Daily Papers

A common use of NLP is to facilitate the understanding of large document collections, with a shift from using traditional topic models to Large Language Models. Yet the effectiveness of using LLM for large corpus understanding in real-world applications remains under-explored. This study measures the knowledge users acquire with unsupervised, supervised LLM-based exploratory approaches or traditional topic models on two datasets. While LLM-based methods generate more human-readable topics and show higher average win probabilities than traditional models for data exploration, they produce overly generic topics for domain-specific datasets that do not easily allow users to learn much about the documents. Adding human supervision to the LLM generation process improves data exploration by mitigating hallucination and over-genericity but requires greater human effort. In contrast, traditional. models like Latent Dirichlet Allocation (LDA) remain effective for exploration but are less user-friendly. We show that LLMs struggle to describe the haystack of large corpora without human help, particularly domain-specific data, and face scaling and hallucination limitations due to context length constraints. Dataset available at https://huggingface. co/datasets/zli12321/Bills.

Multimodal agents are making rapid progress on general computer-use tasks, yet existing benchmarks remain largely confined to browsers and basic desktop applications, falling short in professional software workflows that dominate real-world scientific and industrial practice. To close this gap, we introduce ProSoftArena, a benchmark and platform specifically for evaluating multimodal agents in professional software environments. We establish the first capability hierarchy tailored to agent use of professional software and construct a benchmark of 436 realistic work and research tasks spanning 6 disciplines and 13 core professional applications. To ensure reliable and reproducible assessment, we build an executable real-computer environment with an execution-based evaluation framework and uniquely incorporate a human-in-the-loop evaluation paradigm. Extensive experiments show that even the best-performing agent attains only a 24.4\% success rate on L2 tasks and completely fails on L3 multi-software workflow. In-depth analysis further provides valuable insights for addressing current agent limitations and more effective design principles, paving the way to build more capable agents in professional software settings. This project is available at: https://prosoftarena.github.io.

Despite recent progress, state-of-the-art question answering models remain vulnerable to a variety of adversarial attacks. While dynamic adversarial data collection, in which a human annotator tries to write examples that fool a model-in-the-loop, can improve model robustness, this process is expensive which limits the scale of the collected data. In this work, we are the first to use synthetic adversarial data generation to make question answering models more robust to human adversaries. We develop a data generation pipeline that selects source passages, identifies candidate answers, generates questions, then finally filters or re-labels them to improve quality. Using this approach, we amplify a smaller human-written adversarial dataset to a much larger set of synthetic question-answer pairs. By incorporating our synthetic data, we improve the state-of-the-art on the AdversarialQA dataset by 3.7F1 and improve model generalisation on nine of the twelve MRQA datasets. We further conduct a novel human-in-the-loop evaluation to show that our models are considerably more robust to new human-written adversarial examples: crowdworkers can fool our model only 8.8% of the time on average, compared to 17.6% for a model trained without synthetic data.

Equipping agents with the capacity to justify made decisions using supporting evidence represents a cornerstone of accountable decision-making. Furthermore, ensuring that justifications are in line with human expectations and societal norms is vital, especially in high-stakes situations such as healthcare. In this work, we propose the use of a debate-based reward model for reinforcement learning agents, where the outcome of a zero-sum debate game quantifies the justifiability of a decision in a particular state. This reward model is then used to train a justifiable policy, whose decisions can be more easily corroborated with supporting evidence. In the debate game, two argumentative agents take turns providing supporting evidence for two competing decisions. Given the proposed evidence, a proxy of a human judge evaluates which decision is better justified. We demonstrate the potential of our approach in learning policies for prescribing and justifying treatment decisions of septic patients. We show that augmenting the reward with the feedback signal generated by the debate-based reward model yields policies highly favored by the judge when compared to the policy obtained solely from the environment rewards, while hardly sacrificing any performance. Moreover, in terms of the overall performance and justifiability of trained policies, the debate-based feedback is comparable to the feedback obtained from an ideal judge proxy that evaluates decisions using the full information encoded in the state. This suggests that the debate game outputs key information contained in states that is most relevant for evaluating decisions, which in turn substantiates the practicality of combining our approach with human-in-the-loop evaluations. Lastly, we showcase that agents trained via multi-agent debate learn to propose evidence that is resilient to refutations and closely aligns with human preferences.

Seamless interaction between AI agents and humans using natural language remains a key goal in AI research. This paper addresses the challenges of developing interactive agents capable of understanding and executing grounded natural language instructions through the IGLU competition at NeurIPS. Despite advancements, challenges such as a scarcity of appropriate datasets and the need for effective evaluation platforms persist. We introduce a scalable data collection tool for gathering interactive grounded language instructions within a Minecraft-like environment, resulting in a Multi-Modal dataset with around 9,000 utterances and over 1,000 clarification questions. Additionally, we present a Human-in-the-Loop interactive evaluation platform for qualitative analysis and comparison of agent performance through multi-turn communication with human annotators. We offer to the community these assets referred to as IDAT (IGLU Dataset And Toolkit) which aim to advance the development of intelligent, interactive AI agents and provide essential resources for further research.

Federated Learning (FL) offers a powerful paradigm for training models on decentralized data, but its promise is often undermined by the immense complexity of designing and deploying robust systems. The need to select, combine, and tune strategies for multifaceted challenges like data heterogeneity and system constraints has become a critical bottleneck, resulting in brittle, bespoke solutions. To address this, we introduce Helmsman, a novel multi-agent system that automates the end-to-end synthesis of federated learning systems from high-level user specifications. It emulates a principled research and development workflow through three collaborative phases: (1) interactive human-in-the-loop planning to formulate a sound research plan, (2) modular code generation by supervised agent teams, and (3) a closed-loop of autonomous evaluation and refinement in a sandboxed simulation environment. To facilitate rigorous evaluation, we also introduce AgentFL-Bench, a new benchmark comprising 16 diverse tasks designed to assess the system-level generation capabilities of agentic systems in FL. Extensive experiments demonstrate that our approach generates solutions competitive with, and often superior to, established hand-crafted baselines. Our work represents a significant step towards the automated engineering of complex decentralized AI systems.

The shift of public debate to the digital sphere has been accompanied by a rise in online hate speech. While many promising approaches for hate speech classification have been proposed, studies often focus only on a single language, usually English, and do not address three key concerns: post-deployment performance, classifier maintenance and infrastructural limitations. In this paper, we introduce a new human-in-the-loop BERT-based hate speech classification pipeline and trace its development from initial data collection and annotation all the way to post-deployment. Our classifier, trained using data from our original corpus of over 422k examples, is specifically developed for the inherently multilingual setting of Switzerland and outperforms with its F1 score of 80.5 the currently best-performing BERT-based multilingual classifier by 5.8 F1 points in German and 3.6 F1 points in French. Our systematic evaluations over a 12-month period further highlight the vital importance of continuous, human-in-the-loop classifier maintenance to ensure robust hate speech classification post-deployment.

Handwritten text recognition (HTR) for Arabic-script languages still lags behind Latin-script HTR, despite recent advances in model architectures, datasets, and benchmarks. We show that data quality is a significant limiting factor in many published datasets and propose CER-HV (CER-based Ranking with Human Verification) as a framework to detect and clean label errors. CER-HV combines a CER-based noise detector, built on a carefully configured Convolutional Recurrent Neural Network (CRNN) with early stopping to avoid overfitting noisy samples, and a human-in-the-loop (HITL) step that verifies high-ranking samples. The framework reveals that several existing datasets contain previously underreported problems, including transcription, segmentation, orientation, and non-text content errors. These have been identified with up to 90 percent precision in the Muharaf and 80-86 percent in the PHTI datasets. We also show that our CRNN achieves state-of-the-art performance across five of the six evaluated datasets, reaching 8.45 percent Character Error Rate (CER) on KHATT (Arabic), 8.26 percent on PHTI (Pashto), 10.66 percent on Ajami, and 10.11 percent on Muharaf (Arabic), all without any data cleaning. We establish a new baseline of 11.3 percent CER on the PHTD (Persian) dataset. Applying CER-HV improves the evaluation CER by 0.3-0.6 percent on the cleaner datasets and 1.0-1.8 percent on the noisier ones. Although our experiments focus on documents written in an Arabic-script language, including Arabic, Persian, Urdu, Ajami, and Pashto, the framework is general and can be applied to other text recognition datasets.

Large Language Models have found application in various mundane and repetitive tasks including Human Resource (HR) support. We worked with the domain experts of SAP SE to develop an HR support chatbot as an efficient and effective tool for addressing employee inquiries. We inserted a human-in-the-loop in various parts of the development cycles such as dataset collection, prompt optimization, and evaluation of generated output. By enhancing the LLM-driven chatbot's response quality and exploring alternative retrieval methods, we have created an efficient, scalable, and flexible tool for HR professionals to address employee inquiries effectively. Our experiments and evaluation conclude that GPT-4 outperforms other models and can overcome inconsistencies in data through internal reasoning capabilities. Additionally, through expert analysis, we infer that reference-free evaluation metrics such as G-Eval and Prometheus demonstrate reliability closely aligned with that of human evaluation.

Recent advances in large language models (LLMs) have led to strong reasoning capabilities; however, evaluating such models in low-resource languages remains challenging due to the lack of standardized benchmarks. In particular, Urdu reasoning evaluation has been limited by the sensitivity of machine translation and an emphasis on general language tasks rather than reasoning benchmarks. In this paper, we propose a contextually ensembled translation framework with human-in-the-loop validation that leverages multiple translation systems to develop Urdu reasoning benchmarks while preserving contextual and structural integrity. Using this framework, we translate widely adopted reasoning and question-answering benchmarks, including MGSM, MATH-500, CommonSenseQA, and OpenBookQA, into Urdu, collectively referred to as UrduBench, and conduct a comprehensive evaluation of both reasoning-oriented and instruction-tuned LLMs across multiple prompting strategies. Our analysis reveals performance differences across (1) four datasets, (2) five task difficulty levels, (3) diverse model architectures, (4) multiple model scaling settings, and (5) language consistency tests. We find that multi-step and symbolic reasoning tasks pose significant challenges in Urdu, and that stable language alignment is a critical prerequisite for robust reasoning. Overall, our work establishes a scalable methodology for standardized reasoning evaluation in Urdu and provides empirical insights into multilingual reasoning failures. This experimental setup is also broadly applicable to other low-resource languages. The code and datasets will be publicly released.

As Large Language Models (LLMs) become more pervasive across various users and scenarios, identifying potential issues when using these models becomes essential. Examples of such issues include: bias, inconsistencies, and hallucination. Although auditing the LLM for these problems is often warranted, such a process is neither easy nor accessible for most. An effective method is to probe the LLM using different versions of the same question. This could expose inconsistencies in its knowledge or operation, indicating potential for bias or hallucination. However, to operationalize this auditing method at scale, we need an approach to create those probes reliably and automatically. In this paper we propose the LLMAuditor framework which is an automatic, and scalable solution, where one uses a different LLM along with human-in-the-loop (HIL). This approach offers verifiability and transparency, while avoiding circular reliance on the same LLM, and increasing scientific rigor and generalizability. Specifically, LLMAuditor includes two phases of verification using humans: standardized evaluation criteria to verify responses, and a structured prompt template to generate desired probes. A case study using questions from the TruthfulQA dataset demonstrates that we can generate a reliable set of probes from one LLM that can be used to audit inconsistencies in a different LLM. This process is enhanced by our structured prompt template with HIL, which not only boosts the reliability of our approach in auditing but also yields the delivery of less hallucinated results. The novelty of our research stems from the development of a comprehensive, general-purpose framework that includes a HIL verified prompt template for auditing responses generated by LLMs.

Human evaluation is increasingly critical for assessing large language models, capturing linguistic nuances, and reflecting user preferences more accurately than traditional automated metrics. However, the resource-intensive nature of this type of annotation process poses significant challenges. The key question driving our work: "is it feasible to minimize human-in-the-loop feedback by prioritizing data instances which most effectively distinguish between models?" We evaluate several metric-based methods and find that these metrics enhance the efficiency of human evaluations by minimizing the number of required annotations, thus saving time and cost, while ensuring a robust performance evaluation. We show that our method is effective across widely used model families, reducing instances of indecisive (or "tie") outcomes by up to 54% compared to a random sample when focusing on the top-20 percentile of prioritized instances. This potential reduction in required human effort positions our approach as a valuable strategy in future large language model evaluations.

There has been significant debate in the NLP community about whether or not attention weights can be used as an explanation - a mechanism for interpreting how important each input token is for a particular prediction. The validity of "attention as explanation" has so far been evaluated by computing the rank correlation between attention-based explanations and existing feature attribution explanations using LSTM-based models. In our work, we (i) compare the rank correlation between five more recent feature attribution methods and two attention-based methods, on two types of NLP tasks, and (ii) extend this analysis to also include transformer-based models. We find that attention-based explanations do not correlate strongly with any recent feature attribution methods, regardless of the model or task. Furthermore, we find that none of the tested explanations correlate strongly with one another for the transformer-based model, leading us to question the underlying assumption that we should measure the validity of attention-based explanations based on how well they correlate with existing feature attribution explanation methods. After conducting experiments on five datasets using two different models, we argue that the community should stop using rank correlation as an evaluation metric for attention-based explanations. We suggest that researchers and practitioners should instead test various explanation methods and employ a human-in-the-loop process to determine if the explanations align with human intuition for the particular use case at hand.

Large Language Models (LLMs) based autonomous agents demonstrate multifaceted capabilities to contribute substantially to economic production. However, existing benchmarks remain focused on single agentic capability, failing to capture long-horizon real-world scenarios. Moreover, the reliance on human-in-the-loop feedback for realistic tasks creates a scalability bottleneck, hindering automated rollout collection and evaluation. To bridge this gap, we introduce AgencyBench, a comprehensive benchmark derived from daily AI usage, evaluating 6 core agentic capabilities across 32 real-world scenarios, comprising 138 tasks with specific queries, deliverables, and rubrics. These scenarios require an average of 90 tool calls, 1 million tokens, and hours of execution time to resolve. To enable automated evaluation, we employ a user simulation agent to provide iterative feedback, and a Docker sandbox to conduct visual and functional rubric-based assessment. Experiments reveal that closed-source models significantly outperform open-source models (48.4% vs 32.1%). Further analysis reveals significant disparities across models in resource efficiency, feedback-driven self-correction, and specific tool-use preferences. Finally, we investigate the impact of agentic scaffolds, observing that proprietary models demonstrate superior performance within their native ecosystems (e.g., Claude-4.5-Opus via Claude-Agent-SDK), while open-source models exhibit distinct performance peaks, suggesting potential optimization for specific execution frameworks. AgencyBench serves as a critical testbed for next-generation agents, highlighting the necessity of co-optimizing model architecture with agentic frameworks. We believe this work sheds light on the future direction of autonomous agents, and we release the full benchmark and evaluation toolkit at https://github.com/GAIR-NLP/AgencyBench.

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

End-to-end (E2E) spoken dialogue systems are increasingly replacing cascaded pipelines for voice-based human-AI interaction, processing raw audio directly without intermediate transcription. Existing benchmarks primarily evaluate these models on synthetic speech and single-turn tasks, leaving realistic multi-turn conversational ability underexplored. We introduce Audio MultiChallenge, an open-source benchmark to evaluate E2E spoken dialogue systems under natural multi-turn interaction patterns. Building on the text-based MultiChallenge framework, which evaluates Inference Memory, Instruction Retention, and Self Coherence, we introduce a new axis Voice Editing that tests robustness to mid-utterance speech repairs and backtracking. We further augment each axis to the audio modality, such as introducing Audio-Cue challenges for Inference Memory that require recalling ambient sounds and paralinguistic signals beyond semantic content. We curate 452 conversations from 47 speakers with 1,712 instance-specific rubrics through a hybrid audio-native agentic and human-in-the-loop pipeline that exposes model failures at scale while preserving natural disfluencies found in unscripted human speech. Our evaluation of proprietary and open-source models reveals that even frontier models struggle on our benchmark, with Gemini 3 Pro Preview (Thinking), our highest-performing model achieving a 54.65% pass rate. Error analysis shows that models fail most often on our new axes and that Self Coherence degrades with longer audio context. These failures reflect difficulty of tracking edits, audio cues, and long-range context in natural spoken dialogue. Audio MultiChallenge provides a reproducible testbed to quantify them and drive improvements in audio-native multi-turn interaction capability.

Reinforcement learning (RL), large language models (LLMs), and vision-language models (VLMs) have been widely studied in isolation. However, existing infrastructure lacks the ability to deploy agents from different decision-making paradigms within the same environment, making it difficult to study them in hybrid multi-agent settings or to compare their behaviour fairly under identical conditions. We present MOSAIC, an open-source platform that bridges this gap by incorporating a diverse set of existing reinforcement learning environments and enabling heterogeneous agents (RL policies, LLMs, VLMs, and human players) to operate within them in ad-hoc team settings with reproducible results. MOSAIC introduces three contributions. (i) An IPC-based worker protocol that wraps both native and third-party frameworks as isolated subprocess workers, each executing its native training and inference logic unmodified, communicating through a versioned inter-process protocol. (ii) An operator abstraction that forms an agent-level interface by mapping workers to agents: each operator, regardless of whether it is backed by an RL policy, an LLM, or a human, conforms to a minimal unified interface. (iii) A deterministic cross-paradigm evaluation framework offering two complementary modes: a manual mode that advances up to N concurrent operators in lock-step under shared seeds for fine-grained visual inspection of behavioural differences, and a script mode that drives automated, long-running evaluation through declarative Python scripts, for reproducible experiments. We release MOSAIC as an open, visual-first platform to facilitate reproducible cross-paradigm research across the RL, LLM, and human-in-the-loop communities.

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

Capture The Flag (CTF) challenges are puzzles related to computer security scenarios. With the advent of large language models (LLMs), more and more CTF participants are using LLMs to understand and solve the challenges. However, so far no work has evaluated the effectiveness of LLMs in solving CTF challenges with a fully automated workflow. We develop two CTF-solving workflows, human-in-the-loop (HITL) and fully-automated, to examine the LLMs' ability to solve a selected set of CTF challenges, prompted with information about the question. We collect human contestants' results on the same set of questions, and find that LLMs achieve higher success rate than an average human participant. This work provides a comprehensive evaluation of the capability of LLMs in solving real world CTF challenges, from real competition to fully automated workflow. Our results provide references for applying LLMs in cybersecurity education and pave the way for systematic evaluation of offensive cybersecurity capabilities in LLMs.

Large language models are rapidly evolving into interactive coding agents capable of end-to-end web coding, yet existing benchmarks evaluate only narrow slices of this capability, typically text-conditioned generation with static-correctness metrics, leaving visual fidelity, interaction quality, and codebase-level reasoning largely unmeasured. We introduce WebCompass, a multimodal benchmark that provides unified lifecycle evaluation of web engineering capability. Recognizing that real-world web coding is an iterative cycle of generation, editing, and repair, WebCompass spans three input modalities (text, image, video) and three task types (generation, editing, repair), yielding seven task categories that mirror professional workflows. Through a multi-stage, human-in-the-loop pipeline, we curate instances covering 15 generation domains, 16 editing operation types, and 11 repair defect types, each annotated at Easy/Medium/Hard levels. For evaluation, we adopt a checklist-guided LLM-as-a-Judge protocol for editing and repair, and propose a novel Agent-as-a-Judge paradigm for generation that autonomously executes generated websites in a real browser, explores interactive behaviors via the Model Context Protocol (MCP), and iteratively synthesizes targeted test cases, closely approximating human acceptance testing. We evaluate representative closed-source and open-source models and observe that: (1) closed-source models remain substantially stronger and more balanced; (2) editing and repair exhibit distinct difficulty profiles, with repair preserving interactivity better but remaining execution-challenging; (3) aesthetics is the most persistent bottleneck, especially for open-source models; and (4) framework choice materially affects outcomes, with Vue consistently challenging while React and Vanilla/HTML perform more strongly depending on task type.

We introduce BSDetector, a method for detecting bad and speculative answers from a pretrained Large Language Model by estimating a numeric confidence score for any output it generated. Our uncertainty quantification technique works for any LLM accessible only via a black-box API, whose training data remains unknown. By expending a bit of extra computation, users of any LLM API can now get the same response as they would ordinarily, as well as a confidence estimate that cautions when not to trust this response. Experiments on both closed and open-form Question-Answer benchmarks reveal that BSDetector more accurately identifies incorrect LLM responses than alternative uncertainty estimation procedures (for both GPT-3 and ChatGPT). By sampling multiple responses from the LLM and considering the one with the highest confidence score, we can additionally obtain more accurate responses from the same LLM, without any extra training steps. In applications involving automated evaluation with LLMs, accounting for our confidence scores leads to more reliable evaluation in both human-in-the-loop and fully-automated settings (across both GPT 3.5 and 4).

Evaluating language models and AI agents remains fundamentally challenging because static benchmarks fail to capture real-world uncertainty, distribution shift, and the gap between isolated task accuracy and human-aligned decision-making under evolving conditions. This paper introduces TruthTensor, a novel, reproducible evaluation paradigm that measures reasoning models not only as prediction engines but as human-imitation systems operating in socially-grounded, high-entropy environments. Building on forward-looking, contamination-free tasks, our framework anchors evaluation to live prediction markets and combines probabilistic scoring to provide a holistic view of model behavior. TruthTensor complements traditional correctness metrics with drift-centric diagnostics and explicit robustness checks for reproducibility. It specify human vs. automated evaluation roles, annotation protocols, and statistical testing procedures to ensure interpretability and replicability of results. In experiments across 500+ real markets (political, economic, cultural, technological), TruthTensor demonstrates that models with similar forecast accuracy can diverge markedly in calibration, drift, and risk-sensitivity, underscoring the need to evaluate models along multiple axes (accuracy, calibration, narrative stability, cost, and resource efficiency). TruthTensor therefore operationalizes modern evaluation best practices, clear hypothesis framing, careful metric selection, transparent compute/cost reporting, human-in-the-loop validation, and open, versioned evaluation contracts, to produce defensible assessments of LLMs in real-world decision contexts. We publicly released TruthTensor at https://truthtensor.com.

The emergence of multimodal large language models (MLLMs) has driven breakthroughs in egocentric vision applications. These applications necessitate persistent, context-aware understanding of objects, as users interact with tools in dynamic and cluttered environments. However, existing embodied benchmarks primarily focus on static scene exploration, emphasizing object's appearance and spatial attributes while neglecting the assessment of dynamic changes arising from users' interactions. To address this gap, we introduce EOC-Bench, an innovative benchmark designed to systematically evaluate object-centric embodied cognition in dynamic egocentric scenarios. Specially, EOC-Bench features 3,277 meticulously annotated QA pairs categorized into three temporal categories: Past, Present, and Future, covering 11 fine-grained evaluation dimensions and 3 visual object referencing types. To ensure thorough assessment, we develop a mixed-format human-in-the-loop annotation framework with four types of questions and design a novel multi-scale temporal accuracy metric for open-ended temporal evaluation. Based on EOC-Bench, we conduct comprehensive evaluations of various proprietary, open-source, and object-level MLLMs. EOC-Bench serves as a crucial tool for advancing the embodied object cognitive capabilities of MLLMs, establishing a robust foundation for developing reliable core models for embodied systems.

We present Habitat 3.0: a simulation platform for studying collaborative human-robot tasks in home environments. Habitat 3.0 offers contributions across three dimensions: (1) Accurate humanoid simulation: addressing challenges in modeling complex deformable bodies and diversity in appearance and motion, all while ensuring high simulation speed. (2) Human-in-the-loop infrastructure: enabling real human interaction with simulated robots via mouse/keyboard or a VR interface, facilitating evaluation of robot policies with human input. (3) Collaborative tasks: studying two collaborative tasks, Social Navigation and Social Rearrangement. Social Navigation investigates a robot's ability to locate and follow humanoid avatars in unseen environments, whereas Social Rearrangement addresses collaboration between a humanoid and robot while rearranging a scene. These contributions allow us to study end-to-end learned and heuristic baselines for human-robot collaboration in-depth, as well as evaluate them with humans in the loop. Our experiments demonstrate that learned robot policies lead to efficient task completion when collaborating with unseen humanoid agents and human partners that might exhibit behaviors that the robot has not seen before. Additionally, we observe emergent behaviors during collaborative task execution, such as the robot yielding space when obstructing a humanoid agent, thereby allowing the effective completion of the task by the humanoid agent. Furthermore, our experiments using the human-in-the-loop tool demonstrate that our automated evaluation with humanoids can provide an indication of the relative ordering of different policies when evaluated with real human collaborators. Habitat 3.0 unlocks interesting new features in simulators for Embodied AI, and we hope it paves the way for a new frontier of embodied human-AI interaction capabilities.

The impressive performance of Large Language Models (LLMs) has consistently surpassed numerous human-designed benchmarks, presenting new challenges in assessing the shortcomings of LLMs. Designing tasks and finding LLMs' limitations are becoming increasingly important. In this paper, we investigate the question of whether an LLM can discover its own limitations from the errors it makes. To this end, we propose a Self-Challenge evaluation framework with human-in-the-loop. Starting from seed instances that GPT-4 fails to answer, we prompt GPT-4 to summarize error patterns that can be used to generate new instances and incorporate human feedback on them to refine these patterns for generating more challenging data, iteratively. We end up with 8 diverse patterns, such as text manipulation and questions with assumptions. We then build a benchmark, SC-G4, consisting of 1,835 instances generated by GPT-4 using these patterns, with human-annotated gold responses. The SC-G4 serves as a challenging benchmark that allows for a detailed assessment of LLMs' abilities. Our results show that only 44.96\% of instances in SC-G4 can be answered correctly by GPT-4. Interestingly, our pilot study indicates that these error patterns also challenge other LLMs, such as Claude-3 and Llama-3, and cannot be fully resolved through fine-tuning. Our work takes the first step to demonstrate that LLMs can autonomously identify their inherent flaws and provide insights for future dynamic and automatic evaluation.

Alignment has become a critical step for instruction-tuned Large Language Models (LLMs) to become helpful assistants. However, effective evaluation of alignment for emerging Chinese LLMs is still significantly lacking, calling for real-scenario grounded, open-ended, challenging and automatic evaluations tailored for alignment. To fill in this gap, we introduce AlignBench, a comprehensive multi-dimensional benchmark for evaluating LLMs' alignment in Chinese. Equipped with a human-in-the-loop data curation pipeline, our benchmark employs a rule-calibrated multi-dimensional LLM-as-Judge with Chain-of-Thought to generate explanations and final ratings as evaluations, ensuring high reliability and interpretability. Furthermore, we report AlignBench evaluated by CritiqueLLM, a dedicated Chinese evaluator LLM that recovers 95% of GPT-4's evaluation ability. We will provide public APIs for evaluating AlignBench with CritiqueLLM to facilitate the evaluation of LLMs' Chinese alignment. All evaluation codes, data, and LLM generations are available at https://github.com/THUDM/AlignBench.

Large multimodal models (LMMs) have shown great potential for video reasoning with textual Chain-of-Thought. However, they remain vulnerable to hallucinations, especially when processing long-form videos where evidence is sparse and temporally dispersed. Inspired by how humans comprehend long videos - by first skimming globally and then examining relevant clips for details - we introduce LongVT, an end-to-end agentic framework that enables "Thinking with Long Videos" via interleaved Multimodal Chain-of-Tool-Thought. Specifically, we exploit LMMs' inherent temporal grounding ability as a native video cropping tool to zoom in on a specific video clip and resample finer-grained video frames. This global-to-local reasoning loop continues until answers are grounded in retrieved visual evidence. Given the scarcity of fine-grained question-answering (QA) data for the long video reasoning task, we curate and will release a data suite named VideoSIAH to facilitate both training and evaluation. Specifically, our training dataset consists of 247.9K samples for tool-integrated cold-start supervised fine-tuning, 1.6K samples for agentic reinforcement learning, and 15.4K samples for agentic reinforcement fine-tuning, respectively. Our evaluation benchmark consists of 1,280 QA pairs that are carefully curated through a semi-automatic data pipeline with human-in-the-loop validation. With a meticulously designed three-stage training strategy and extensive empirical validation, LongVT consistently outperforms existing strong baselines across four challenging long-video understanding and reasoning benchmarks. Our codes, data, and model checkpoints are publicly available at https://github.com/EvolvingLMMs-Lab/LongVT .

Recent studies have shown the impressive efficacy of counterfactually augmented data (CAD) for reducing NLU models' reliance on spurious features and improving their generalizability. However, current methods still heavily rely on human efforts or task-specific designs to generate counterfactuals, thereby impeding CAD's applicability to a broad range of NLU tasks. In this paper, we present AutoCAD, a fully automatic and task-agnostic CAD generation framework. AutoCAD first leverages a classifier to unsupervisedly identify rationales as spans to be intervened, which disentangles spurious and causal features. Then, AutoCAD performs controllable generation enhanced by unlikelihood training to produce diverse counterfactuals. Extensive evaluations on multiple out-of-domain and challenge benchmarks demonstrate that AutoCAD consistently and significantly boosts the out-of-distribution performance of powerful pre-trained models across different NLU tasks, which is comparable or even better than previous state-of-the-art human-in-the-loop or task-specific CAD methods. The code is publicly available at https://github.com/thu-coai/AutoCAD.

We present POTATO, a task- and languageindependent framework for human-in-the-loop (HITL) learning of rule-based text classifiers using graph-based features. POTATO handles any type of directed graph and supports parsing text into Abstract Meaning Representations (AMR), Universal Dependencies (UD), and 4lang semantic graphs. A streamlit-based user interface allows users to build rule systems from graph patterns, provides real-time evaluation based on ground truth data, and suggests rules by ranking graph features using interpretable machine learning models. Users can also provide patterns over graphs using regular expressions, and POTATO can recommend refinements of such rules. POTATO is applied in projects across domains and languages, including classification tasks on German legal text and English social media data. All components of our system are written in Python, can be installed via pip, and are released under an MIT License on GitHub.

Formulating effective search queries remains a challenging task, particularly when users lack expertise in a specific domain or are not proficient in the language of the content. Providing example documents of interest might be easier for a user. However, such query-by-example scenarios are prone to concept drift, and the retrieval effectiveness is highly sensitive to the query generation method, without a clear way to incorporate user feedback. To enable exploration and to support Human-In-The-Loop experiments we propose QueryExplorer -- an interactive query generation, reformulation, and retrieval interface with support for HuggingFace generation models and PyTerrier's retrieval pipelines and datasets, and extensive logging of human feedback. To allow users to create and modify effective queries, our demo supports complementary approaches of using LLMs interactively, assisting the user with edits and feedback at multiple stages of the query formulation process. With support for recording fine-grained interactions and user annotations, QueryExplorer can serve as a valuable experimental and research platform for annotation, qualitative evaluation, and conducting Human-in-the-Loop (HITL) experiments for complex search tasks where users struggle to formulate queries.

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

With the rise of vision-language models (VLM), their application for autonomous driving (VLM4AD) has gained significant attention. Meanwhile, in autonomous driving, closed-loop evaluation has become widely recognized as a more reliable validation method than open-loop evaluation, as it can evaluate the performance of the model under cumulative errors and out-of-distribution inputs. However, existing VLM4AD benchmarks evaluate the model`s scene understanding ability under open-loop, i.e., via static question-answer (QA) dataset. This kind of evaluation fails to assess the VLMs performance under out-of-distribution states rarely appeared in the human collected datasets.To this end, we present Bench2Drive-VL, an extension of Bench2Drive that brings closed-loop evaluation to VLM-based driving, which introduces: (1) DriveCommenter, a closed-loop generator that automatically generates diverse, behavior-grounded question-answer pairs for all driving situations in CARLA,including severe off-route and off-road deviations previously unassessable in simulation. (2) A unified protocol and interface that allows modern VLMs to be directly plugged into the Bench2Drive closed-loop environment to compare with traditional agents. (3) A flexible reasoning and control framework, supporting multi-format visual inputs and configurable graph-based chain-of-thought execution. (4) A complete development ecosystem. Together, these components form a comprehensive closed-loop benchmark for VLM4AD. All codes and annotated datasets are open sourced.

We introduce Dynatask: an open source system for setting up custom NLP tasks that aims to greatly lower the technical knowledge and effort required for hosting and evaluating state-of-the-art NLP models, as well as for conducting model in the loop data collection with crowdworkers. Dynatask is integrated with Dynabench, a research platform for rethinking benchmarking in AI that facilitates human and model in the loop data collection and evaluation. To create a task, users only need to write a short task configuration file from which the relevant web interfaces and model hosting infrastructure are automatically generated. The system is available at https://dynabench.org/ and the full library can be found at https://github.com/facebookresearch/dynabench.

Scams exploiting real-time social engineering -- such as phishing, impersonation, and phone fraud -- remain a persistent and evolving threat across digital platforms. Existing defenses are largely reactive, offering limited protection during active interactions. We propose a privacy-preserving, AI-in-the-loop framework that proactively detects and disrupts scam conversations in real time. The system combines instruction-tuned artificial intelligence with a safety-aware utility function that balances engagement with harm minimization, and employs federated learning to enable continual model updates without raw data sharing. Experimental evaluations show that the system produces fluent and engaging responses (perplexity as low as 22.3, engagement approx0.80), while human studies confirm significant gains in realism, safety, and effectiveness over strong baselines. In federated settings, models trained with FedAvg sustain up to 30 rounds while preserving high engagement (approx0.80), strong relevance (approx0.74), and low PII leakage (leq0.0085). Even with differential privacy, novelty and safety remain stable, indicating that robust privacy can be achieved without sacrificing performance. The evaluation of guard models (LlamaGuard, LlamaGuard2/3, MD-Judge) shows a straightforward pattern: stricter moderation settings reduce the chance of exposing personal information, but they also limit how much the model engages in conversation. In contrast, more relaxed settings allow longer and richer interactions, which improve scam detection, but at the cost of higher privacy risk. To our knowledge, this is the first framework to unify real-time scam-baiting, federated privacy preservation, and calibrated safety moderation into a proactive defense paradigm.

Large Language Models are increasingly used as judges to evaluate code artifacts when exhaustive human review or executable test coverage is unavailable. LLM-judge is increasingly relevant in agentic software engineering workflows, where it can help rank candidate solutions and guide patch selection. While attractive for scale, current practice lacks a principled account of reliability and bias: repeated evaluations of the same case can disagree; small prompt edits can swing outcomes; and seemingly semantics-preserving, human-equivalent perturbations may elicit divergent verdicts. This paper studies LLM-as-a-Judge for code through a measurement-first lens. We analyze two pointwise judging regimes across code generation, code repair task, and test generation, and we systematically probe prompt-induced biases. Our study considers difficulty levels for repeated runs and controlled prompt interventions that isolate one presentation cue at a time, and it evaluates judges using consistency and sensitivity to bias. We find that judge decisions are highly sensitive to prompt biases even when the underlying code snippet is unchanged. Across all three tasks, several biases systematically shift preferences toward the option favored by the prompt, improving accuracy when that option aligns with the gold answer but substantially reducing it otherwise. In some settings, these effects are large enough to change task-level conclusions and alter relative model rankings. These findings show that reported judge performance may reflect prompt artifacts rather than stable assessment ability, posing a direct threat to the validity and reproducibility of code evaluation. We therefore argue that LLM-as-a-Judge studies should report bias sensitivity alongside accuracy and incorporate explicit controls to support more trustworthy model comparison in software engineering.

Rigorously evaluating machine intelligence against the broad spectrum of human general intelligence has become increasingly important and challenging in this era of rapid technological advance. Conventional AI benchmarks typically assess only narrow capabilities in a limited range of human activity. Most are also static, quickly saturating as developers explicitly or implicitly optimize for them. We propose that a more promising way to evaluate human-like general intelligence in AI systems is through a particularly strong form of general game playing: studying how and how well they play and learn to play all conceivable human games, in comparison to human players with the same level of experience, time, or other resources. We define a "human game" to be a game designed by humans for humans, and argue for the evaluative suitability of this space of all such games people can imagine and enjoy -- the "Multiverse of Human Games". Taking a first step towards this vision, we introduce the AI GameStore, a scalable and open-ended platform that uses LLMs with humans-in-the-loop to synthesize new representative human games, by automatically sourcing and adapting standardized and containerized variants of game environments from popular human digital gaming platforms. As a proof of concept, we generated 100 such games based on the top charts of Apple App Store and Steam, and evaluated seven frontier vision-language models (VLMs) on short episodes of play. The best models achieved less than 10\% of the human average score on the majority of the games, and especially struggled with games that challenge world-model learning, memory and planning. We conclude with a set of next steps for building out the AI GameStore as a practical way to measure and drive progress toward human-like general intelligence in machines.

Automated machine learning (AutoML) seeks to build ML models with minimal human effort. While considerable research has been conducted in the area of AutoML in general, aiming to take humans out of the loop when building artificial intelligence (AI) applications, scant literature has focused on how AutoML works well in open-environment scenarios such as the process of training and updating large models, industrial supply chains or the industrial metaverse, where people often face open-loop problems during the search process: they must continuously collect data, update data and models, satisfy the requirements of the development and deployment environment, support massive devices, modify evaluation metrics, etc. Addressing the open-environment issue with pure data-driven approaches requires considerable data, computing resources, and effort from dedicated data engineers, making current AutoML systems and platforms inefficient and computationally intractable. Human-computer interaction is a practical and feasible way to tackle the problem of open-environment AI. In this paper, we introduce OmniForce, a human-centered AutoML (HAML) system that yields both human-assisted ML and ML-assisted human techniques, to put an AutoML system into practice and build adaptive AI in open-environment scenarios. Specifically, we present OmniForce in terms of ML version management; pipeline-driven development and deployment collaborations; a flexible search strategy framework; and widely provisioned and crowdsourced application algorithms, including large models. Furthermore, the (large) models constructed by OmniForce can be automatically turned into remote services in a few minutes; this process is dubbed model as a service (MaaS). Experimental results obtained in multiple search spaces and real-world use cases demonstrate the efficacy and efficiency of OmniForce.

Large language models (LLMs) have enabled agent-based systems that aim to automate scientific research workflows. Most existing approaches focus on fully autonomous discovery, where AI systems generate research ideas, conduct analyses, and produce manuscripts with minimal human involvement. However, empirical research in economics and the social sciences poses additional constraints: research questions must be grounded in available datasets, identification strategies require careful design, and human judgment remains essential for evaluating economic significance. We introduce HLER (Human-in-the-Loop Economic Research), a multi-agent architecture that supports empirical research automation while preserving critical human oversight. The system orchestrates specialized agents for data auditing, data profiling, hypothesis generation, econometric analysis, manuscript drafting, and automated review. A key design principle is dataset-aware hypothesis generation, where candidate research questions are constrained by dataset structure, variable availability, and distributional diagnostics, reducing infeasible or hallucinated hypotheses. HLER further implements a two-loop architecture: a question quality loop that screens and selects feasible hypotheses, and a research revision loop where automated review triggers re-analysis and manuscript revision. Human decision gates are embedded at key stages, allowing researchers to guide the automated pipeline. Experiments on three empirical datasets show that dataset-aware hypothesis generation produces feasible research questions in 87% of cases (versus 41% under unconstrained generation), while complete empirical manuscripts can be produced at an average API cost of 0.8-1.5 per run. These results suggest that Human-AI collaborative pipelines may provide a practical path toward scalable empirical research.

Revision is an essential part of the human writing process. It tends to be strategic, adaptive, and, more importantly, iterative in nature. Despite the success of large language models on text revision tasks, they are limited to non-iterative, one-shot revisions. Examining and evaluating the capability of large language models for making continuous revisions and collaborating with human writers is a critical step towards building effective writing assistants. In this work, we present a human-in-the-loop iterative text revision system, Read, Revise, Repeat (R3), which aims at achieving high quality text revisions with minimal human efforts by reading model-generated revisions and user feedbacks, revising documents, and repeating human-machine interactions. In R3, a text revision model provides text editing suggestions for human writers, who can accept or reject the suggested edits. The accepted edits are then incorporated into the model for the next iteration of document revision. Writers can therefore revise documents iteratively by interacting with the system and simply accepting/rejecting its suggested edits until the text revision model stops making further revisions or reaches a predefined maximum number of revisions. Empirical experiments show that R3 can generate revisions with comparable acceptance rate to human writers at early revision depths, and the human-machine interaction can get higher quality revisions with fewer iterations and edits. The collected human-model interaction dataset and system code are available at https://github.com/vipulraheja/IteraTeR. Our system demonstration is available at https://youtu.be/lK08tIpEoaE.

As AI systems become more integrated into daily life, the need for safer and more reliable moderation has never been greater. Large Language Models (LLMs) have demonstrated remarkable capabilities, surpassing earlier models in complexity and performance. Their evaluation across diverse tasks has consistently showcased their potential, enabling the development of adaptive and personalized agents. However, despite these advancements, LLMs remain prone to errors, particularly in areas requiring nuanced moral reasoning. They struggle with detecting implicit hate, offensive language, and gender biases due to the subjective and context-dependent nature of these issues. Moreover, their reliance on training data can inadvertently reinforce societal biases, leading to inconsistencies and ethical concerns in their outputs. To explore the limitations of LLMs in this role, we developed an experimental framework based on state-of-the-art (SOTA) models to assess human emotions and offensive behaviors. The framework introduces a unified benchmark dataset encompassing 49 distinct categories spanning the wide spectrum of human emotions, offensive and hateful text, and gender and racial biases. Furthermore, we introduced SafePhi, a QLoRA fine-tuned version of Phi-4, adapting diverse ethical contexts and outperforming benchmark moderators by achieving a Macro F1 score of 0.89, where OpenAI Moderator and Llama Guard score 0.77 and 0.74, respectively. This research also highlights the critical domains where LLM moderators consistently underperformed, pressing the need to incorporate more heterogeneous and representative data with human-in-the-loop, for better model robustness and explainability.

We present IMTLab, an open-source end-to-end interactive machine translation (IMT) system platform that enables researchers to quickly build IMT systems with state-of-the-art models, perform an end-to-end evaluation, and diagnose the weakness of systems. IMTLab treats the whole interactive translation process as a task-oriented dialogue with a human-in-the-loop setting, in which human interventions can be explicitly incorporated to produce high-quality, error-free translations. To this end, a general communication interface is designed to support the flexible IMT architectures and user policies. Based on the proposed design, we construct a simulated and real interactive environment to achieve end-to-end evaluation and leverage the framework to systematically evaluate previous IMT systems. Our simulated and manual experiments show that the prefix-constrained decoding approach still gains the lowest editing cost in the end-to-end evaluation, while BiTIIMT achieves comparable editing cost with a better interactive experience.

Although Multimodal Large Language Models (MLLMs) demonstrate strong omni-modal perception, their ability to forecast future events from audio-visual cues remains largely unexplored, as existing benchmarks focus mainly on retrospective understanding. To bridge this gap, we introduce FutureOmni, the first benchmark designed to evaluate omni-modal future forecasting from audio-visual environments. The evaluated models are required to perform cross-modal causal and temporal reasoning, as well as effectively leverage internal knowledge to predict future events. FutureOmni is constructed via a scalable LLM-assisted, human-in-the-loop pipeline and contains 919 videos and 1,034 multiple-choice QA pairs across 8 primary domains. Evaluations on 13 omni-modal and 7 video-only models show that current systems struggle with audio-visual future prediction, particularly in speech-heavy scenarios, with the best accuracy of 64.8% achieved by Gemini 3 Flash. To mitigate this limitation, we curate a 7K-sample instruction-tuning dataset and propose an Omni-Modal Future Forecasting (OFF) training strategy. Evaluations on FutureOmni and popular audio-visual and video-only benchmarks demonstrate that OFF enhances future forecasting and generalization. We publicly release all code (https://github.com/OpenMOSS/FutureOmni) and datasets (https://huggingface.co/datasets/OpenMOSS-Team/FutureOmni).

To address the global issue of hateful content proliferating in online platforms, hate speech detection (HSD) models are typically developed on datasets collected in the United States, thereby failing to generalize to English dialects from the Majority World. Furthermore, HSD models are often evaluated on curated samples, raising concerns about overestimating model performance in real-world settings. In this work, we introduce NaijaHate, the first dataset annotated for HSD which contains a representative sample of Nigerian tweets. We demonstrate that HSD evaluated on biased datasets traditionally used in the literature largely overestimates real-world performance on representative data. We also propose NaijaXLM-T, a pretrained model tailored to the Nigerian Twitter context, and establish the key role played by domain-adaptive pretraining and finetuning in maximizing HSD performance. Finally, we show that in this context, a human-in-the-loop approach to content moderation where humans review 1% of Nigerian tweets flagged as hateful would enable to moderate 60% of all hateful content. Taken together, these results pave the way towards robust HSD systems and a better protection of social media users from hateful content in low-resource settings.

Evaluating aligned large language models' (LLMs) ability to recognize and reject unsafe user requests is crucial for safe, policy-compliant deployments. Existing evaluation efforts, however, face three limitations that we address with SORRY-Bench, our proposed benchmark. First, existing methods often use coarse-grained taxonomies of unsafe topics, and are over-representing some fine-grained topics. For example, among the ten existing datasets that we evaluated, tests for refusals of self-harm instructions are over 3x less represented than tests for fraudulent activities. SORRY-Bench improves on this by using a fine-grained taxonomy of 45 potentially unsafe topics, and 450 class-balanced unsafe instructions, compiled through human-in-the-loop methods. Second, linguistic characteristics and formatting of prompts are often overlooked, like different languages, dialects, and more -- which are only implicitly considered in many evaluations. We supplement SORRY-Bench with 20 diverse linguistic augmentations to systematically examine these effects. Third, existing evaluations rely on large LLMs (e.g., GPT-4) for evaluation, which can be computationally expensive. We investigate design choices for creating a fast, accurate automated safety evaluator. By collecting 7K+ human annotations and conducting a meta-evaluation of diverse LLM-as-a-judge designs, we show that fine-tuned 7B LLMs can achieve accuracy comparable to GPT-4 scale LLMs, with lower computational cost. Putting these together, we evaluate over 40 proprietary and open-source LLMs on SORRY-Bench, analyzing their distinctive refusal behaviors. We hope our effort provides a building block for systematic evaluations of LLMs' safety refusal capabilities, in a balanced, granular, and efficient manner.

The autonomous discovery of bugs remains a significant challenge in modern software development. Compared to code generation, the complexity of dynamic runtime environments makes bug discovery considerably harder for large language models (LLMs). In this paper, we take game development as a representative domain and introduce the Game Benchmark for Quality Assurance (GBQA), a benchmark containing 30 games and 124 human-verified bugs across three difficulty levels, to evaluate whether LLMs can autonomously detect software bugs. The benchmark is constructed using a multi-agent system that develops games and injects bugs in a scalable manner, with human experts in the loop to ensure correctness. Moreover, we provide a baseline interactive agent equipped with a multi-round ReAct loop and a memory mechanism, enabling long-horizon exploration of game environments for bug detection across different LLMs. Extensive experiments on frontier LLMs demonstrate that autonomous bug discovery remains highly challenging: the best-performing model, Claude-4.6-Opus in thinking mode, identifies only 48.39% of the verified bugs. We believe GBQA provides an adequate testbed and evaluation criterion, and that further progress on it will help close the gap in autonomous software engineering.

Evaluation of large language models (LLMs) for code has primarily relied on static benchmarks, including HumanEval (Chen et al., 2021), which measure the ability of LLMs to generate complete code that passes unit tests. As LLMs are increasingly used as programmer assistants, we study whether gains on existing benchmarks translate to gains in programmer productivity when coding with LLMs, including time spent coding. In addition to static benchmarks, we investigate the utility of preference metrics that might be used as proxies to measure LLM helpfulness, such as code acceptance or copy rates. To do so, we introduce RealHumanEval, a web interface to measure the ability of LLMs to assist programmers, through either autocomplete or chat support. We conducted a user study (N=213) using RealHumanEval in which users interacted with six LLMs of varying base model performance. Despite static benchmarks not incorporating humans-in-the-loop, we find that improvements in benchmark performance lead to increased programmer productivity; however gaps in benchmark versus human performance are not proportional -- a trend that holds across both forms of LLM support. In contrast, we find that programmer preferences do not correlate with their actual performance, motivating the need for better, human-centric proxy signals. We also open-source RealHumanEval to enable human-centric evaluation of new models and the study data to facilitate efforts to improve code models.

Human-Scene Interaction (HSI) seeks to generate realistic human behaviors within complex environments, yet it faces significant challenges in handling long-horizon, high-level tasks and generalizing to unseen scenes. To address these limitations, we introduce FantasyHSI, a novel HSI framework centered on video generation and multi-agent systems that operates without paired data. We model the complex interaction process as a dynamic directed graph, upon which we build a collaborative multi-agent system. This system comprises a scene navigator agent for environmental perception and high-level path planning, and a planning agent that decomposes long-horizon goals into atomic actions. Critically, we introduce a critic agent that establishes a closed-loop feedback mechanism by evaluating the deviation between generated actions and the planned path. This allows for the dynamic correction of trajectory drifts caused by the stochasticity of the generative model, thereby ensuring long-term logical consistency. To enhance the physical realism of the generated motions, we leverage Direct Preference Optimization (DPO) to train the action generator, significantly reducing artifacts such as limb distortion and foot-sliding. Extensive experiments on our custom SceneBench benchmark demonstrate that FantasyHSI significantly outperforms existing methods in terms of generalization, long-horizon task completion, and physical realism. Ours project page: https://fantasy-amap.github.io/fantasy-hsi/

This paper presents a conceptual and operational framework for developing and operating safe and trustworthy AI agents based on a Three-Pillar Model grounded in transparency, accountability, and trustworthiness. Building on prior work in Human-in-the-Loop systems, reinforcement learning, and collaborative AI, the framework defines an evolutionary path toward autonomous agents that balances increasing automation with appropriate human oversight. The paper argues that safe agent autonomy must be achieved through progressive validation, analogous to the staged development of autonomous driving, rather than through immediate full automation. Transparency and accountability are identified as foundational requirements for establishing user trust and for mitigating known risks in generative AI systems, including hallucinations, data bias, and goal misalignment, such as the inversion problem. The paper further describes three ongoing work streams supporting this framework: public deliberation on AI agents conducted by the Stanford Deliberative Democracy Lab, cross-industry collaboration through the Safe AI Agent Consortium, and the development of open tooling for an agent operating environment aligned with the Three-Pillar Model. Together, these contributions provide both conceptual clarity and practical guidance for enabling the responsible evolution of AI agents that operate transparently, remain aligned with human values, and sustain societal trust.

Scalar reward models compress multi-dimensional human preferences into a single opaque score, creating an information bottleneck that often leads to brittleness and reward hacking in open-ended alignment. We argue that robust alignment for non-verifiable tasks is fundamentally a principle generalization problem: reward should not be a learned function internalized into a judge, but an explicit reasoning process executed under inspectable principles. To operationalize this view, we present the Open Rubric System (OpenRS), a plug-and-play, rubrics-based LLM-as-a-Judge framework built around Pairwise Adaptive Meta-Rubrics (PAMR) and lightweight Pointwise Verifiable Rubrics (PVRs), which provide both hard-constraint guardrails and verifiable reward components when ground-truth or programmatic checks are available. OpenRS uses an explicit meta-rubric -- a constitution-like specification that governs how rubrics are instantiated, weighted, and enforced -- and instantiates adaptive rubrics on the fly by conditioning on the semantic differences between two candidate responses. It then performs criterion-wise pairwise comparisons and aggregates criterion-level preferences externally, avoiding pointwise weighted scalarization while improving discriminability in open-ended settings. To keep principles consistent yet editable across various domains, we introduce a two-level meta-rubric refinement pipeline (automated evolutionary refinement for general principles and a reproducible human-in-the-loop procedure for domain principles), complemented with pointwise verifiable rubrics that act as both guardrails against degenerate behaviors and a source of verifiable reward for objective sub-tasks. Finally, we instantiate OpenRS as reward supervision in pairwise RL training.

Recent advances in large language models have enabled their application to a range of healthcare tasks. However, aligning LLMs with the nuanced demands of medical ethics, especially under complex real world scenarios, remains underexplored. In this work, we present MedES, a dynamic, scenario-centric benchmark specifically constructed from 260 authoritative Chinese medical, ethical, and legal sources to reflect the challenges in clinical decision-making. To facilitate model alignment, we introduce a guardian-in-the-loop framework that leverages a dedicated automated evaluator (trained on expert-labeled data and achieving over 97% accuracy within our domain) to generate targeted prompts and provide structured ethical feedback. Using this pipeline, we align a 7B-parameter LLM through supervised fine-tuning and domain-specific preference optimization. Experimental results, conducted entirely within the Chinese medical ethics context, demonstrate that our aligned model outperforms notably larger baselines on core ethical tasks, with observed improvements in both quality and composite evaluation metrics. Our work offers a practical and adaptable framework for aligning LLMs with medical ethics in the Chinese healthcare domain, and suggests that similar alignment pipelines may be instantiated in other legal and cultural environments through modular replacement of the underlying normative corpus.

Exposure to large language model output is rapidly increasing. How will seeing AI-generated ideas affect human ideas? We conducted an experiment (800+ participants, 40+ countries) where participants viewed creative ideas that were from ChatGPT or prior experimental participants and then brainstormed their own idea. We varied the number of AI-generated examples (none, low, or high exposure) and if the examples were labeled as 'AI' (disclosure). Our dynamic experiment design -- ideas from prior participants in an experimental condition are used as stimuli for future participants in the same experimental condition -- mimics the interdependent process of cultural creation: creative ideas are built upon prior ideas. Hence, we capture the compounding effects of having LLMs 'in the culture loop'. We find that high AI exposure (but not low AI exposure) did not affect the creativity of individual ideas but did increase the average amount and rate of change of collective idea diversity. AI made ideas different, not better. There were no main effects of disclosure. We also found that self-reported creative people were less influenced by knowing an idea was from AI, and that participants were more likely to knowingly adopt AI ideas when the task was difficult. Our findings suggest that introducing AI ideas into society may increase collective diversity but not individual creativity.

When developing new large language models (LLMs), a key step is evaluating their final performance, often by computing the win-rate against a reference model based on external feedback. Human feedback is the gold standard, particularly for capturing nuanced qualities like coherence, readability, and alignment with human expectations. However, human evaluations are costly -- even for large tech companies -- and when conducted with active users, they may negatively impact user experience. A promising alternative is synthetic feedback, where evaluations are conducted by other large language models, including reward models. While this eliminates the need for costly human annotations, it introduces biases that may distort the evaluation process. In this work, we propose a statistically principled framework that integrates human and synthetic feedback to reduce reliance on human annotations while maintaining unbiased win-rate calculations. Our experiments demonstrate a reduction in human annotations by up to 12.2% with an off-the-shelf synthetic evaluator and up to 24.8% with a finetuned variant. Apart from being generalizable, scalable, and free of hyper-parameter tuning, our method offers predictable annotation savings, which can be estimated based on data-dependent characteristics.

Large Language Models (LLMs) and other automated techniques have been increasingly used to support software developers by generating software artifacts such as code snippets, patches, and comments. However, accurately assessing the correctness of these generated artifacts remains a significant challenge. On one hand, human evaluation provides high accuracy but is labor-intensive and lacks scalability. On the other hand, many automatic evaluation metrics are scalable and require minimal human effort, but they often fail to accurately reflect the actual correctness of generated software artifacts. In this paper, we present SE-Jury, the first evaluation metric for LLM-as-Ensemble-Judge specifically designed to accurately assess the correctness of generated software artifacts. SE-Jury first defines five distinct evaluation strategies, each implemented by an independent judge. A dynamic team selection mechanism then identifies the most appropriate subset of judges as a team to produce a final correctness score through ensembling. We evaluate SE-Jury across a diverse set of software engineering (SE) benchmarks that span three popular SE tasks: code generation, automated program repair, and code summarization. Results demonstrate that SE-Jury consistently achieves a higher correlation with human judgments, with improvements ranging from 29.6% to 140.8% over existing automatic metrics. SE-Jury reaches agreement levels with human annotators that are close to inter-annotator agreement in code generation and program repair. These findings underscore SE-Jury's potential as a scalable and reliable alternative to human evaluation in these SE tasks.

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

To evaluate Large Language Models (LLMs) for question answering (QA), traditional methods typically focus on directly assessing the immediate responses generated by the models based on the given question and context. In the common use case of humans seeking AI assistant's help in finding information, these non-interactive evaluations do not account for the dynamic nature of human-model conversations, and interaction-aware evaluations have shown that accurate QA models are preferred by humans (Lee et al., 2023). Recent works in human-computer interaction (HCI) have employed human evaluators to conduct interactions and evaluations, but they are often prohibitively expensive and time-consuming to scale. In this work, we introduce an automatic evaluation framework IQA-EVAL to Interactive Question Answering Evaluation. More specifically, we introduce LLM-based Evaluation Agent (LEA) that can: (1) simulate human behaviors to generate interactions with IQA models; (2) automatically evaluate the generated interactions. Moreover, we propose assigning personas to LEAs to better simulate groups of real human evaluators. We show that: (1) our evaluation framework with GPT-4 (or Claude) as the backbone model achieves a high correlation with human evaluations on the IQA task; (2) assigning personas to LEA to better represent the crowd further significantly improves correlations. Finally, we use our automatic metric to evaluate five recent representative LLMs with over 1000 questions from complex and ambiguous question answering tasks, which comes with a substantial cost of $5k if evaluated by humans.

Human evaluation is critical for validating the performance of text-to-image generative models, as this highly cognitive process requires deep comprehension of text and images. However, our survey of 37 recent papers reveals that many works rely solely on automatic measures (e.g., FID) or perform poorly described human evaluations that are not reliable or repeatable. This paper proposes a standardized and well-defined human evaluation protocol to facilitate verifiable and reproducible human evaluation in future works. In our pilot data collection, we experimentally show that the current automatic measures are incompatible with human perception in evaluating the performance of the text-to-image generation results. Furthermore, we provide insights for designing human evaluation experiments reliably and conclusively. Finally, we make several resources publicly available to the community to facilitate easy and fast implementations.

As NLP evaluation shifts from static benchmarks to multi-turn interactive settings, LLM-based simulators have become widely used as user proxies, serving two roles: generating user turns and providing evaluation signals. Yet, these simulations are frequently assumed to be faithful to real human behaviors, often without rigorous verification. We formalize the Sim2Real gap in user simulation and present the first study running the full τ-bench protocol with real humans (451 participants, 165 tasks), benchmarking 31 LLM simulators across proprietary, open-source, and specialized families using the User-Sim Index (USI), a metric we introduce to quantify how well LLM simulators resemble real user interactive behaviors and feedback. Behaviorally, LLM simulators are excessively cooperative, stylistically uniform, and lack realistic frustration or ambiguity, creating an "easy mode" that inflates agent success rates above the human baseline. In evaluations, real humans provide nuanced judgments across eight quality dimensions while simulated users produce uniformly more positive feedback; rule-based rewards are failing to capture rich feedback signals generated by human users. Overall, higher general model capability does not necessarily yield more faithful user simulation. These findings highlight the importance of human validation when using LLM-based user simulators in the agent development cycle and motivate improved models for user simulation.

Large language models (LLMs) such as ChatGPT have seen widespread adoption due to their ability to follow user instructions well. Developing these LLMs involves a complex yet poorly understood workflow requiring training with human feedback. Replicating and understanding this instruction-following process faces three major challenges: the high cost of data collection, the lack of trustworthy evaluation, and the absence of reference method implementations. We address these challenges with AlpacaFarm, a simulator that enables research and development for learning from feedback at a low cost. First, we design LLM prompts to simulate human feedback that are 45x cheaper than crowdworkers and display high agreement with humans. Second, we propose an automatic evaluation and validate it against human instructions obtained on real-world interactions. Third, we contribute reference implementations for several methods (PPO, best-of-n, expert iteration, and more) that learn from pairwise feedback. Finally, as an end-to-end validation of AlpacaFarm, we train and evaluate eleven models on 10k pairs of real human feedback and show that rankings of models trained in AlpacaFarm match rankings of models trained on human data. As a demonstration of the research possible in AlpacaFarm, we find that methods that use a reward model can substantially improve over supervised fine-tuning and that our reference PPO implementation leads to a +10% improvement in win-rate against Davinci003. We release all components of AlpacaFarm at https://github.com/tatsu-lab/alpaca_farm.

Large Language Models (LLMs) and code agents in software development are rapidly evolving from generating isolated code snippets to producing full-fledged software applications with graphical interfaces, interactive logic, and dynamic behaviors. However, current benchmarks fall short in evaluating such production-ready software, as they often rely on static checks or binary pass/fail scripts, failing to capture the interactive behaviors and runtime dynamics that define real-world usability - qualities that only emerge when an application is actively used. This is the blind spot of current evaluation: you don't know if an app works until you click through it, interact with it, and observe how it responds. To bridge this gap, we introduce RealDevWorld, a novel evaluation framework for automated end-to-end assessment of LLMs' ability to generate production-ready repositories from scratch. It features two key components: (1) RealDevBench, a diverse collection of 194 open-ended software engineering tasks across multiple domains, incorporating multimodal elements to reflect real-world complexity; and (2) AppEvalPilot, a new agent-as-a-judge evaluation system that simulates realistic, GUI-based user interactions to automatically and holistically assess software functional correctness, visual fidelity, and runtime behavior. The framework delivers fine-grained, task-specific diagnostic feedback, supporting nuanced evaluation beyond simple success/failure judgments. Empirical results show that RealDevWorld delivers effective, automatic, and human-aligned evaluations, achieving an accuracy of 0.92 and a correlation of 0.85 with expert human assessments, while significantly reducing the reliance on manual review. This enables scalable, human-aligned assessment of production-level software generated by LLMs. Our code is available on GitHub.

With the advent of large language models (LLMs) like GPT-3, a natural question is the extent to which these models can be utilized for source code optimization. This paper presents methodologically stringent case studies applied to well-known open source python libraries pillow and numpy. We find that contemporary LLM ChatGPT-4 (state September and October 2023) is surprisingly adept at optimizing energy and compute efficiency. However, this is only the case in interactive use, with a human expert in the loop. Aware of experimenter bias, we document our qualitative approach in detail, and provide transcript and source code. We start by providing a detailed description of our approach in conversing with the LLM to optimize the _getextrema function in the pillow library, and a quantitative evaluation of the performance improvement. To demonstrate qualitative replicability, we report further attempts on another locus in the pillow library, and one code locus in the numpy library, to demonstrate generalization within and beyond a library. In all attempts, the performance improvement is significant (factor up to 38). We have also not omitted reporting of failed attempts (there were none). We conclude that LLMs are a promising tool for code optimization in open source libraries, but that the human expert in the loop is essential for success. Nonetheless, we were surprised by how few iterations were required to achieve substantial performance improvements that were not obvious to the expert in the loop. We would like bring attention to the qualitative nature of this study, more robust quantitative studies would need to introduce a layer of selecting experts in a representative sample -- we invite the community to collaborate.

We introduce Vibe-Eval: a new open benchmark and framework for evaluating multimodal chat models. Vibe-Eval consists of 269 visual understanding prompts, including 100 of hard difficulty, complete with gold-standard responses authored by experts. Vibe-Eval is open-ended and challenging with dual objectives: (i) vibe checking multimodal chat models for day-to-day tasks and (ii) rigorously testing and probing the capabilities of present frontier models. Notably, our hard set contains >50% questions that all frontier models answer incorrectly. We explore the nuances of designing, evaluating, and ranking models on ultra challenging prompts. We also discuss trade-offs between human and automatic evaluation, and show that automatic model evaluation using Reka Core roughly correlates to human judgment. We offer free API access for the purpose of lightweight evaluation and plan to conduct formal human evaluations for public models that perform well on the Vibe-Eval's automatic scores. We release the evaluation code and data, see https://github.com/reka-ai/reka-vibe-eval

Evaluating LLMs is challenging, as benchmark scores often fail to capture models' real-world usefulness. Instead, users often rely on ``vibe-testing'': informal experience-based evaluation, such as comparing models on coding tasks related to their own workflow. While prevalent, vibe-testing is often too ad hoc and unstructured to analyze or reproduce at scale. In this work, we study how vibe-testing works in practice and then formalize it to support systematic analysis. We first analyze two empirical resources: (1) a survey of user evaluation practices, and (2) a collection of in-the-wild model comparison reports from blogs and social media. Based on these resources, we formalize vibe-testing as a two-part process: users personalize both what they test and how they judge responses. We then introduce a proof-of-concept evaluation pipeline that follows this formulation by generating personalized prompts and comparing model outputs using user-aware subjective criteria. In experiments on coding benchmarks, we find that combining personalized prompts and user-aware evaluation can change which model is preferred, reflecting the role of vibe-testing in practice. These findings suggest that formalized vibe-testing can serve as a useful approach for bridging benchmark scores and real-world experience.

How can we design Natural Language Processing (NLP) systems that learn from human feedback? There is a growing research body of Human-in-the-loop (HITL) NLP frameworks that continuously integrate human feedback to improve the model itself. HITL NLP research is nascent but multifarious -- solving various NLP problems, collecting diverse feedback from different people, and applying different methods to learn from collected feedback. We present a survey of HITL NLP work from both Machine Learning (ML) and Human-Computer Interaction (HCI) communities that highlights its short yet inspiring history, and thoroughly summarize recent frameworks focusing on their tasks, goals, human interactions, and feedback learning methods. Finally, we discuss future directions for integrating human feedback in the NLP development loop.

Human evaluation is the gold-standard for evaluating text generation models. It is also expensive, and to fit budgetary constraints, a random subset of the test data is often chosen in practice. The randomly selected data may not accurately represent test performance, making this approach economically inefficient for model comparison. Thus, in this work, we develop a suite of selectors to get the most informative datapoints for human evaluation while taking the evaluation costs into account. We show that selectors based on variance in automated metric scores, diversity in model outputs, or Item Response Theory outperform random selection. We further develop an approach to distill these selectors to the scenario where the model outputs are not yet available. In particular, we introduce source-based estimators, which predict item usefulness for human evaluation just based on the source texts. We demonstrate the efficacy of our selectors in two common NLG tasks, machine translation and summarization, and show that up to only ~50% of the test data is needed to produce the same evaluation result as the entire data. Our implementations are published in the subset2evaluate package.

Template mining is one of the foundational tasks to support log analysis, which supports the diagnosis and troubleshooting of large scale Web applications. This paper develops a human-in-the-loop template mining framework to support interactive log analysis, which is highly desirable in real-world diagnosis or troubleshooting of Web applications but yet previous template mining algorithms fails to support it. We formulate three types of light-weight user feedbacks and based on them we design three atomic human-in-the-loop template mining algorithms. We derive mild conditions under which the outputs of our proposed algorithms are provably correct. We also derive upper bounds on the computational complexity and query complexity of each algorithm. We demonstrate the versatility of our proposed algorithms by combining them to improve the template mining accuracy of five representative algorithms over sixteen widely used benchmark datasets.

This paper introduces LegalRikai: Open Benchmark, a new benchmark comprising four complex tasks that emulate Japanese corporate legal practices. The benchmark was created by legal professionals under the supervision of an attorney. This benchmark has 100 samples that require long-form, structured outputs, and we evaluated them against multiple practical criteria. We conducted both human and automated evaluations using leading LLMs, including GPT-5, Gemini 2.5 Pro, and Claude Opus 4.1. Our human evaluation revealed that abstract instructions prompted unnecessary modifications, highlighting model weaknesses in document-level editing that were missed by conventional short-text tasks. Furthermore, our analysis reveals that automated evaluation aligns well with human judgment on criteria with clear linguistic grounding, and assessing structural consistency remains a challenge. The result demonstrates the utility of automated evaluation as a screening tool when expert availability is limited. We propose a dataset evaluation framework to promote more practice-oriented research in the legal domain.

While LLM-Based agents, which use external tools to solve complex problems, have made significant progress, benchmarking their ability is challenging, thereby hindering a clear understanding of their limitations. In this paper, we propose an interactive evaluation framework, named CIBench, to comprehensively assess LLMs' ability to utilize code interpreters for data science tasks. Our evaluation framework includes an evaluation dataset and two evaluation modes. The evaluation dataset is constructed using an LLM-human cooperative approach and simulates an authentic workflow by leveraging consecutive and interactive IPython sessions. The two evaluation modes assess LLMs' ability with and without human assistance. We conduct extensive experiments to analyze the ability of 24 LLMs on CIBench and provide valuable insights for future LLMs in code interpreter utilization.

Language models are increasingly used in scientific discovery to generate hypotheses, propose candidate solutions, implement systems, and iteratively refine them. At the core of these trial-and-error loops lies evaluation: the process of obtaining feedback on candidate solutions via verifiers, simulators, or task-specific scoring functions. While prior work has highlighted the importance of evaluation, it has not explicitly formulated the problem of how evaluation-driven discovery loops can be scaled up in a principled and effective manner to push the boundaries of scientific discovery, a problem this paper seeks to address. We introduce Simple Test-time Evaluation-driven Scaling (SimpleTES), a general framework that strategically combines parallel exploration, feedback-driven refinement, and local selection, revealing substantial gains unlocked by scaling evaluation-driven discovery loops along the right dimensions. Across 21 scientific problems spanning six domains, SimpleTES discovers state-of-the-art solutions using gpt-oss models, consistently outperforming both frontier-model baselines and sophisticated optimization pipelines. Particularly, we sped up the widely used LASSO algorithm by over 2x, designed quantum circuit routing policies that reduce gate overhead by 24.5%, and discovered new Erdos minimum overlap constructions that surpass the best-known results. Beyond novel discoveries, SimpleTES produces trajectory-level histories that naturally supervise feedback-driven learning. When post-trained on successful trajectories, models not only improve efficiency on seen problems but also generalize to unseen problems, discovering solutions that base models fail to uncover. Together, our results establish effective evaluation-driven loop scaling as a central axis for advancing LLM-driven scientific discovery, and provide a simple yet practical framework for realizing these gains.

As large language models (LLMs) continue to advance, accurately and comprehensively evaluating their performance becomes increasingly challenging. Human evaluations are conventionally considered the gold standard in natural language generation, but recent advancements incorporate state-of-the-art LLMs as proxies for human judges in evaluation processes. However, the extent to which humans and LLMs are capable evaluators remains uncertain. This study investigates the behavior of crowd-sourced and expert annotators, as well as LLMs, when comparing outputs from different models. To achieve this, we curate a dataset of intentionally flawed machine-generated answers. Our findings reveal a concerning bias in the evaluation process, as answers with factual errors are rated more favorably than answers that are too short or contained grammatical errors. To address this issue, we propose independently evaluating machine-generated text across multiple dimensions, rather than merging all the evaluation aspects into a single score. We instantiate this idea with the Elo rating system, resulting in the Multi-Elo Rating System. Empirical results from our study reveal that this proposed approach significantly enhances the quality of LLM-based evaluations, particularly in terms of factual accuracy. However, there is no significant improvement in crowd-sourced-based evaluations, indicating the need for further investigation and refinement.

The increasing adoption of agentic workflows across diverse domains brings a critical need to scalably and systematically evaluate the complex traces these systems generate. Current evaluation methods depend on manual, domain-specific human analysis of lengthy workflow traces - an approach that does not scale with the growing complexity and volume of agentic outputs. Error analysis in these settings is further complicated by the interplay of external tool outputs and language model reasoning, making it more challenging than traditional software debugging. In this work, we (1) articulate the need for robust and dynamic evaluation methods for agentic workflow traces, (2) introduce a formal taxonomy of error types encountered in agentic systems, and (3) present a set of 148 large human-annotated traces (TRAIL) constructed using this taxonomy and grounded in established agentic benchmarks. To ensure ecological validity, we curate traces from both single and multi-agent systems, focusing on real-world applications such as software engineering and open-world information retrieval. Our evaluations reveal that modern long context LLMs perform poorly at trace debugging, with the best Gemini-2.5-pro model scoring a mere 11% on TRAIL. Our dataset and code are made publicly available to support and accelerate future research in scalable evaluation for agentic workflows.

We present the workflow of Online Iterative Reinforcement Learning from Human Feedback (RLHF) in this technical report, which is widely reported to outperform its offline counterpart by a large margin in the recent large language model (LLM) literature. However, existing open-source RLHF projects are still largely confined to the offline learning setting. In this technical report, we aim to fill in this gap and provide a detailed recipe that is easy to reproduce for online iterative RLHF. In particular, since online human feedback is usually infeasible for open-source communities with limited resources, we start by constructing preference models using a diverse set of open-source datasets and use the constructed proxy preference model to approximate human feedback. Then, we discuss the theoretical insights and algorithmic principles behind online iterative RLHF, followed by a detailed practical implementation. Our trained LLM, SFR-Iterative-DPO-LLaMA-3-8B-R, achieves impressive performance on LLM chatbot benchmarks, including AlpacaEval-2, Arena-Hard, and MT-Bench, as well as other academic benchmarks such as HumanEval and TruthfulQA. We have shown that supervised fine-tuning (SFT) and iterative RLHF can obtain state-of-the-art performance with fully open-source datasets. Further, we have made our models, curated datasets, and comprehensive step-by-step code guidebooks publicly available. Please refer to https://github.com/RLHFlow/RLHF-Reward-Modeling and https://github.com/RLHFlow/Online-RLHF for more detailed information.

Retriever-augmented instruction-following models are attractive alternatives to fine-tuned approaches for information-seeking tasks such as question answering (QA). By simply prepending retrieved documents in its input along with an instruction, these models can be adapted to various information domains and tasks without additional fine-tuning. While the model responses tend to be natural and fluent, the additional verbosity makes traditional QA evaluation metrics such as exact match (EM) and F1 unreliable for accurately quantifying model performance. In this work, we investigate the performance of instruction-following models across three information-seeking QA tasks. We use both automatic and human evaluation to evaluate these models along two dimensions: 1) how well they satisfy the user's information need (correctness), and 2) whether they produce a response based on the provided knowledge (faithfulness). Guided by human evaluation and analysis, we highlight the shortcomings of traditional metrics for both correctness and faithfulness. We then propose simple token-overlap based and model-based metrics that reflect the true performance of these models. Our analysis reveals that instruction-following models are competitive, and sometimes even outperform fine-tuned models for correctness. However, these models struggle to stick to the provided knowledge and often hallucinate in their responses. We hope our work encourages a more holistic evaluation of instruction-following models for QA. Our code and data is available at https://github.com/McGill-NLP/instruct-qa

Among the hardest tasks for humans are those found in competitive programming where problems require sophisticated algorithmic thinking, puzzle solving, and the creation of effective code. As a domain to assess language models (LMs), it has not received enough attention, though. This study presents the ICPC benchmark, which consists of 254 international collegiate programming contest (ICPC) tasks. Each problem includes official analysis, reference code, and sample, high-quality unit, and hidden tests. We are able to develop and evaluate a variety of LM inference techniques for competitive programming with these resources. With zero-shot chain-of-thought prompting, we find that o1 only achieves a 19.1\% pass@1 solve rate. With our best inference technique, which combines multi-turn self-judge with reflection and retrieval over episodic information, raises this to 42.2\%. Furthermore, we conduct a new human-in-the-loop investigation to gain a deeper understanding of the remaining difficulties. Surprisingly, we discover that o1 can solve 17 out of 18 problems that were previously unsolvable by any model or technique with just a few specific instructions. A footstep toward LMs with grounded, imaginative, and algorithmic thinking is provided by our quantitative findings and qualitative research. We open-source our code and data at https://github.com/kraritt/zolve.

One core capability of Large Language Models (LLMs) is to follow natural language instructions. However, the evaluation of such abilities is not standardized: Human evaluations are expensive, slow, and not objectively reproducible, while LLM-based auto-evaluation is potentially biased or limited by the ability of the evaluator LLM. To overcome these issues, we introduce Instruction-Following Eval (IFEval) for large language models. IFEval is a straightforward and easy-to-reproduce evaluation benchmark. It focuses on a set of "verifiable instructions" such as "write in more than 400 words" and "mention the keyword of AI at least 3 times". We identified 25 types of those verifiable instructions and constructed around 500 prompts, with each prompt containing one or more verifiable instructions. We show evaluation results of two widely available LLMs on the market. Our code and data can be found at https://github.com/google-research/google-research/tree/master/instruction_following_eval

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

Reliable evaluation of large language models (LLMs) is impeded by two key challenges: objective metrics often fail to reflect human perception of natural language, and exhaustive human labeling is prohibitively expensive. Here, we propose a sample-efficient human evaluation method for LLMs based on the principle of MAximum Discrepancy (MAD) Competition. Our method automatically and adaptively selects a compact set of input instructions that maximize semantic discrepancy between pairs of LLM responses. Human evaluators then perform three-alternative forced choices on these paired responses, which are aggregated into a global ranking using Elo rating. We apply our approach to compare eight widely used LLMs across four tasks: scientific knowledge understanding, mathematical reasoning, creative and functional writing, and code generation and explanation. Experimental results show that our sample-efficient evaluation method recovers "gold-standard" model rankings with a handful of MAD-selected instructions, reveals respective strengths and weaknesses of each LLM, and offers nuanced insights to guide future LLM development. Code is available at https://github.com/weiji-Feng/MAD-Eval .

Analyzing Open Source Intelligence (OSINT) from large volumes of data is critical for drafting and publishing comprehensive CTI reports. This process usually follows a three-stage workflow -- triage, deep search and TI drafting. While Large Language Models (LLMs) offer a promising route toward automation, existing benchmarks still have limitations. These benchmarks often consist of tasks that do not reflect real-world analyst workflows. For example, human analysts rarely receive tasks in the form of multiple-choice questions. Also, existing benchmarks often rely on model-centric metrics that emphasize lexical overlap rather than actionable, detailed insights essential for security analysts. Moreover, they typically fail to cover the complete three-stage workflow. To address these issues, we introduce CyberThreat-Eval, which is collected from the daily CTI workflow of a world-leading company. This expert-annotated benchmark assesses LLMs on practical tasks across all three stages as mentioned above. It utilizes analyst-centric metrics that measure factual accuracy, content quality, and operational costs. Our evaluation using this benchmark reveals important insights into the limitations of current LLMs. For example, LLMs often lack the nuanced expertise required to handle complex details and struggle to distinguish between correct and incorrect information. To address these challenges, the CTI workflow incorporates both external ground-truth databases and human expert knowledge. TRA allows human experts to iteratively provide feedback for continuous improvement. The code is available at https://github.com/xschen-beb/CyberThreat-Eval{GitHub} and https://huggingface.co/datasets/xse/CyberThreat-Eval{HuggingFace}.

Human evaluation is the gold standard for multilingual NLP, but is often skipped in practice and substituted with automatic metrics, because it is notoriously complex and slow to set up with existing tools with substantial engineering and operational overhead. We introduce Pearmut, a lightweight yet feature-rich platform that makes end-to-end human evaluation as easy to run as automatic evaluation. Pearmut removes common entry barriers and provides support for evaluating multilingual tasks, with a particular focus on machine translation. The platform implements standard evaluation protocols, including DA, ESA, or MQM, but is also extensible to allow prototyping new protocols. It features document-level context, absolute and contrastive evaluation, attention checks, ESAAI pre-annotations and both static and active learning-based assignment strategies. Pearmut enables reliable human evaluation to become a practical, routine component of model development and diagnosis rather than an occasional effort.

Deep neural networks have demonstrated promising performance on image recognition tasks. However, they may heavily rely on confounding factors, using irrelevant artifacts or bias within the dataset as the cue to improve performance. When a model performs decision-making based on these spurious correlations, it can become untrustable and lead to catastrophic outcomes when deployed in the real-world scene. In this paper, we explore and try to solve this problem in the context of skin cancer diagnosis. We introduce a human-in-the-loop framework in the model training process such that users can observe and correct the model's decision logic when confounding behaviors happen. Specifically, our method can automatically discover confounding factors by analyzing the co-occurrence behavior of the samples. It is capable of learning confounding concepts using easily obtained concept exemplars. By mapping the black-box model's feature representation onto an explainable concept space, human users can interpret the concept and intervene via first order-logic instruction. We systematically evaluate our method on our newly crafted, well-controlled skin lesion dataset and several public skin lesion datasets. Experiments show that our method can effectively detect and remove confounding factors from datasets without any prior knowledge about the category distribution and does not require fully annotated concept labels. We also show that our method enables the model to focus on clinical-related concepts, improving the model's performance and trustworthiness during model inference.

Large language models (LLMs) are increasingly used in interactive applications, and human evaluation remains the gold standard for assessing their performance in multi-turn conversations. Since human studies are costly, time-consuming, and hard to reproduce, recent work explores using LLMs to simulate users for automatic assistant evaluation. However, there is no benchmark or systematic study to evaluate whether these simulated users are reliable stand-ins for real users. To address this, we introduce SimulatorArena, a benchmark of 909 annotated human-LLM conversations on two interactive tasks -- math tutoring and document creation. SimulatorArena evaluates simulators based on how closely their messages match human behavior and how well their assistant ratings align with human judgments. Experiments on various simulator methods show that simulators conditioned on user profiles, capturing traits like background and message styles, align closely with human judgments. They reach Spearman's rho of 0.7 on both tasks, providing a practical, scalable alternative to human evaluation. Using the best simulator for each task, we benchmark 18 assistants, including the latest LLMs such as GPT-5, Claude 4.1 Opus, and Gemini 2.5 Pro.

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

Following the rapid progress in natural language processing (NLP) models, language models are applied to increasingly more complex interactive tasks such as negotiations and conversation moderations. Having human evaluators directly interact with these NLP models is essential for adequately evaluating the performance on such interactive tasks. We develop BotEval, an easily customizable, open-source, evaluation toolkit that focuses on enabling human-bot interactions as part of the evaluation process, as opposed to human evaluators making judgements for a static input. BotEval balances flexibility for customization and user-friendliness by providing templates for common use cases that span various degrees of complexity and built-in compatibility with popular crowdsourcing platforms. We showcase the numerous useful features of BotEval through a study that evaluates the performance of various chatbots on their effectiveness for conversational moderation and discuss how BotEval differs from other annotation tools.

The paradigm of LLM-as-a-judge is emerging as a scalable and efficient alternative to human evaluation, demonstrating strong performance on well-defined tasks. However, its reliability in open-ended tasks with dynamic environments and complex interactions remains unexplored. To bridge the gap, we introduce WebDevJudge, a systematic benchmark for assessing LLM-as-a-judge performance in web development, with support for both non-interactive evaluation based on static observations and continuous interactive evaluation with a dynamic web environment. WebDevJudge comprises human preference labels over paired web implementations, annotated with structured and query-grounded rubrics to ensure high-quality ground truth. Using this benchmark, we comprehensively evaluate various evaluators, including LLMs, MLLMs, and agentic workflows. We systematically investigate the impact of different paradigms and guidance mechanisms. Our experiments reveal a significant gap between LLM judges and human experts. In-depth analysis indicates this gap stems from fundamental model limitations, including failures in recognizing functional equivalence, verifying task feasibility, and mitigating bias. Overall, WebDevJudge presents a significant challenge to LLM-as-a-judge, offering insights to guide future research toward developing more reliable and capable automated evaluators for complicated scenarios. Code and data are available at https://github.com/lcy2723/WebDevJudge.

Model-based evaluation is at the heart of successful model development -- as a reward model for training, and as a replacement for human evaluation. To train such evaluators, the standard approach is to collect a large amount of human preference judgments over model responses, which is costly and the data becomes stale as models improve. In this work, we present an approach that aims to im-prove evaluators without human annotations, using synthetic training data only. Starting from unlabeled instructions, our iterative self-improvement scheme generates contrasting model outputs and trains an LLM-as-a-Judge to produce reasoning traces and final judgments, repeating this training at each new iteration using the improved predictions. Without any labeled preference data, our Self-Taught Evaluator can improve a strong LLM (Llama3-70B-Instruct) from 75.4 to 88.3 (88.7 with majority vote) on RewardBench. This outperforms commonly used LLM judges such as GPT-4 and matches the performance of the top-performing reward models trained with labeled examples.

Large language models that exhibit instruction-following behaviour represent one of the biggest recent upheavals in conversational interfaces, a trend in large part fuelled by the release of OpenAI's ChatGPT, a proprietary large language model for text generation fine-tuned through reinforcement learning from human feedback (LLM+RLHF). We review the risks of relying on proprietary software and survey the first crop of open-source projects of comparable architecture and functionality. The main contribution of this paper is to show that openness is differentiated, and to offer scientific documentation of degrees of openness in this fast-moving field. We evaluate projects in terms of openness of code, training data, model weights, RLHF data, licensing, scientific documentation, and access methods. We find that while there is a fast-growing list of projects billing themselves as 'open source', many inherit undocumented data of dubious legality, few share the all-important instruction-tuning (a key site where human annotation labour is involved), and careful scientific documentation is exceedingly rare. Degrees of openness are relevant to fairness and accountability at all points, from data collection and curation to model architecture, and from training and fine-tuning to release and deployment.

Natural language (NL) feedback contains rich information about the user experience. Existing studies focus on an instance-level approach, where feedback is used to refine specific examples, disregarding its system-wide application. This paper proposes a general framework for unlocking the system-level use of NL feedback. We show how to use feedback to formalize system-level design decisions in a human-in-the-loop-process -- in order to produce better models. In particular this is done through: (i) metric design for tasks; and (ii) language model prompt design for refining model responses. We conduct two case studies of this approach for improving search query generation and dialog response generation, demonstrating the effectiveness of the use of system-level feedback. We show the combination of system-level feedback and instance-level feedback brings further gains, and that human written instance-level feedback results in more grounded refinements than GPT-3.5 written ones, underlying the importance of human feedback for building systems.

Language models (LMs) as conversational assistants recently became popular tools that help people accomplish a variety of tasks. These typically result from adapting LMs pretrained on general domain text sequences through further instruction-tuning and possibly preference optimisation methods. The evaluation of such LMs would ideally be performed using human judgement, however, this is not scalable. On the other hand, automatic evaluation featuring auxiliary LMs as judges and/or knowledge-based tasks is scalable but struggles with assessing conversational ability and adherence to instructions. To help accelerate the development of LMs as conversational assistants, we propose a novel automatic evaluation task: HumanRankEval (HRE). It consists of a large-scale, diverse and high-quality set of questions, each with several answers authored and scored by humans. To perform evaluation, HRE ranks these answers based on their log-likelihood under the LM's distribution, and subsequently calculates their correlation with the corresponding human rankings. We support HRE's efficacy by investigating how efficiently it separates pretrained and instruction-tuned LMs of various sizes. We show that HRE correlates well with human judgements and is particularly responsive to model changes following instruction-tuning.

Humans are integral components of the transportation ecosystem, and understanding their behaviors is crucial to facilitating the development of safe driving systems. Although recent progress has explored various aspects of human behaviorx2014such as motion, trajectories, and intentionx2014a comprehensive benchmark for evaluating human behavior understanding in autonomous driving remains unavailable. In this work, we propose MMHU, a large-scale benchmark for human behavior analysis featuring rich annotations, such as human motion and trajectories, text description for human motions, human intention, and critical behavior labels relevant to driving safety. Our dataset encompasses 57k human motion clips and 1.73M frames gathered from diverse sources, including established driving datasets such as Waymo, in-the-wild videos from YouTube, and self-collected data. A human-in-the-loop annotation pipeline is developed to generate rich behavior captions. We provide a thorough dataset analysis and benchmark multiple tasksx2014ranging from motion prediction to motion generation and human behavior question answeringx2014thereby offering a broad evaluation suite. Project page : https://MMHU-Benchmark.github.io.

Multimodal systems have great potential to assist humans in procedural activities, where people follow instructions to achieve their goals. Despite diverse application scenarios, systems are typically evaluated on traditional classification tasks, e.g., action recognition or temporal action segmentation. In this paper, we present a novel evaluation dataset, ProMQA, to measure system advancements in application-oriented scenarios. ProMQA consists of 401 multimodal procedural QA pairs on user recording of procedural activities coupled with their corresponding instruction. For QA annotation, we take a cost-effective human-LLM collaborative approach, where the existing annotation is augmented with LLM-generated QA pairs that are later verified by humans. We then provide the benchmark results to set the baseline performance on ProMQA. Our experiment reveals a significant gap between human performance and that of current systems, including competitive proprietary multimodal models. We hope our dataset sheds light on new aspects of models' multimodal understanding capabilities.

Retrieval-Augmented Generation (RAG) has advanced significantly in recent years. The complexity of RAG systems, which involve multiple components-such as indexing, retrieval, and generation-along with numerous other parameters, poses substantial challenges for systematic evaluation and quality enhancement. Previous research highlights that evaluating RAG systems is essential for documenting advancements, comparing configurations, and identifying effective approaches for domain-specific applications. This study systematically reviews 63 academic articles to provide a comprehensive overview of state-of-the-art RAG evaluation methodologies, focusing on four key areas: datasets, retrievers, indexing and databases, and the generator component. We observe the feasibility of an automated evaluation approach for each component of a RAG system, leveraging an LLM capable of both generating evaluation datasets and conducting evaluations. In addition, we found that further practical research is essential to provide companies with clear guidance on the do's and don'ts of implementing and evaluating RAG systems. By synthesizing evaluation approaches for key RAG components and emphasizing the creation and adaptation of domain-specific datasets for benchmarking, we contribute to the advancement of systematic evaluation methods and the improvement of evaluation rigor for RAG systems. Furthermore, by examining the interplay between automated approaches leveraging LLMs and human judgment, we contribute to the ongoing discourse on balancing automation and human input, clarifying their respective contributions, limitations, and challenges in achieving robust and reliable evaluations.

Large language models (LLMs) combined with retrieval augmented generation have enabled the deployment of domain-specific chatbots, but these systems remain prone to generating unsupported or incorrect answers. Reliable evaluation is therefore critical, yet manual review is costly and existing frameworks often depend on curated test sets and static metrics, limiting scalability. We propose an end-to-end automatic evaluator designed to substantially reduce human effort. Our system generates Q\&A pairs directly from the underlying knowledge base, uses LLMs to judge chatbot responses against reference answers, and applies confidence-based filtering to highlight uncertain cases. Applied to a Vietnamese news dataset, the evaluator achieves high agreement with human judgments while significantly lowering review overhead. The framework is modular and language-agnostic, making it readily adaptable to diverse domains. This work introduces a practical, scalable solution for evaluating chatbots with minimal reliance on manual intervention.

To solve complex tasks, large language models (LLMs) often require multiple rounds of interactions with the user, sometimes assisted by external tools. However, current evaluation protocols often emphasize benchmark performance with single-turn exchanges, neglecting the nuanced interactions among the user, LLMs, and external tools, while also underestimating the importance of natural language feedback from users. These oversights contribute to discrepancies between research benchmark evaluations and real-world use cases. We introduce MINT, a benchmark that evaluates LLMs' ability to solve tasks with multi-turn interactions by (1) using tools and (2) leveraging natural language feedback. To ensure reproducibility, we provide an evaluation framework where LLMs can access tools by executing Python code and receive users' natural language feedback simulated by GPT-4. We repurpose a diverse set of established evaluation datasets focusing on reasoning, coding, and decision-making and carefully curate them into a compact subset for efficient evaluation. Our analysis of 20 open- and closed-source LLMs offers intriguing findings. (a) LLMs generally benefit from tools and language feedback, with performance gains (absolute, same below) of 1-8% for each turn of tool use and 2-17% with natural language feedback. (b) Better single-turn performance does not guarantee better multi-turn performance. (c) Surprisingly, on the LLMs evaluated, supervised instruction-finetuning (SIFT) and reinforcement learning from human feedback (RLHF) generally hurt multi-turn capabilities. We expect MINT can help measure progress and incentivize research in improving LLMs' capabilities in multi-turn interactions, especially for open-source communities where multi-turn human evaluation can be less accessible compared to commercial LLMs with a larger user base.

While Multimodal Large Language Models (MLLMs) show immense promise for achieving truly human-like interactions, progress is hindered by the lack of fine-grained evaluation frameworks for human-centered scenarios, encompassing both the understanding of complex human intentions and the provision of empathetic, context-aware responses. Here we introduce HumanSense, a comprehensive benchmark designed to evaluate the human-centered perception and interaction capabilities of MLLMs, with a particular focus on deep understanding of extended multimodal contexts and the formulation of rational feedback. Our evaluation reveals that leading MLLMs still have considerable room for improvement, particularly for advanced interaction-oriented tasks. Supplementing visual input with audio and text information yields substantial improvements, and Omni-modal models show advantages on these tasks. Furthermore, we argue that appropriate feedback stems from a contextual analysis of the interlocutor's needs and emotions, with reasoning ability serving as the key to unlocking it. Accordingly, we employ a multi-stage, modality-progressive reinforcement learning to enhance the reasoning abilities of an Omni model, achieving substantial gains on evaluation results. Additionally, we observe that successful reasoning processes exhibit highly consistent thought patterns. By designing corresponding prompts, we also enhance the performance of non-reasoning models in a training-free manner. Project page: brightpinkhttps://digital-avatar.github.io/ai/HumanSense/

Large Language Models (LLMs) achieve strong performance on diverse tasks but often exhibit cognitive inertia, struggling to follow instructions that conflict with the standardized patterns learned during supervised fine-tuning (SFT). To evaluate this limitation, we propose Inverse IFEval, a benchmark that measures models Counter-intuitive Abilitytheir capacity to override training-induced biases and comply with adversarial instructions. Inverse IFEval introduces eight types of such challenges, including Question Correction, Intentional Textual Flaws, Code without Comments, and Counterfactual Answering. Using a human-in-the-loop pipeline, we construct a dataset of 1012 high-quality Chinese and English questions across 23 domains, evaluated under an optimized LLM-as-a-Judge framework. Experiments on existing leading LLMs demonstrate the necessity of our proposed Inverse IFEval benchmark. Our findings emphasize that future alignment efforts should not only pursue fluency and factual correctness but also account for adaptability under unconventional contexts. We hope that Inverse IFEval serves as both a diagnostic tool and a foundation for developing methods that mitigate cognitive inertia, reduce overfitting to narrow patterns, and ultimately enhance the instruction-following reliability of LLMs in diverse and unpredictable real-world scenarios.

Human evaluation is the foundation upon which the evaluation of both summarization systems and automatic metrics rests. However, existing human evaluation studies for summarization either exhibit a low inter-annotator agreement or have insufficient scale, and an in-depth analysis of human evaluation is lacking. Therefore, we address the shortcomings of existing summarization evaluation along the following axes: (1) We propose a modified summarization salience protocol, Atomic Content Units (ACUs), which is based on fine-grained semantic units and allows for a high inter-annotator agreement. (2) We curate the Robust Summarization Evaluation (RoSE) benchmark, a large human evaluation dataset consisting of 22,000 summary-level annotations over 28 top-performing systems on three datasets. (3) We conduct a comparative study of four human evaluation protocols, underscoring potential confounding factors in evaluation setups. (4) We evaluate 50 automatic metrics and their variants using the collected human annotations across evaluation protocols and demonstrate how our benchmark leads to more statistically stable and significant results. The metrics we benchmarked include recent methods based on large language models (LLMs), GPTScore and G-Eval. Furthermore, our findings have important implications for evaluating LLMs, as we show that LLMs adjusted by human feedback (e.g., GPT-3.5) may overfit unconstrained human evaluation, which is affected by the annotators' prior, input-agnostic preferences, calling for more robust, targeted evaluation methods.

The evaluation of machine learning models using human-labeled validation data can be expensive and time-consuming. AI-labeled synthetic data can be used to decrease the number of human annotations required for this purpose in a process called autoevaluation. We suggest efficient and statistically principled algorithms for this purpose that improve sample efficiency while remaining unbiased. These algorithms increase the effective human-labeled sample size by up to 50% on experiments with GPT-4.

Recent benchmark studies have claimed that AI has approached or even surpassed human-level performances on various cognitive tasks. However, this position paper argues that current AI evaluation paradigms are insufficient for assessing human-like cognitive capabilities. We identify a set of key shortcomings: a lack of human-validated labels, inadequate representation of human response variability and uncertainty, and reliance on simplified and ecologically-invalid tasks. We support our claims by conducting a human evaluation study on ten existing AI benchmarks, suggesting significant biases and flaws in task and label designs. To address these limitations, we propose five concrete recommendations for developing future benchmarks that will enable more rigorous and meaningful evaluations of human-like cognitive capacities in AI with various implications for such AI applications.

A recent focus of large language model (LLM) development, as exemplified by generative search engines, is to incorporate external references to generate and support their claims. However, evaluating the attribution, i.e., verifying whether the generated statement is indeed fully supported by the cited reference, remains an open problem. Although human evaluation is common practice, it is costly and time-consuming. In this paper, we investigate the automatic evaluation of attribution by LLMs. We begin by providing a definition of attribution and then explore two approaches for automatic evaluation: prompting LLMs and fine-tuning smaller LMs. The fine-tuning data is repurposed from related tasks, such as question answering, fact-checking, natural language inference, and summarization. To facilitate the evaluation, we manually curate a set of test examples covering 12 domains from a generative search engine, New Bing. Our results on the curated test set and simulated test examples from existing benchmark questions highlight both promising signals as well as remaining challenges for the automatic evaluation of attribution. We hope our testbed, modeling methodology, and insights will help lay the foundation for future studies on this important problem.

Crowdsourced model evaluation platforms, such as Chatbot Arena, enable real-time evaluation from human perspectives to assess the quality of model responses. In the coding domain, manually examining the quality of LLM-generated content is extremely challenging, as it requires understanding long chunks of raw code and deliberately simulating code execution. To this end, we introduce BigCodeArena, an open human evaluation platform for code generation backed by a comprehensive and on-the-fly execution environment. Built on top of Chatbot Arena, BigCodeArena enables the execution of LLM-generated code and allows humans to interact with the execution process and outcomes. We collected over 14,000 raw code-centric conversation sessions across 10 widely used LLMs, spanning 10 languages and 8 types of execution environments. Among these conversations, we identified more than 4,700 multi-turn samples with pairwise human preferences. Further analysis uncovers underexplored preferences of LLMs in fine-grained domains characterized by tasks, languages, and frameworks. To systematically examine code understanding and generation capabilities of frontier LLMs, we curated two benchmarks based on the collected data, namely BigCodeReward and AutoCodeArena. For BigCodeReward, we post-processed the 4,700 conversations and evaluated the consistency between reward models and human preferences. The evaluation shows that most LLMs have superior performance in judging coding preferences when the execution results are available. Inspired by these findings, we propose AutoCodeArena, an automatic Elo rating benchmark designed to assess the coding quality of LLMs without human involvement. We find that proprietary LLMs like GPT-5, Claude-Sonnet-4, and Claude-Opus-4 still lead in code generation performance among recent emerging models.

High-quality training triplets (source-target image pairs with precise editing instructions) are a critical bottleneck for scaling instruction-guided image editing models. Vision-language models (VLMs) are widely used for automated instruction synthesis, but we identify three systematic failure modes in image-pair settings: orientation inconsistency (e.g., left/right confusion), viewpoint ambiguity, and insufficient fine-grained attribute description. Human evaluation shows that over 47% of instructions from strong baseline VLMs contain critical errors unusable for downstream training. We propose EditCaption, a scalable two-stage post-training pipeline for VLM-based instruction synthesis. Stage 1 builds a 100K supervised fine-tuning (SFT) dataset by combining GLM automatic annotation, EditScore-based filtering, and human refinement for spatial, directional, and attribute-level accuracy. Stage 2 collects 10K human preference pairs targeting the three failure modes and applies direct preference optimization (DPO) for alignment beyond SFT alone. On Eval-400, ByteMorph-Bench, and HQ-Edit, fine-tuned Qwen3-VL models outperform open-source baselines; the 235B model reaches 4.712 on Eval-400 (vs. Gemini-3-Pro 4.706, GPT-4.1 4.220, Kimi-K2.5 4.111) and 4.588 on ByteMorph-Bench (vs. Gemini-3-Pro 4.522, GPT-4.1 3.412). Human evaluation shows critical errors falling from 47.75% to 23% and correctness rising from 41.75% to 66%. The work offers a practical path to scalable, human-aligned instruction synthesis for image editing data.

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

Language models are capable of iteratively improving their outputs based on natural language feedback, thus enabling in-context optimization of user preference. In place of human users, a second language model can be used as an evaluator, providing feedback along with numerical ratings which the generator attempts to optimize. However, because the evaluator is an imperfect proxy of user preference, this optimization can lead to reward hacking, where the evaluator's ratings improve while the generation quality remains stagnant or even decreases as judged by actual user preference. The concern of reward hacking is heightened in iterative self-refinement where the generator and the evaluator use the same underlying language model, in which case the optimization pressure can drive them to exploit shared vulnerabilities. Using an essay editing task, we show that iterative self-refinement leads to deviation between the language model evaluator and human judgment, demonstrating that reward hacking can occur spontaneously in-context with the use of iterative self-refinement. In addition, we study conditions under which reward hacking occurs and observe two factors that affect reward hacking severity: model size and context sharing between the generator and the evaluator.

Assistants on assembly tasks have a large potential to benefit humans from everyday tasks to industrial settings. However, no testbeds support application-oriented system evaluation in a practical setting, especially in assembly. To foster the development, we propose a new multimodal QA dataset on assembly activities. Our dataset, ProMQA-Assembly, consists of 391 QA pairs that require the multimodal understanding of human-activity recordings and their instruction manuals in an online-style manner. In the development, we adopt a semi-automated QA annotation approach, where LLMs generate candidates and humans verify them, as a cost-effective method, and further improve it by integrating fine-grained action labels to diversify question types. Furthermore, we create instruction task graphs for the target tasks of assembling toy vehicles. These newly created task graphs are used in our benchmarking experiment, as well as to facilitate the human verification process in the QA annotation. Utilizing our dataset, we benchmark models, including competitive proprietary multimodal models. Our results suggest great room for improvement for the current models. We believe our new evaluation dataset can contribute to the further development of procedural-activity assistants.

Large language models (LLMs) are increasingly used as simulated participants in social science experiments, but their behavior is often unstable and highly sensitive to design choices. Prior evaluations frequently conflate base-model capabilities with experimental instantiation, obscuring whether outcomes reflect the model itself or the agent setup. We instead frame participant simulation as an agent-design problem over full experimental protocols, where an agent is defined by a base model and a specification (e.g., participant attributes) that encodes behavioral assumptions. We introduce HUMANSTUDY-BENCH, a benchmark and execution engine that orchestrates LLM-based agents to reconstruct published human-subject experiments via a Filter--Extract--Execute--Evaluate pipeline, replaying trial sequences and running the original analysis pipeline in a shared runtime that preserves the original statistical procedures end to end. To evaluate fidelity at the level of scientific inference, we propose new metrics to quantify how much human and agent behaviors agree. We instantiate 12 foundational studies as an initial suite in this dynamic benchmark, spanning individual cognition, strategic interaction, and social psychology, and covering more than 6,000 trials with human samples ranging from tens to over 2,100 participants.

We introduce iTeach, a human-in-the-loop Mixed Reality (MR) system that enhances robot perception through interactive teaching. Our system enables users to visualize robot perception outputs such as object detection and segmentation results using a MR device. Therefore, users can inspect failures of perception models using the system on real robots. Moreover, iTeach facilitates real-time, informed data collection and annotation, where users can use hand gesture, eye gaze and voice commands to annotate images collected from robots. The annotated images can be used to fine-tune perception models to improve their accuracy and adaptability. The system continually improves perception models by collecting annotations of failed examples from users. When applied to object detection and unseen object instance segmentation (UOIS) tasks, iTeach demonstrates encouraging results in improving pre-trained vision models for these two tasks. These results highlight the potential of MR to make robotic perception systems more capable and adaptive in real-world environments. Project page at https://irvlutd.github.io/iTeach.

In this position paper, we argue that human evaluation of generative large language models (LLMs) should be a multidisciplinary undertaking that draws upon insights from disciplines such as user experience research and human behavioral psychology to ensure that the experimental design and results are reliable. The conclusions from these evaluations, thus, must consider factors such as usability, aesthetics, and cognitive biases. We highlight how cognitive biases can conflate fluent information and truthfulness, and how cognitive uncertainty affects the reliability of rating scores such as Likert. Furthermore, the evaluation should differentiate the capabilities and weaknesses of increasingly powerful large language models -- which requires effective test sets. The scalability of human evaluation is also crucial to wider adoption. Hence, to design an effective human evaluation system in the age of generative NLP, we propose the ConSiDERS-The-Human evaluation framework consisting of 6 pillars -- Consistency, Scoring Criteria, Differentiating, User Experience, Responsible, and Scalability.

Automatically generated questions often suffer from problems such as unclear expression or factual inaccuracies, requiring a reliable and comprehensive evaluation of their quality. Human evaluation is frequently used in the field of question generation (QG) and is one of the most accurate evaluation methods. It also serves as the standard for automatic metrics. However, there is a lack of unified evaluation criteria, which hampers the development of both QG technologies and automatic evaluation methods. To address this, we propose QGEval, a multi-dimensional Evaluation benchmark for Question Generation, which evaluates both generated questions and existing automatic metrics across 7 dimensions: fluency, clarity, conciseness, relevance, consistency, answerability, and answer consistency. We demonstrate the appropriateness of these dimensions by examining their correlations and distinctions. Analysis with QGEval reveals that 1) most QG models perform unsatisfactorily in terms of answerability and answer consistency, and 2) existing metrics fail to align well with human assessments when evaluating generated questions across the 7 dimensions. We expect this work to foster the development of both QG technologies and automatic metrics for QG.

Frontier coding agents solve complex tasks when given complete context but collapse when specifications are incomplete or ambiguous. The bottleneck is not raw capability, but judgment: knowing when to act autonomously and when to ask for help. Current benchmarks are blind to this failure mode. They supply unambiguous detailed instructions and solely reward execution correctness, so an agent that makes a lucky guess for a missing requirement will score identically to one that would have asked to be certain. We present HiL-Bench (Human-in-the-Loop Benchmark) to measure this selective escalation skill. Each task contains human-validated blockers (missing information, ambiguous requests, contradictory information) that surface only through progressive exploration, not upfront inspection. Our core metric, Ask-F1, the harmonic mean of question precision and blocker recall, captures the tension between over-asking and silent guessing; its structure architecturally prevents gaming through question spam. Evaluation across SWE and text-to-SQL domains reveals a large universal judgment gap: no frontier model recovers more than a fraction of its full-information performance when deciding whether to ask. Failure analysis identifies three key help-seeking patterns: overconfident wrong beliefs with no gap detection; high uncertainty detection yet persistent errors; broad, imprecise escalation without self-correction. These consistent patterns confirm poor help-seeking is a model-level flaw, not task-specific. RL training on shaped Ask-F1 reward shows judgment is trainable: a 32B model improves both help-seeking quality and task pass rate, with gains that transfer across domains. The model does not learn domain-specific heuristics for when to ask; it learns to detect unresolvable uncertainty and act on it.

Large language models (LLMs) have emerged as a promising alternative to expensive human evaluations. However, the alignment and coverage of LLM-based evaluations are often limited by the scope and potential bias of the evaluation prompts and criteria. To address this challenge, we propose HD-Eval, a novel framework that iteratively aligns LLM-based evaluators with human preference via Hierarchical Criteria Decomposition. HD-Eval inherits the essence from the evaluation mindset of human experts and enhances the alignment of LLM-based evaluators by decomposing a given evaluation task into finer-grained criteria, aggregating them according to estimated human preferences, pruning insignificant criteria with attribution, and further decomposing significant criteria. By integrating these steps within an iterative alignment training process, we obtain a hierarchical decomposition of criteria that comprehensively captures aspects of natural language at multiple levels of granularity. Implemented as a white box, the human preference-guided aggregator is efficient to train and more explainable than relying solely on prompting, and its independence from model parameters makes it applicable to closed-source LLMs. Extensive experiments on three evaluation domains demonstrate the superiority of HD-Eval in further aligning state-of-the-art evaluators and providing deeper insights into the explanation of evaluation results and the task itself.

Benchmarks are essential for consistent evaluation and reproducibility. The integration of Artificial Intelligence into Software Engineering (AI4SE) has given rise to numerous benchmarks for tasks such as code generation and bug fixing. However, this surge presents challenges: (1) scattered benchmark knowledge across tasks, (2) difficulty in selecting relevant benchmarks, (3) the absence of a uniform standard for benchmark development, and (4) limitations of existing benchmarks. In this paper, we review 173 studies and identify 204 AI4SE benchmarks. We classify these benchmarks, analyze their limitations, and expose gaps in practices. Based on our review, we created BenchScout, a semantic search tool to find relevant benchmarks, using automated clustering of the contexts from associated studies. We conducted a user study with 22 participants to evaluate BenchScout's usability, effectiveness, and intuitiveness which resulted in average scores of 4.5, 4.0, and 4.1 out of 5. To advance benchmarking standards, we propose BenchFrame, a unified method to enhance benchmark quality. As a case study, we applied BenchFrame to the HumanEval benchmark and addressed its main limitations. This led to HumanEvalNext, featuring (1) corrected errors, (2) improved language conversion, (3) expanded test coverage, and (4) increased difficulty. We then evaluated ten state-of-the-art code language models on HumanEval, HumanEvalPlus, and HumanEvalNext. On HumanEvalNext, models showed a pass@1 score reduction of 31.22% and 19.94% compared to HumanEval and HumanEvalPlus, respectively.

While current Computer Use Agent (CUA) benchmarks measure task completion effectively, they provide limited assessment of enterprise deployment readiness, emphasizing functional correctness over the operational reliability required for production systems. We present UI-CUBE (UiPath Computer Use BEnchmark), a systematic benchmark comprising 226 tasks across two difficulty tiers designed to expose fundamental architectural limitations in current CUAs. Our evaluation covers simple UI interactions (136 tasks) and complex workflows including copy-paste tasks (50 tasks) and enterprise application scenarios (40 tasks), with systematic interface variation coverage, multi-resolution testing and automated validation of task success through the application state. Evaluation of five state-of-the-art models reveals a sharp capability cliff rather than gradual performance degradation. Simple UI interactions achieve 67-85% success rates (compared to 97.9% human performance), but complex workflows drop precipitously to 9-19%. Human evaluators with no prior application experience achieve only 61.2% on complex tasks despite near-perfect performance on simple tasks, establishing realistic performance ceilings. This discontinuous performance pattern -- where agents achieve 68-87% of human performance on simple tasks but only 15-32% on complex workflows -- indicates fundamental architectural limitations in memory management, hierarchical planning, and state coordination rather than incremental capability gaps addressable through better training or prompting. UI-CUBE functions as an enterprise-readiness diagnostic, revealing that while current CUAs can manipulate individual interface elements, they cannot yet function as reliable workflow automation tools. These findings provide architectural insights essential for developing production-ready CUAs capable of managing complex, multi-step enterprise processes.

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

The assessment of evaluation metrics (meta-evaluation) is crucial for determining the suitability of existing metrics in text-to-image (T2I) generation tasks. Human-based meta-evaluation is costly and time-intensive, and automated alternatives are scarce. We address this gap and propose CROC: a scalable framework for automated Contrastive Robustness Checks that systematically probes and quantifies metric robustness by synthesizing contrastive test cases across a comprehensive taxonomy of image properties. With CROC, we generate a pseudo-labeled dataset (CROC^{syn}) of over one million contrastive prompt-image pairs to enable a fine-grained comparison of evaluation metrics. We also use the dataset to train CROCScore, a new metric that achieves state-of-the-art performance among open-source methods, demonstrating an additional key application of our framework. To complement this dataset, we introduce a human-supervised benchmark (CROC^{hum}) targeting especially challenging categories. Our results highlight robustness issues in existing metrics: for example, many fail on prompts involving negation, and all tested open-source metrics fail on at least 25% of cases involving correct identification of body parts.

Large Language Models (LLMs) with agentic web search capabilities show strong potential for tasks requiring real-time information access and complex fact retrieval, yet evaluating such systems remains challenging. We introduce \bench, a rigorous and regularly updated benchmark designed to assess the agentic web search abilities of LLMs. \bench automatically generates fresh question-answer pairs from recent news articles, ensuring that questions require information beyond an LLM's training data and enabling clear separation between internal knowledge and search capability. The benchmark features intentionally difficult questions requiring multi-hop search queries, page visits, and reasoning, making it well-suited for evaluating agentic search behavior. Our automated data curation and question generation pipeline enables frequent benchmark updates and supports construction of a large-scale training dataset for agentic web search models, addressing the scarcity of such data in the research community. To ensure reliable evaluation, we include a subset of human-verified samples in the test set. We evaluate a broad range of systems using \bench, including commercial and open-weight LLMs as well as LLM-based web search APIs. The leaderboard, datasets, and code are publicly available at livenewsbench.com.

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

Evaluating Natural Language Generation (NLG) is crucial for the practical adoption of AI, but has been a longstanding research challenge. While human evaluation is considered the de-facto standard, it is expensive and lacks scalability. Practical applications have driven the development of various automatic evaluation metrics (AEM), designed to compare the model output with human-written references, generating a score which approximates human judgment. Over time, AEMs have evolved from simple lexical comparisons, to semantic similarity models and, more recently, to LLM-based evaluators. However, it seems that no single metric has emerged as a definitive solution, resulting in studies using different ones without fully considering the implications. This paper aims to show this by conducting a thorough examination of the methodologies of existing metrics, their documented strengths and limitations, validation methods, and correlations with human judgment. We identify several key challenges: metrics often capture only specific aspects of text quality, their effectiveness varies by task and dataset, validation practices remain unstructured, and correlations with human judgment are inconsistent. Importantly, we find that these challenges persist in the most recent type of metric, LLM-as-a-Judge, as well as in the evaluation of Retrieval Augmented Generation (RAG), an increasingly relevant task in academia and industry. Our findings challenge the quest for the 'perfect metric'. We propose selecting metrics based on task-specific needs and leveraging complementary evaluations and advocate that new metrics should focus on enhanced validation methodologies.

Language models can distinguish between testing and deployment phases -- a capability known as evaluation awareness. This has significant safety and policy implications, potentially undermining the reliability of evaluations that are central to AI governance frameworks and voluntary industry commitments. In this paper, we study evaluation awareness in Llama-3.3-70B-Instruct. We show that linear probes can separate real-world evaluation and deployment prompts, suggesting that current models internally represent this distinction. We also find that current safety evaluations are correctly classified by the probes, suggesting that they already appear artificial or inauthentic to models. Our findings underscore the importance of ensuring trustworthy evaluations and understanding deceptive capabilities. More broadly, our work showcases how model internals may be leveraged to support blackbox methods in safety audits, especially for future models more competent at evaluation awareness and deception.

LLM-as-a-Judge models generate chain-of-thought (CoT) sequences intended to capture the step-bystep reasoning process that underlies the final evaluation of a response. However, due to the lack of human annotated CoTs for evaluation, the required components and structure of effective reasoning traces remain understudied. Consequently, previous approaches often (1) constrain reasoning traces to hand-designed components, such as a list of criteria, reference answers, or verification questions and (2) structure them such that planning is intertwined with the reasoning for evaluation. In this work, we propose EvalPlanner, a preference optimization algorithm for Thinking-LLM-as-a-Judge that first generates an unconstrained evaluation plan, followed by its execution, and then the final judgment. In a self-training loop, EvalPlanner iteratively optimizes over synthetically constructed evaluation plans and executions, leading to better final verdicts. Our method achieves a new state-of-the-art performance for generative reward models on RewardBench (with a score of 93.9), despite being trained on fewer amount of, and synthetically generated, preference pairs. Additional experiments on other benchmarks like RM-Bench, JudgeBench, and FollowBenchEval further highlight the utility of both planning and reasoning for building robust LLM-as-a-Judge reasoning models.

Interactive Natural Language Processing (iNLP) has emerged as a novel paradigm within the field of NLP, aimed at addressing limitations in existing frameworks while aligning with the ultimate goals of artificial intelligence. This paradigm considers language models as agents capable of observing, acting, and receiving feedback iteratively from external entities. Specifically, language models in this context can: (1) interact with humans for better understanding and addressing user needs, personalizing responses, aligning with human values, and improving the overall user experience; (2) interact with knowledge bases for enriching language representations with factual knowledge, enhancing the contextual relevance of responses, and dynamically leveraging external information to generate more accurate and informed responses; (3) interact with models and tools for effectively decomposing and addressing complex tasks, leveraging specialized expertise for specific subtasks, and fostering the simulation of social behaviors; and (4) interact with environments for learning grounded representations of language, and effectively tackling embodied tasks such as reasoning, planning, and decision-making in response to environmental observations. This paper offers a comprehensive survey of iNLP, starting by proposing a unified definition and framework of the concept. We then provide a systematic classification of iNLP, dissecting its various components, including interactive objects, interaction interfaces, and interaction methods. We proceed to delve into the evaluation methodologies used in the field, explore its diverse applications, scrutinize its ethical and safety issues, and discuss prospective research directions. This survey serves as an entry point for researchers who are interested in this rapidly evolving area and offers a broad view of the current landscape and future trajectory of iNLP.

Recent large language models often answer factual questions correctly. But users can't trust any given claim a model makes without fact-checking, because language models can hallucinate convincing nonsense. In this work we use reinforcement learning from human preferences (RLHP) to train "open-book" QA models that generate answers whilst also citing specific evidence for their claims, which aids in the appraisal of correctness. Supporting evidence is drawn from multiple documents found via a search engine, or from a single user-provided document. Our 280 billion parameter model, GopherCite, is able to produce answers with high quality supporting evidence and abstain from answering when unsure. We measure the performance of GopherCite by conducting human evaluation of answers to questions in a subset of the NaturalQuestions and ELI5 datasets. The model's response is found to be high-quality 80\% of the time on this Natural Questions subset, and 67\% of the time on the ELI5 subset. Abstaining from the third of questions for which it is most unsure improves performance to 90\% and 80\% respectively, approaching human baselines. However, analysis on the adversarial TruthfulQA dataset shows why citation is only one part of an overall strategy for safety and trustworthiness: not all claims supported by evidence are true.