and derive four design insights for reasoning evaluation in coding tasks. Guided by these insights, we propose VERA, a two-stage evaluator that combines evidence-grounded verification with ambiguity-aware score correction. Experiments on CodeRQ-Bench show that VERA consistently outperforms strong baselines across four datasets, improving AUCROC by up to 0.26 and AUPRC by up to 0.21. We release CodeRQ-Bench at https://github.com/MrLYG/CodeRQ-Bench, supporting future investigations.

Role-playing agents (RPAs) have become one of the most important consumer applications of large language models. Users engage in multi-turn conversations with RPAs for experiences such as emotional comfort, making reliable evaluation essential for measuring capability, comparing systems, and guiding further improvement. Existing benchmarks, however, typically require an RPA to continue a fixed dialogue history and then evaluate the continuation using a fixed rubric detached from the user. We identify and empirically demonstrate two limitations of this design. First, an RPA's output is shaped by the preceding dialogue history, preventing a scientifically grounded assessment of its role-playing ability in real multi-turn settings. Second, user experience varies substantially across individuals, and conventional fixed rubrics need not align with user satisfaction. We therefore introduce PALATE (Person-Aligned LLM-Simulated-User Assessment with Tailored Evaluation), a scalable RPA benchmark built on user simulators. PALATE is accompanied by a pool of 300 character profiles. Its main evaluation trains five per-user simulators and lets them engage candidate RPAs in free-form, multi-turn conversations over a pre-frozen panel of character profiles. Alongside a general quality rubric, we construct personalized rubrics to measure user satisfaction; on held-out annotated data, the personalized rubrics show higher agreement with human judgments than the general rubric. In the main evaluation of 16 candidates, PALATE separately characterizes generic turn quality, long-horizon session capability, and per-user experience on multi-turn trajectories co-constructed by each candidate. It thereby produces interpretable evaluations of specific user-RPA pairs rather than compressing systems into a single user-independent ranking.

As AI systems are increasingly used to draft radiology reports, reliably evaluating their clinical quality remains a critical challenge. Large language model (LLM)-based metrics are now the best-correlated with radiologist judgment, yet they output a single opaque score that neither a clinician nor a model builder can easily interpret or audit. We introduce RadMatch, a multi-stage, LLM-based metric that decomposes report comparison into a structured finding-level matching with significance-aware scoring and error characterization across seven clinical attribute dimensions (status, location, severity, morphology, certainty, longitudinal comparison, and measurement). The main score is the actionable-error count, both interpretable and auditable. Candidate findings are graded correct, partial, or incorrect, and unmatched findings are counted as missed or hallucinated. Triage and actionable safety recall/precision and per-subset views add complementary, deployment-oriented lenses. Across two expert benchmarks, RadMatch is the most clinically aligned metric, matching inter-radiologist agreement on ReXVal and more than doubling the best prior metric on the harder RadEvalExpert. Relying only on few-shot prompting, it is designed to extend to other modalities and anatomies. We will release RadMatch as open-source code with an interactive dashboard for inspecting results.

Both capability and safety benchmarks rest upon the assumption that the behavior of language models undergoing a test is informative about their behavior in deployment. This assumption can fail, should models infer that they are being evaluated and condition their response on such context. This hypothesis, termed ``evaluation awareness'', has been observed in frontier and open-weight language models alike. We provide a systematic study of this phenomenon, by probing for it across six language models (from four families and three sizes) and three metrics. More precisely, we examine whether (i) being under evaluation is linearly represented within the models' activations space, (ii) it is verbalized in their output tokens (as scored by an LLM-as-judge), and (iii) steering causally affects their behavior. For the open-checkpoint Olmo models, we further test these measures at every training stage. In doing so, we report that evaluation awareness is linearly decodable from the residual streams of every model (best AUROC geq 0.7). By contrast, these representations align only in part with verbalization: their correlations and mutual information are nonzero in some settings, yet vary substantially across models, layers, and readout choices. Nevertheless, steering along probe-derived directions can shift the verbalization scores. Finally, a comparison across the Olmo checkpoints reveals that evaluation awareness is already present within base models, becomes amplified throughout the stages of supervised fine-tuning, and remains stable thereafter---unlike the effects of steering, that grow more pronounced at every successive training stage. These results show the need for evaluations to account for the disjunction between what models represent internally, what they verbalize, and their steering.

The WMT26 General MT task evaluates systems on 10 language pairs that have no human references (neither translated from scratch nor post-edited from MT output by humans). We describe how we built the pseudo-references for these pairs and six other language pairs (in which some forms of human references are available): seven models translate the 3,277 official documents under up to five prompt conditions, giving a total of 26 system-prompt combinations; then three reference-free quality estimation (QE) models score every candidate; and a per-document selector picks one translation, which GPT-5.5 post-edits where needed. Working without references exposed a failure mode of QE-guided selection: the metrics rank fluent output in the wrong language above correct translations. Adding a confidence-scaled language identification penalty to the score fusion drives the wrong-language count to zero, and the resulting selector still scores better on MetricX than the rank-fusion baseline it replaces. Since no references were available for these pairs while we were building them, we calibrate every selection decision on last year's WMT25 human judgments. The human evaluation, released after construction, shows the cost of getting selection wrong: our references stand with the strongest participating systems when the selector kept a frontier-model candidate, and fall up to 17 ESA points below them when it did not. We release the selection method and the provenance of every reference (https://github.com/surrey-nlp/PseudoRef)

Trainable evaluation metrics for machine translation (MT) exhibit strong correlation with human judgements, but they are often hard to interpret and might produce unreliable scores under noisy or out-of-domain data. Recent work has attempted to mitigate this with simple uncertainty quantification techniques (Monte Carlo dropout and deep ensembles), however these techniques (as we show) are limited in several ways -- for example, they are unable to distinguish between different kinds of uncertainty, and they are time and memory consuming. In this paper, we propose more powerful and efficient uncertainty predictors for MT evaluation, and we assess their ability to target different sources of aleatoric and epistemic uncertainty. To this end, we develop and compare training objectives for the COMET metric to enhance it with an uncertainty prediction output, including heteroscedastic regression, divergence minimization, and direct uncertainty prediction. Our experiments show improved results on uncertainty prediction for the WMT metrics task datasets, with a substantial reduction in computational costs. Moreover, they demonstrate the ability of these predictors to address specific uncertainty causes in MT evaluation, such as low quality references and out-of-domain data.

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.

Large Language Models (LLMs) applied to code-related applications have emerged as a prominent field, attracting significant interest from both academia and industry. However, as new and improved LLMs are developed, existing evaluation benchmarks (e.g., HumanEval, MBPP) are no longer sufficient for assessing their capabilities. In this work, we propose LiveCodeBench, a comprehensive and contamination-free evaluation of LLMs for code, which continuously collects new problems over time from contests across three competition platforms, namely LeetCode, AtCoder, and CodeForces. Notably, our benchmark also focuses on a broader range of code related capabilities, such as self-repair, code execution, and test output prediction, beyond just code generation. Currently, LiveCodeBench hosts four hundred high-quality coding problems that were published between May 2023 and February 2024. We have evaluated 9 base LLMs and 20 instruction-tuned LLMs on LiveCodeBench. We present empirical findings on contamination, holistic performance comparisons, potential overfitting in existing benchmarks as well as individual model comparisons. We will release all prompts and model completions for further community analysis, along with a general toolkit for adding new scenarios and model

Recently, there have been notable advancements in large language models (LLMs), demonstrating their growing abilities in complex reasoning. However, existing research largely overlooks a thorough and systematic comparison of these models' reasoning processes and outputs, particularly regarding their self-reflection pattern (also termed "Aha moment") and the interconnections across diverse domains. This paper proposes a novel framework for analyzing the reasoning characteristics of four cutting-edge large reasoning models (GPT-o1, DeepSeek-R1, Kimi-k1.5, and Grok-3) using keywords statistic and LLM-as-a-judge paradigm. Our approach connects their internal thinking processes with their final outputs. A diverse dataset consists of real-world scenario-based questions covering logical deduction, causal inference, and multi-step problem-solving. Additionally, a set of metrics is put forward to assess both the coherence of reasoning and the accuracy of the outputs. The research results uncover various patterns of how these models balance exploration and exploitation, deal with problems, and reach conclusions during the reasoning process. Through quantitative and qualitative comparisons, disparities among these models are identified in aspects such as the depth of reasoning, the reliance on intermediate steps, and the degree of similarity between their thinking processes and output patterns and those of GPT-o1. This work offers valuable insights into the trade-off between computational efficiency and reasoning robustness and provides practical recommendations for enhancing model design and evaluation in practical applications. We publicly release our project at: https://github.com/ChangWenhan/FromThinking2Output

Large language models have demonstrated remarkable capabilities in complex mathematical reasoning tasks, but they inevitably generate errors throughout multi-step solutions. Process-level Reward Models (PRMs) have shown great promise by providing supervision and evaluation at each intermediate step, thereby effectively improving the models' reasoning abilities. However, training effective PRMs requires high-quality process reward data, yet existing methods for constructing such data are often labour-intensive or inefficient. In this paper, we propose an uncertainty-driven framework for automated process reward data construction, encompassing both data generation and annotation processes for PRMs. Additionally, we identify the limitations of both majority vote and PRMs, and introduce two generic uncertainty-aware output aggregation methods: Hybrid Majority Reward Vote and Weighted Reward Frequency Vote, which combine the strengths of majority vote with PRMs. Extensive experiments on ProcessBench, MATH, and GSMPlus show the effectiveness and efficiency of the proposed PRM data construction framework, and demonstrate that the two output aggregation methods further improve the mathematical reasoning abilities across diverse PRMs. The code and data will be publicly available at https://github.com/Jiuzhouh/UnPRM.

Episodic memory -- the ability to recall specific events grounded in time and space -- is a cornerstone of human cognition, enabling not only coherent storytelling, but also planning and decision-making. Despite their remarkable capabilities, Large Language Models (LLMs) lack a robust mechanism for episodic memory: we argue that integrating episodic memory capabilities into LLM is essential for advancing AI towards human-like cognition, increasing their potential to reason consistently and ground their output in real-world episodic events, hence avoiding confabulations. To address this challenge, we introduce a comprehensive framework to model and evaluate LLM episodic memory capabilities. Drawing inspiration from cognitive science, we develop a structured approach to represent episodic events, encapsulating temporal and spatial contexts, involved entities, and detailed descriptions. We synthesize a unique episodic memory benchmark, free from contamination, and release open source code and datasets to assess LLM performance across various recall and episodic reasoning tasks. Our evaluation of state-of-the-art models, including GPT-4 and Claude variants, Llama 3.1, and o1-mini, reveals that even the most advanced LLMs struggle with episodic memory tasks, particularly when dealing with multiple related events or complex spatio-temporal relationships -- even in contexts as short as 10k-100k tokens.

Large language models (LLMs) are increasingly adept at following instructions containing task descriptions to solve complex problems, such as mathematical reasoning and automatic evaluation (LLM-as-a-Judge). However, as prompts grow more complex, models often struggle to adhere to all instructions. This difficulty is especially common when instructive prompts intertwine reasoning directives -- specifying what the model should solve -- with rigid formatting requirements that dictate how the solution must be presented. The entanglement creates competing goals for the model, suggesting that more explicit separation of these two aspects could lead to improved performance. To this front, we introduce Deco-G, a decoding framework that explicitly decouples format adherence from task solving. Deco-G handles format compliance with a separate tractable probabilistic model (TPM), while prompts LLMs with only task instructions. At each decoding step, Deco-G combines next token probabilities from the LLM with the TPM calculated format compliance likelihood to form the output probability. To make this approach both practical and scalable for modern instruction-tuned LLMs, we introduce three key innovations: instruction-aware distillation, a flexible trie-building algorithm, and HMM state pruning for computational efficiency. We demonstrate the effectiveness of Deco-G across a wide range of tasks with diverse format requirements, including mathematical reasoning, LLM-as-a-judge, and event argument extraction. Overall, our approach yields 1.0% to 6.0% relative gain over regular prompting practice with guaranteed format compliance.

This study presents PARROT (Persuasion and Agreement Robustness Rating of Output Truth), a robustness focused framework designed to measure the degradation in accuracy that occurs under social pressure exerted on users through authority and persuasion in large language models (LLMs) the phenomenon of sycophancy (excessive conformity). PARROT (i) isolates causal effects by comparing the neutral version of the same question with an authoritatively false version using a double-blind evaluation, (ii) quantifies confidence shifts toward the correct and imposed false responses using log-likelihood-based calibration tracking, and (iii) systematically classifies failure modes (e.g., robust correct, sycophantic agreement, reinforced error, stubborn error, self-correction, etc.) using an eight-state behavioral taxonomy. We evaluated 22 models using 1,302 MMLU-style multiple-choice questions across 13 domains and domain-specific authority templates. Findings show marked heterogeneity: advanced models (e.g., GPT-5, GPT-4.1, Claude Sonnet 4.5) exhibit low "follow rates" (leq 11%, GPT-5: 4\%) and minimal accuracy loss, while older/smaller models show severe epistemic collapse (GPT-4: 80\%, Qwen 2.5-1.5B: 94\%). The danger is not limited to response changes; weak models reduce confidence in the correct response while increasing confidence in the imposed incorrect response. While international law and global knowledge at the domain level exhibit high fragility, elementary mathematics is relatively resilient. Consequently, we argue that the goal of "resistance to overfitting pressure" should be addressed as a primary objective alongside accuracy, harm avoidance, and privacy for safe deployment in the real world.

Document parsing aims to transform unstructured PDF images into semi-structured data, facilitating the digitization and utilization of information in diverse domains. While vision language models (VLMs) have significantly advanced this task, achieving reliable, high-quality parsing in real-world scenarios remains challenging. Common practice often selects the top-performing model on standard benchmarks. However, these benchmarks may carry dataset-specific biases, leading to inconsistent model rankings and limited correlation with real-world performance. Moreover, benchmark metrics typically provide only overall scores, which can obscure distinct error patterns in output. This raises a key challenge: how can we reliably and comprehensively assess document parsing quality in the wild? We address this problem with DOCR-Inspector, which formalizes document parsing assessment as fine-grained error detection and analysis. Leveraging VLM-as-a-Judge, DOCR-Inspector analyzes a document image and its parsed output, identifies all errors, assigns them to one of 28 predefined types, and produces a comprehensive quality assessment. To enable this capability, we construct DOCRcase-200K for training and propose the Chain-of-Checklist reasoning paradigm to enable the hierarchical structure of parsing quality assessment. For empirical validation, we introduce DOCRcaseBench, a set of 882 real-world document parsing cases with manual annotations. On this benchmark, DOCR-Inspector-7B outperforms commercial models like Gemini 2.5 Pro, as well as leading open-source models. Further experiments demonstrate that its quality assessments provide valuable guidance for parsing results refinement, making DOCR-Inspector both a practical evaluator and a driver for advancing document parsing systems at scale. Model and code are released at: https://github.com/ZZZZZQT/DOCR-Inspector.

Output diversity is crucial for Large Language Models as it underpins pluralism and creativity. In this work, we reveal that controlling the language used during model thinking-the language of thought-provides a novel and structural source of output diversity. Our preliminary study shows that different thinking languages occupy distinct regions in a model's thinking space. Based on this observation, we study two repeated sampling strategies under multilingual thinking-Single-Language Sampling and Mixed-Language Sampling-and conduct diversity evaluation on outputs that are controlled to be in English, regardless of the thinking language used. Across extensive experiments, we demonstrate that switching the thinking language from English to non-English languages consistently increases output diversity, with a clear and consistent positive correlation such that languages farther from English in the thinking space yield larger gains. We further show that aggregating samples across multiple thinking languages yields additional improvements through compositional effects, and that scaling sampling with linguistic heterogeneity expands the model's diversity ceiling. Finally, we show that these findings translate into practical benefits in pluralistic alignment scenarios, leading to broader coverage of cultural knowledge and value orientations in LLM outputs. Our code is publicly available at https://github.com/iNLP-Lab/Multilingual-LoT-Diversity.

Many benchmarks for automated causal inference evaluate a system's performance based on a single numerical output, such as an Average Treatment Effect (ATE). This approach conflates two distinct steps in causal analysis: identification-formulating a valid research design under stated assumptions-and estimation-implementing that design numerically on finite data. We introduce CausalReasoningBenchmark, a benchmark of 173 queries across 138 real-world datasets, curated from 85 peer-reviewed research papers and four widely-used causal-inference textbooks. For each query a system must produce (i) a structured identification specification that names the strategy, the treatment, outcome, and control variables, and all design-specific elements, and (ii) a point estimate with a standard error. By scoring these two components separately, our benchmark enables granular diagnosis: it distinguishes failures in causal reasoning from errors in numerical execution. Baseline results with a state-of-the-art LLM show that, while the model correctly identifies the high-level strategy in 84 % of cases, full identification-specification correctness drops to only 30 %, revealing that the bottleneck lies in the nuanced details of research design rather than in computation. CausalReasoningBenchmark is publicly available on Hugging Face and is designed to foster the development of more robust automated causal-inference systems.

Immense effort has been dedicated to minimizing the presence of harmful or biased generative content and better aligning AI output to human intention; however, research investigating the cultural values of LLMs is still in very early stages. Cultural values underpin how societies operate, providing profound insights into the norms, priorities, and decision making of their members. In recognition of this need for further research, we draw upon cultural psychology theory and the empirically-validated GLOBE framework to propose the LLM-GLOBE benchmark for evaluating the cultural value systems of LLMs, and we then leverage the benchmark to compare the values of Chinese and US LLMs. Our methodology includes a novel "LLMs-as-a-Jury" pipeline which automates the evaluation of open-ended content to enable large-scale analysis at a conceptual level. Results clarify similarities and differences that exist between Eastern and Western cultural value systems and suggest that open-generation tasks represent a more promising direction for evaluation of cultural values. We interpret the implications of this research for subsequent model development, evaluation, and deployment efforts as they relate to LLMs, AI cultural alignment more broadly, and the influence of AI cultural value systems on human-AI collaboration outcomes.

This paper introduces BioAgent Bench, a benchmark dataset and an evaluation suite designed for measuring the performance and robustness of AI agents in common bioinformatics tasks. The benchmark contains curated end-to-end tasks (e.g., RNA-seq, variant calling, metagenomics) with prompts that specify concrete output artifacts to support automated assessment, including stress testing under controlled perturbations. We evaluate frontier closed-source and open-weight models across multiple agent harnesses, and use an LLM-based grader to score pipeline progress and outcome validity. We find that frontier agents can complete multi-step bioinformatics pipelines without elaborate custom scaffolding, often producing the requested final artifacts reliably. However, robustness tests reveal failure modes under controlled perturbations (corrupted inputs, decoy files, and prompt bloat), indicating that correct high-level pipeline construction does not guarantee reliable step-level reasoning. Finally, because bioinformatics workflows may involve sensitive patient data, proprietary references, or unpublished IP, closed-source models can be unsuitable under strict privacy constraints; in such settings, open-weight models may be preferable despite lower completion rates. We release the dataset and evaluation suite publicly.

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.

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.

Recent studies demonstrate that tool-calling capability enables large language models (LLMs) to interact with external environments for long-horizon financial tasks. While existing benchmarks have begun evaluating financial tool calling, they focus on limited scenarios and rely on call-level metrics that fail to capture trajectory-level reasoning quality. To address this gap, we introduce FinTrace, a benchmark comprising 800 expert-annotated trajectories spanning 34 real-world financial task categories across multiple difficulty levels. FinTrace employs a rubric-based evaluation protocol with nine metrics organized along four axes -- action correctness, execution efficiency, process quality, and output quality -- enabling fine-grained assessment of LLM tool-calling behavior. Our evaluation of 13 LLMs reveals that while frontier models achieve strong tool selection, all models struggle with information utilization and final answer quality, exposing a critical gap between invoking the right tools and reasoning effectively over their outputs. To move beyond diagnosis, we construct FinTrace-Training, the first trajectory-level preference dataset for financial tool-calling, containing 8,196 curated trajectories with tool-augmented contexts and preference pairs. We fine-tune Qwen-3.5-9B using supervised fine-tuning followed by direct preference optimization (DPO) and show that training on FinTrace-Training consistently improves intermediate reasoning metrics, with DPO more effectively suppressing failure modes. However, end-to-end answer quality remains a bottleneck, indicating that trajectory-level improvements do not yet fully propagate to final output quality.

With PRECISE, we extended Prediction-Powered Inference to produce bias-corrected estimates of ranking evaluation metrics by combining a small human-labeled set with a large LLM-judged set. PPI is provably unbiased regardless of the LLM judge's error profile. We make it applicable to hierarchical metrics like Precision@K, where annotations are per-document but the metric is per-query, by reducing the output-space computation from O(2^|C|) to O(2^K). On the ESCI benchmark, augmenting 30 human annotations with Claude 3 Sonnet judgments reduces the standard error of Precision@4 estimates from 4.45 to 3.50 (a 21% relative reduction). In a production system, our framework correctly identified the best of three system variants from 100 human labels and 2 hours of domain-expert annotation; A/B testing confirmed this ranking with +407 bps in daily sales.

Existing evaluation frameworks for large language models -- including HELM, MT-Bench, AgentBench, and BIG-bench -- are designed for controlled, single-session, lab-scale settings. They do not address the evaluation challenges that emerge when agentic AI systems operate continuously in production: compounding decision errors, tool failure cascades, non-deterministic output drift, and the absence of ground truth for long-horizon tasks. This paper makes three contributions. First, we present a taxonomy of seven failure modes unique to production agentic systems, each grounded in observations from systems operating at billion-event scale. Second, we demonstrate empirically where standard metrics -- ROUGE, BERTScore, accuracy/AUC, and the agentic benchmarks above -- fail to detect each failure mode. Third, we propose PAEF (Production Agentic Evaluation Framework), a five-dimension evaluation framework with an open-source reference implementation, designed for continuous evaluation on production traffic rather than episodic benchmark runs. Our analysis shows that standard metrics fail to detect four of the seven failure modes entirely and detect three others only after a lag of multiple evaluation cycles.

Vision-Language-Action (VLA) models acquire broad embodied capabilities through large-scale pretraining, yet their generalization remains far more fragile than that of LLMs and VLMs. The prevailing remedy, post-training via supervised fine-tuning or reinforcement learning, improves task-specific performance but narrows the generalist capability that makes pretraining valuable. We identify a key bottleneck: VLA failures stem not only from action generation but also from action evaluation. A diagnostic pass@k study confirms that frozen VLAs already contain competent behaviors in their output distribution, with overall success rates rising from 33% at pass@1 to 92% at pass@32. Inspired by this, we propose SVA (Search, Value, and Act), a simple framework that equips frozen VLA policies with long-term consequence awareness. SVA first uses Monte-Carlo tree search in simulation to fully explore the VLA's output distribution and collect diverse trajectories annotated with empirical returns; this knowledge is then distilled into a lightweight Q-value model that predicts the expected consequence of candidate actions; at deployment, the frozen VLA proposes multiple candidates and the evaluator selects the one with the highest uncertainty-regularized Q-value, requiring no simulator access. By decoupling action proposal from consequence evaluation, SVA preserves the generalization capacity of the VLA backbone while substantially improving task success rates. Experiments across embodied benchmarks show that SVA consistently improves generalization on unseen tasks and exhibits strong test-time scaling behavior. Strikingly, SVA enables a 9B VLA to outperform a 27B VLA by 7 points at 27% lower inference latency, suggesting that scaling test-time evaluation is more cost-effective than scaling model size.

Reproducibility crises across sciences highlight the limitations of the paper-centric review system in assessing the rigor and reproducibility of research. AI agents that autonomously design and generate large volumes of research outputs exacerbate these challenges. In this work, we address the growing challenges of scalability and rigor by flipping the dynamic and developing AI agents as research evaluators. We propose the first execution-grounded evaluation framework that verifies research beyond narrative review by examining code and data alongside the paper. We use mechanistic interpretability research as a testbed, build standardized research output, and develop MechEvalAgent, an automated evaluation framework that assesses the coherence of the experimental process, the reproducibility of results, and the generalizability of findings. We show that our framework achieves above 80% agreement with human judges, identifies substantial methodological problems, and surfaces 51 additional issues that human reviewers miss. Our work demonstrates the potential of AI agents to transform research evaluation and pave the way for rigorous scientific practices.

Agent benchmarks often evaluate only final answers even when agents run on stateful runtimes. We argue this under-specifies what is being evaluated: the proper unit is a declared model-plus-runtime configuration whose failures can occur in evidence acquisition, runtime routing, safety boundaries, or repeated execution. We present ClawProBench, a trace-aware benchmark for runtime-native agent evaluation instantiated on OpenClaw, a live agent runtime with workspace tools and native surfaces for browsing, memory, messaging, scheduling, skills, and subagents. ClawProBench defines two tracks: a 102-scenario full profile with live workspace and native-runtime routing tasks, and a frozen 68-scenario holdout with closed-world JSON output contracts for robust ranking. Trials are scored from execution traces via a safety-gated formula combining correctness, process quality, and efficiency, preserving failure evidence for audit. Our anonymous artifact includes benchmark definitions, scoring code, manifests and sanitized traces. We evaluate 68 configurations on the full profile and 37 on holdout. The top safety-gated average trace score is 0.7671. Native-runtime tasks underperform workspace-live tasks (0.5238 vs. 0.6415). On holdout, pass@k-any outperforms strict three-trial pass (0.6638 vs. 0.2890), while full-profile and holdout rankings show weak alignment (Spearman 0.1300). Rankings based purely on correctness differ substantially from process-aware, safety-gated and strict-pass views. Final-answer leaderboards may hide native-surface weaknesses, one-off successes and trace-local agent failure modes.

Evaluating LLM agents is essential for guiding their development, yet it has grown prohibitively expensive: a single pass of a frontier model over an agentic benchmark can cost hundreds to thousands of dollars, a price paid repeatedly across iterative development cycles. Prior efforts, centered on benchmark distillation, reduce the number of evaluation tasks but leave the cost of executing each retained task untouched. In this work, we introduce early outcome prediction, a complementary axis of efficiency that instead cuts cost within each task. Our key insight is that an agent's final outcome is often evident from its intermediate behavior well before execution completes. We instantiate this idea in EarlyEval, a lightweight framework that trains a pair of LightGBM success and failure classifiers over behavioral, textual, and reference-solution features, and halts an agent run the moment either classifier crosses a calibrated confidence threshold, adding negligible per-step overhead. Across three benchmarks, SWE-bench Verified, TerminalBench, and Toolathlon, EarlyEval can eliminate 13%-26% of agent steps and up to 44.1% input tokens and 29.4% output tokens at 89%-97% prediction accuracy, while perturbing per-agent resolve rates by only one to two percentage points on average.

Evaluating generative models, such as large language models (LLMs), commonly involves question-answering tasks where the final answer is selected based on probability of answer choices. On the other hand, for models requiring reasoning, the method of answer extraction plays a critical role. Our research reveals that the performance of reasoning models and their final answer distributions are highly sensitive to the answer extraction algorithm employed. In order to mitigate this, we propose a basic framework: Answer Regeneration. The method uses an additional model inference, providing the prior input and output prefaced by the prompt "Answer:". The final answer is then selected or extracted from the regenerated output. We show that this extraction-rule-agnostic approach exhibits improved performance and enhanced robustness. Furthermore, we have applied this framework to general math problems and open-ended question answering tasks. Our analysis and this framework could offer a more reliable results for model evaluation.

Retrieval-augmented generation (RAG) evaluations often compare readers after a compressor has changed their evidence. This mixes two questions: which complete pipeline works best, and how much of a reader upgrade survives compression. We show that fixed compression can raise average pipeline accuracy while hiding most of a reader upgrade. We hold questions, retrieved candidates, prompts, scoring, and compressed text fixed while comparing 8-20 readers across five question-answering benchmarks and five compression families. HotpotQA and MuSiQue are the main benchmarks, analyzed under a plan fixed in advance. In the 20-reader HotpotQA panel, the lowest- and highest-scoring readers under raw evidence are 31.8 percentage points (pp) apart before compression but only 7.8pp apart on the same stored output from a HotpotQA-trained RECOMP compressor. On both main benchmarks, lower raw-scoring readers gain more under exact match and token-overlap F1. Reader pairs also change order more often under compressed evidence than in raw replays across question halves. Row-level accounting explains how higher average accuracy and smaller upgrades coexist: compression rescues some wrong answers and damages some correct answers, with a different balance for each reader. We release ragscale to audit declared reader upgrades under raw and fixed compressed evidence. Reader evaluations should run this audit before attributing a compressed-only result to the reader.

Existing synthetic image evaluators typically provide only a scalar quality score and do not identify the image regions that support it. We introduce VIEScore2, a unified evaluator for image generation and editing tasks with optional conditioning images. VIEScore2 represents an image as an N x N grid and jointly predicts quality scores and defect locations in a single model pass. Its text-native grid representation provides a common interface for heterogeneous spatial supervision and enables directly verifiable post-training objectives. We train on 38K examples spanning score-only, localization-only, and joint supervision across generation and editing tasks. Starting from supervised fine-tuning, we further apply GRPO to improve defect localization using rewards that combine cell-level Dice overlap, score accuracy, and output-format validity. A parameter-free parser converts the structured predictions into readable explanations. On the primary suite, VIEScore2 achieves an overall-score SRCC of 0.601, compared with 0.491 for Gemini-3-Flash, the strongest zero-shot general-purpose VLM baseline under matched inputs. For defect localization, VIEScore2 outperforms both general-purpose VLMs and specialized spatial evaluators on three of six benchmarks in per-image grid IoU and ranks among the top three on five, including datasets beyond its training sources.

Automated research systems support literature synthesis, ideation, experiments, writing, and peer review, but their evaluation is dispersed across tasks, benchmarks, and studies that are difficult to compare directly. We review this literature from the perspective of evaluation design and evidence, covering six targets: literature synthesis, research ideation, executable workflows, scholarly writing and communication, automatic peer review, and end-to-end research. We compare task construction, evidence sources, evaluators, and scoring procedures to explain the capabilities assessed by different designs. Our synthesis highlights three recurring lessons: output checks, process checks, and human studies provide complementary information; evaluator calibration is specific to the property being assessed; and resource budgets and attempt selection are integral to interpreting performance comparisons. We identify diagnostic evaluation designs and documented gaps in supporting evidence, and translate these comparisons into reporting and audit recommendations for specific evaluation settings. The survey helps readers navigate existing evaluations, select appropriate benchmarks, and design subsequent studies.

This research investigates prompt designs of evaluating generated texts using large language models (LLMs). While LLMs are increasingly used for scoring various inputs, creating effective prompts for open-ended text evaluation remains challenging due to model sensitivity and subjectivity in evaluation of text generation. Our study experimented with different prompt structures, altering the sequence of output instructions and including explanatory reasons. We found that the order of presenting reasons and scores significantly influences LLMs' scoring, with a different level of rule understanding in the prompt. An additional optimization may enhance scoring alignment if sufficient data is available. This insight is crucial for improving the accuracy and consistency of LLM-based evaluations.

Perceiving and generating diverse modalities are crucial for AI models to effectively learn from and engage with real-world signals, necessitating reliable evaluations for their development. We identify two major issues in current evaluations: (1) inconsistent standards, shaped by different communities with varying protocols and maturity levels; and (2) significant query, grading, and generalization biases. To address these, we introduce MixEval-X, the first any-to-any real-world benchmark designed to optimize and standardize evaluations across input and output modalities. We propose multi-modal benchmark mixture and adaptation-rectification pipelines to reconstruct real-world task distributions, ensuring evaluations generalize effectively to real-world use cases. Extensive meta-evaluations show our approach effectively aligns benchmark samples with real-world task distributions and the model rankings correlate strongly with that of crowd-sourced real-world evaluations (up to 0.98). We provide comprehensive leaderboards to rerank existing models and organizations and offer insights to enhance understanding of multi-modal evaluations and inform future research.

We present the AI weather and climate model intercomparison project (AIMIP), phase 1. Drawing from the rich tradition of intercomparisons in climate model development, we specify a common experiment, output data format, and training constraints (namely, training against historical reanalysis data) for AIMIP Phase 1 models. We aim to identify differences in modeling frameworks and AI architectural choices that influence model behavior, and build trust in AI weather and climate models through open data and evaluation. AIMIP Phase 1 models must simulate the atmosphere given specified historical sea surface temperatures over 1979-2024. We evaluate the models' performance using five major evaluation criteria: biases, trends, response to El Niño-related sea surface temperature anomalies, temporal variability, and out-of-sample generalization tests. We find that the AI models are able to simulate the historical climate and response to forcing as well as a conventional physically-based model, but some AI models underestimate historical warming trends, and their predictions diverge in the out-of-sample generalization tests. We describe the AIMIP Phase 1 dataset that is publicly available for additional evaluations.

Large Language Models (LLMs) have demonstrated impressive performance on Natural Language Processing (NLP) tasks, such as Question Answering, Summarization, and Classification. The use of LLMs as evaluators, that can rank or score the output of other models (usually LLMs) has become increasingly popular, due to the limitations of current evaluation techniques including the lack of appropriate benchmarks, metrics, cost, and access to human annotators. While LLMs are capable of handling approximately 100 languages, the majority of languages beyond the top 20 lack systematic evaluation across various tasks, metrics, and benchmarks. This creates an urgent need to scale up multilingual evaluation to ensure a precise understanding of LLM performance across diverse languages. LLM-based evaluators seem like the perfect solution to this problem, as they do not require human annotators, human-created references, or benchmarks and can theoretically be used to evaluate any language covered by the LLM. In this paper, we investigate whether LLM-based evaluators can help scale up multilingual evaluation. Specifically, we calibrate LLM-based evaluation against 20k human judgments of five metrics across three text-generation tasks in eight languages. Our findings indicate that LLM-based evaluators may exhibit bias towards higher scores and should be used with caution and should always be calibrated with a dataset of native speaker judgments, particularly in low-resource and non-Latin script languages.

We introduce Llama Guard, an LLM-based input-output safeguard model geared towards Human-AI conversation use cases. Our model incorporates a safety risk taxonomy, a valuable tool for categorizing a specific set of safety risks found in LLM prompts (i.e., prompt classification). This taxonomy is also instrumental in classifying the responses generated by LLMs to these prompts, a process we refer to as response classification. For the purpose of both prompt and response classification, we have meticulously gathered a dataset of high quality. Llama Guard, a Llama2-7b model that is instruction-tuned on our collected dataset, albeit low in volume, demonstrates strong performance on existing benchmarks such as the OpenAI Moderation Evaluation dataset and ToxicChat, where its performance matches or exceeds that of currently available content moderation tools. Llama Guard functions as a language model, carrying out multi-class classification and generating binary decision scores. Furthermore, the instruction fine-tuning of Llama Guard allows for the customization of tasks and the adaptation of output formats. This feature enhances the model's capabilities, such as enabling the adjustment of taxonomy categories to align with specific use cases, and facilitating zero-shot or few-shot prompting with diverse taxonomies at the input. We are making Llama Guard model weights available and we encourage researchers to further develop and adapt them to meet the evolving needs of the community for AI safety.

Large Language Models (LLMs) have demonstrated great capabilities in solving a wide range of tasks in a resource-efficient manner through prompting, which does not require task-specific training, but suffers from performance fluctuation when there are multiple prompt candidates. Previous works have introduced gradient-free probability-based prompt selection methods that aim to choose the optimal prompt among the candidates for a given task but fail to provide a comprehensive and fair comparison between each other. In this paper, we propose a unified framework to interpret and evaluate the existing probability-based prompt selection methods by performing extensive experiments on 13 common NLP tasks. We find that all existing methods can be unified into some variant of the method that maximizes the mutual information between the input and the corresponding model output (denoted as MI). Using the finding, we develop several variants of MI and increases the effectiveness of the best prompt selection method from 87.79% to 94.98%, measured as the ratio of the performance of the selected prompt to that of the optimal oracle prompt. Furthermore, we propose a novel calibration method called Calibration by Marginalization (CBM) that is orthogonal to existing methods and helps increase the prompt selection effectiveness of the best method by 99.44%. The code and datasets used in our work will be released at https://github.com/soheeyang/unified-prompt-selection.

Assessing long-form responses generated by Vision-Language Models (VLMs) is challenging. It not only requires checking whether the VLM follows the given instruction but also verifying whether the text output is properly grounded on the given image. Inspired by the recent approach of evaluating LMs with LMs, in this work, we propose to evaluate VLMs with VLMs. For this purpose, we present a new feedback dataset called the Perception Collection, encompassing 15K customized score rubrics that users might care about during assessment. Using the Perception Collection, we train Prometheus-Vision, the first open-source VLM evaluator model that can understand the user-defined score criteria during evaluation. Prometheus-Vision shows the highest Pearson correlation with human evaluators and GPT-4V among open-source models, showing its effectiveness for transparent and accessible evaluation of VLMs. We open-source our code, dataset, and model at https://github.com/kaistAI/prometheus-vision

Evaluating text-to-image models is notoriously difficult. A strong recent approach for assessing text-image faithfulness is based on QG/A (question generation and answering), which uses pre-trained foundational models to automatically generate a set of questions and answers from the prompt, and output images are scored based on whether these answers extracted with a visual question answering model are consistent with the prompt-based answers. This kind of evaluation is naturally dependent on the quality of the underlying QG and VQA models. We identify and address several reliability challenges in existing QG/A work: (a) QG questions should respect the prompt (avoiding hallucinations, duplications, and omissions) and (b) VQA answers should be consistent (not asserting that there is no motorcycle in an image while also claiming the motorcycle is blue). We address these issues with Davidsonian Scene Graph (DSG), an empirically grounded evaluation framework inspired by formal semantics, which is adaptable to any QG/A frameworks. DSG produces atomic and unique questions organized in dependency graphs, which (i) ensure appropriate semantic coverage and (ii) sidestep inconsistent answers. With extensive experimentation and human evaluation on a range of model configurations (LLM, VQA, and T2I), we empirically demonstrate that DSG addresses the challenges noted above. Finally, we present DSG-1k, an open-sourced evaluation benchmark that includes 1,060 prompts, covering a wide range of fine-grained semantic categories with a balanced distribution. We release the DSG-1k prompts and the corresponding DSG questions.

Large language models, particularly multilingual ones, are designed, claimed, and expected to cater to native speakers of varied languages. We hypothesise that the current practices of fine-tuning and evaluating these models may mismatch this intention owing to a heavy reliance on translation, which can introduce translation artefacts and defects. It remains unknown whether the nature of the instruction data has an impact on the model output; on the other hand, it remains questionable whether translated test sets can capture such nuances. Due to the often coupled practices of using translated data in both stages, such imperfections could have been overlooked. This work investigates these issues by using controlled native or translated data during instruction tuning and evaluation stages and observing model results. Experiments on eight base models and eight different benchmarks reveal that native or generation benchmarks display a notable difference between native and translated instruction data especially when model performance is high, whereas other types of test sets cannot. Finally, we demonstrate that regularization is beneficial to bridging this gap on structured but not generative tasks.

As large language models (LLMs) advance, it becomes more challenging to reliably evaluate their output due to the high costs of human evaluation. To make progress towards better LLM autoraters, we introduce FLAMe, a family of Foundational Large Autorater Models. FLAMe is trained on our large and diverse collection of 100+ quality assessment tasks comprising 5M+ human judgments, curated and standardized using publicly released human evaluations from previous research. FLAMe significantly improves generalization to a wide variety of held-out tasks, outperforming LLMs trained on proprietary data like GPT-4 and Claude-3 on many tasks. We show that FLAMe can also serve as a powerful starting point for further downstream fine-tuning, using reward modeling evaluation as a case study (FLAMe-RM). Notably, on RewardBench, our FLAMe-RM-24B model (with an accuracy of 87.8%) is the top-performing generative model trained exclusively on permissively licensed data, outperforming both GPT-4-0125 (85.9%) and GPT-4o (84.7%). Additionally, we explore a more computationally efficient approach using a novel tail-patch fine-tuning strategy to optimize our FLAMe multitask mixture for reward modeling evaluation (FLAMe-Opt-RM), offering competitive RewardBench performance while requiring approximately 25x less training datapoints. Overall, our FLAMe variants outperform all popular proprietary LLM-as-a-Judge models we consider across 8 out of 12 autorater evaluation benchmarks, encompassing 53 quality assessment tasks, including RewardBench and LLM-AggreFact. Finally, our analysis reveals that FLAMe is significantly less biased than these LLM-as-a-Judge models on the CoBBLEr autorater bias benchmark, while effectively identifying high-quality responses for code generation.

Large Language Models (LLM) have become a popular approach for implementing Retrieval Augmented Generation (RAG) systems, and a significant amount of effort has been spent on building good models and metrics. In spite of increased recognition of the need for rigorous evaluation of RAG systems, few tools exist that go beyond the creation of model output and automatic calculation. We present InspectorRAGet, an introspection platform for RAG evaluation. InspectorRAGet allows the user to analyze aggregate and instance-level performance of RAG systems, using both human and algorithmic metrics as well as annotator quality. InspectorRAGet is suitable for multiple use cases and is available publicly to the community. The demo video is available at https://youtu.be/MJhe8QIXcEc

As large language models rapidly evolve to support longer context, there is a notable disparity in their capability to generate output at greater lengths. Recent study suggests that the primary cause for this imbalance may arise from the lack of data with long-output during alignment training. In light of this observation, attempts are made to re-align foundation models with data that fills the gap, which result in models capable of generating lengthy output when instructed. In this paper, we explore the impact of data-quality in tuning a model for long output, and the possibility of doing so from the starting points of human-aligned (instruct or chat) models. With careful data curation, we show that it possible to achieve similar performance improvement in our tuned models, with only a small fraction of training data instances and compute. In addition, we assess the generalizability of such approaches by applying our tuning-recipes to several models. our findings suggest that, while capacities for generating long output vary across different models out-of-the-box, our approach to tune them with high-quality data using lite compute, consistently yields notable improvement across all models we experimented on. We have made public our curated dataset for tuning long-writing capability, the implementations of model tuning and evaluation, as well as the fine-tuned models, all of which can be openly-accessed.