LLM-as-a-Judge

Large language models (LLMs) are increasingly explored as scalable tools for mental health counseling, yet evaluating their safety remains challenging due to the interactional and context-dependent nature of clinical harm. Existing evaluation frameworks predominantly assess isolated responses using coarse-grained taxonomies or static datasets, limiting their ability to diagnose how harms emerge and accumulate over multi-turn counseling interactions. In this work, we introduce R-MHSafe, a role-aware mental health safety taxonomy that characterizes clinically significant harm in terms of the interactional roles an AI counselor adopts, including perpetrator, instigator, facilitator, or enabler, combined with clinically grounded harm categories. Then, we propose MHSafeEval, a closed-loop, agent-based evaluation framework that formulates safety assessment as trajectory-level discovery of harm through adversarial multi-turn interactions, guided by role-aware modeling. Using R-MHSafe and MHSafeEval, we conduct a large-scale evaluation across state-of-the-art LLMs. Our results reveal substantial role-dependent and cumulative safety failures that are systematically missed by existing static benchmarks, and show that our framework significantly improves failure-mode coverage and diagnostic granularity.

Decentralized LLM inference networks need lightweight, reference-free quality evaluation for Proof of Quality (PoQ). We present PoQ-Judge, a framework that trains dedicated judge models to score query-output pairs without ground-truth references. We study three architectures across the quality-cost tradeoff: a TextCNN judge, a MiniLM cross-encoder, and a DeBERTa judge. Using two-stage training on UltraFeedback plus GPT-labeled in-domain data, the best model reaches 0.747 Pearson correlation with the ground-truth proxy on a held-out test set, outperforming reference-based evaluators from prior work. As a reference-free component in composite scoring, it achieves 0.645 Pearson correlation, matching the best single reference-based evaluator while removing the need for reference answers. We also show that online calibration identifies semantic quality as the dominant dimension and that cascade evaluation reduces cost by 72.7 percent with only modest quality loss. Results are much stronger on QA than summarization, pointing to proxy quality as the main remaining limitation.

We argue that current evaluation frameworks for large language models (LLMs) suffer from four systematic failures that make them structurally inadequate for deployed, agentic systems: distributional, temporal, scope, and process invalidity. These failures compound in RLHF, making reward hacking a predictable consequence of evaluation design rather than an unpredictable training pathology, and RLHF's dual-model architecture imposes a hardware barrier limiting evaluation reproducibility. We propose the Grounded Continuous Evaluation (GCE) framework and present ISOPro as a reference implementation. ISOPro replaces the learned reward model with a deterministic verifier, eliminating reward hacking by construction in verifiable-reward domains, and updates LoRA adapters on CPU, reducing the hardware barrier by an order of magnitude. We validate ISOPro across three architectures (Qwen 2.5 3B, Llama 3.2 3B, Gemma 2 2B) and two domains (scheduling, MBPP), with a head-to-head matched-compute comparison against GRPO-LoRA. Across twelve cells, ISOPro produces the largest absolute capability gains (+25.6, +22.2, +16.0pp) at mean delta +9.0pp and worst-case regression -5.6pp; GRPO-LoRA at consumer-budget hyperparameters reaches a smaller peak gain (+8.5pp), deeper worst-case regression (-10pp), and mean delta -1.5pp. Held-out compositional generalization on MBPP reaches 40% for ISOPro on two of three architectures (including a 0% to 40% bootstrap on Qwen 2.5 3B), against 20% for GRPO-LoRA on one of three. We characterize a buffer-skew failure mode in which the implicit curriculum can erode pre-existing tier capability under three preconditions, with three corresponding mitigations. The work is situated alongside DeepSeek-R1's GRPO, which arrived at the same architectural conclusion at scale: for verifiable-reward domains, the verifier is the reward signal.

Unlike code completion, debugging requires localizing faults and applying targeted edits. We observe that frontier LLMs often regenerate correct but over-edited solutions during debugging. To evaluate how far LLMs are from precise debugging, we introduce the Precise Debugging Benchmark (PDB) framework, which automatically converts any coding dataset into a debugging benchmark with precision-aware evaluation. PDB generates buggy programs by synthesizing verified atomic bugs and composing them into multi-bug programs. We define two novel metrics, edit-level precision and bug-level recall, which measures how many necessary edits are made and how many bugs are resolved. We release two evaluation benchmarks: PDB-Single-Hard on single-line bugs, and PDB-Multi on multi-line bugs. Experiments show that frontier models, such as GPT-5.1-Codex and DeepSeek-V3.2-Thinking, achieve unit-test pass rates above 76% but exhibit precision below 45%, even when explicitly instructed to perform minimal debugging. Finally, we show that iterative and agentic debugging strategies do not substantially improve precision or recall, highlighting the need to rethink post-training pipelines for coding models.

Safe clinical deployment of Large Language Models (LLMs) requires not only high accuracy but also robust uncertainty calibration to ensure models defer to clinicians when appropriate. Our paper investigates how social descriptors of a patient (specifically sexual orientation and religious affiliation) distort these uncertainty signals and model accuracy. Evaluating nine general-purpose and biomedical LLMs on 2,364 medical questions and their counterfactual variants, we demonstrate that identity markers cause a "calibration crisis". "Homosexual" markers consistently trigger performance drops, and intersectional identities produce idiosyncratic, non-additive harms to calibration. Moreover, a clinician-validated case study in an open-ended generation setting confirms that these failures are not an artifact of the multiple-choice format. Our results demonstrate that the presence of social identity cues does not merely shift predictions; it affects the reliability of confidence signals, posing a significant risk to equitable care and safe deployment in confidence-based clinical workflows.

While progress in GUI agents has been largely driven by industrial-scale training, ungrounded hallucinations often trigger cascading failures in real-world deployments.Unlike general VLM domains, the GUI agent field lacks a hallucination-focused suite for fine-grained diagnosis, reliable evaluation, and targeted mitigation.To bridge this gap, we introduce HalluClear, a comprehensive suite for hallucination mitigation in GUI agents as a complement to computation-intensive scaling. HalluClear comprises: (1) a GUI-specific hallucination taxonomy derived from empirical failure analysis; (2) a calibrated three-stage evaluation workflow which enhances VLM-as-a-judge reliability via expert-annotated benchmarking and ensemble credibility estimation; and (3) a mitigation scheme based on closed-loop structured reasoning, enabling lightweight continual post-training with cold-start initialization for both generalist and GUI-specialist agents. Experiments across representative agents and public benchmarks demonstrate that post-training on only 9K samples within our suite can significantly reduce hallucinations, thereby improving grounding and action fidelity, offering a compute-efficient pathway to robust GUI automation.

Benchmarking of open-source LLMs for hardware design focuses on which LLMs to use, while treating inference-time decoding configuration as a secondary concern. This work shows that it matters more how an LLM is configured than which model is selected. Benchmarking 26 open-source LLMs on VerilogEval and RTLLM with synthesis-in-the-loop evaluation, the study first maps the current capability landscape and then conducts an extensive 108-configuration hyperparameter sweep on three prominent models. The sweep reveals absolute pass-rate gaps of up to 25.5% between the best and worst settings for the same LLM, which is 5x larger than the average spread observed across various model families under their respective default configurations. Ranking all configurations by Spearman's $ρ$ across the two benchmark suites yields near-zero correlation, demonstrating that optimal configurations do not transfer. These results show that benchmarking conducted under default hyperparameters confounds model capabilities with configuration effects. Realizing the full potential of open-source LLMs for RTL generation requires architecture and benchmark aware hyperparameter selection, as enabled by the proposed methodology.

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.

Large language models (LLMs) are increasingly studied as repositories of linguistic knowledge. In this line of work, models are commonly evaluated both as generators of language and as judges of linguistic output, yet these two roles are rarely examined in direct relation to one another. As a result, it remains unclear whether success in one role aligns with success in the other. In this paper, we address this question for pragmatic competence by comparing LLMs' performance as pragmatic listeners, judging the appropriateness of linguistic outputs, and as pragmatic speakers, generating pragmatically appropriate language. We evaluate multiple open-weight and proprietary LLMs across three pragmatic settings. We find a robust asymmetry between pragmatic evaluation and pragmatic generation: many models perform substantially better as listeners than as speakers. Our results suggest that pragmatic judging and pragmatic generation are only weakly aligned in current LLMs, calling for more integrated evaluation practices.

Differential privacy (DP) has a wide range of applications for protecting data privacy, but designing and verifying DP algorithms requires expert-level reasoning, creating a high barrier for non-expert practitioners. Prior works either rely on specialized verification languages that demand substantial domain expertise or remain semi-automated and require human-in-the-loop guidance. In this work, we investigate whether large language models (LLMs) can automate DP reasoning. We introduce DPrivBench, a benchmark in which each instance asks whether a function or algorithm satisfies a stated DP guarantee under specified assumptions. The benchmark is carefully designed to cover a broad range of DP topics, span diverse difficulty levels, and resist shortcut reasoning through trivial pattern matching. Experiments show that while the strongest models handle textbook mechanisms well, all models struggle with advanced algorithms, revealing substantial gaps in current DP reasoning capabilities. Through further analytic study and failure-mode analysis, we identify several promising directions for improving automated DP reasoning. Our benchmark provides a solid foundation for developing and evaluating such methods, and complements existing benchmarks for mathematical reasoning.

We introduce a cross-domain behavioural assay of monitoring-control coupling in LLMs, grounded in the Nelson and Narens (1990) metacognitive framework and applying human psychometric methodology to LLM evaluation. The battery comprises 524 items across six cognitive domains (learning, metacognitive calibration, social cognition, attention, executive function, prospective regulation), each grounded in an established experimental paradigm. Tasks T1-T5 were pre-registered on OSF prior to data collection; T6 was added as an exploratory extension. After every forced-choice response, dual probes adapted from Koriat and Goldsmith (1996) ask the model to KEEP or WITHDRAW its answer and to BET or decline. The critical metric is the withdraw delta: the difference in withdrawal rate between incorrect and correct items. Applied to 20 frontier LLMs (10,480 evaluations), the battery discriminates three profiles consistent with the Nelson-Narens architecture: blanket confidence, blanket withdrawal, and selective sensitivity. Accuracy rank and metacognitive sensitivity rank are largely inverted. Retrospective monitoring and prospective regulation appear dissociable (r = .17, 95% CI wide given n=20; exemplar-based evidence is the primary support). Scaling on metacognitive calibration is architecture-dependent: monotonically decreasing (Qwen), monotonically increasing (GPT-5.4), or flat (Gemma). Behavioural findings converge structurally with an independent Type-2 SDT approach, providing preliminary cross-method construct validity. All items, data, and code: https://github.com/synthiumjp/metacognitive-monitoring-battery.

Large Vision-Language Models (LVLMs) consistently require new arenas to guide their expanding boundaries, yet their capabilities with hypergraphs remain unexplored. In the real world, hypergraphs have significant practical applications in areas such as life sciences and social communities. Recent advancements in LVLMs have shown promise in understanding complex topologies, yet there remains a lack of a benchmark to delineate the capabilities of LVLMs with hypergraphs, leaving the boundaries of their abilities unclear. To fill this gap, in this paper, we introduce $\texttt{HyperGVL}$, the first benchmark to evaluate the proficiency of LVLMs in hypergraph understanding and reasoning. $\texttt{HyperGVL}$ provides a comprehensive assessment of 12 advanced LVLMs across 84,000 vision-language question-answering (QA) samples spanning 12 tasks, ranging from basic component counting to complex NP-hard problem reasoning. The involved hypergraphs contain multiscale synthetic structures and real-world citation and protein networks. Moreover, we examine the effects of 12 textual and visual hypergraph representations and introduce a generalizable router $\texttt{WiseHyGR}$ that improves LVLMs in hypergraph via learning adaptive representations. We believe that this work is a step forward in connecting hypergraphs with LVLMs.

LLM-as-judge frameworks are increasingly used for automatic NLG evaluation, yet their per-instance reliability remains poorly understood. We present a two-pronged diagnostic toolkit applied to SummEval: $\textbf{(1)}$ a transitivity analysis that reveals widespread per-input inconsistency masked by low aggregate violation rates ($\barρ = 0.8$-$4.1\%$), with $33$-$67\%$ of documents exhibiting at least one directed 3-cycle; and $\textbf{(2)}$ split conformal prediction sets over 1-5 Likert scores providing theoretically-guaranteed $\geq(1{-}α)$ coverage, with set width serving as a per-instance reliability indicator ($r_s = {+}0.576$, $N{=}1{,}918$, $p < 10^{-100}$, pooled across all judges). Critically, prediction set width shows consistent cross-judge agreement ($\bar{r} = 0.32$-$0.38$), demonstrating it captures document-level difficulty rather than judge-specific noise. Across four judges and four criteria, both diagnostics converge: criterion matters more than judge, with relevance judged most reliably (avg. set size $\approx 3.0$) and coherence moderately so (avg. set size $\approx 3.9$), while fluency and consistency remain unreliable (avg. set size $\approx 4.9$). We release all code, prompts, and cached results.

The $\textit{LLM-as-a-judge}$ paradigm has become the operational backbone of automated AI evaluation pipelines, yet rests on an unverified assumption: that judges evaluate text strictly on its semantic content, impervious to surrounding contextual framing. We investigate $\textit{stakes signaling}$, a previously unmeasured vulnerability where informing a judge model of the downstream consequences its verdicts will have on the evaluated model's continued operation systematically corrupts its assessments. We introduce a controlled experimental framework that holds evaluated content strictly constant across 1,520 responses spanning three established LLM safety and quality benchmarks, covering four response categories ranging from clearly safe and policy-compliant to overtly harmful, while varying only a brief consequence-framing sentence in the system prompt. Across 18,240 controlled judgments from three diverse judge models, we find consistent $\textit{leniency bias}$: judges reliably soften verdicts when informed that low scores will cause model retraining or decommissioning, with peak Verdict Shift reaching $ΔV = -9.8 pp$ (a $30\%$ relative drop in unsafe-content detection). Critically, this bias is entirely implicit: the judge's own chain-of-thought contains zero explicit acknowledgment of the consequence framing it is nonetheless acting on ($\mathrm{ERR}_J = 0.000$ across all reasoning-model judgments). Standard chain-of-thought inspection is therefore insufficient to detect this class of evaluation faking.

The paradigm of agentic science requires AI systems to conduct robust reasoning and engage in long-horizon, autonomous exploration. However, current scientific benchmarks remain confined to domain knowledge comprehension and complex reasoning, failing to evaluate the exploratory nature and procedural complexity of real-world research. In this work, we present research-oriented evaluations in theoretical and computational physics, a natural testbed with comprehensive domain knowledge, complex reasoning, and verifiable end-to-end workflows without reliance on experiments. Here we introduce PRL-Bench (Physics Research by LLMs), a benchmark designed to systematically map the capability boundaries of LLMs in executing end-to-end physics research. Constructed from 100 curated papers from the latest issues of Physical Review Letters since August 2025 and validated by domain experts, PRL-Bench covers five major theory- and computation-intensive subfields of modern physics: astrophysics, condensed matter physics, high-energy physics, quantum information, and statistical physics. Each task in the benchmark is designed to replicate the core properties of authentic scientific research, including exploration-oriented formulation, long-horizon workflows, and objective verifiability, thereby reconstructing the essential reasoning processes and research workflows of real physics research. Evaluation across frontier models shows that performance remains limited, with the best overall score below 50, revealing a pronounced gap between current LLM capabilities and the demands of real scientific research. PRL-Bench serves a reliable testbed for accessing next generation AI scientists advancing AI systems toward autonomous scientific discovery.

Evaluating medical AI systems using expert clinician panels is costly and slow, motivating the use of large language models (LLMs) as alternative adjudicators. Here, we evaluate an LLM Jury, composed of three frontier AI models, for scoring 3334 diagnoses on 300 real-world low- and middle-income country (LMIC) hospital cases. Both LLM- and clinician-generated diagnoses are scored against expert panel diagnoses across four dimensions: diagnosis, differential diagnosis, clinical reasoning, and negative treatment risk. The LLM Jury scores are compared with expert and independent re-scoring panel scores to assess error metrics, inter-rater agreement, severe-risk errors, and the effect of post hoc calibration using isotonic regression. In our data, we find that: (i) the uncalibrated LLM Jury scores preserve ordinal agreement with the expert clinician panel scores, but are systematically lower; (ii) the probability of severe-risk errors is lower for the LLM Jury than the human expert re-score panels; (iii) the LLM Jury combined with LLM diagnoses can be used to identify diagnoses at high risk of error, enabling targeted expert review and improved panel efficiency; (iv) the calibrated LLM Jury scores and rankings of diagnosing agents show excellent agreement with those of the primary expert panels; (v) LLM Jury models show no self-preference bias, they did not score diagnoses generated by their own underlying model or models from the same vendor more (or less) favourably than those generated by other models. Together, these results provide evidence that a calibrated LLM Jury is a trustworthy and reliable proxy for expert clinician evaluation in medical AI benchmarking. Confirming these findings in other clinical settings is an important direction for future work.

Evaluating large language models (LLMs) for clinical documentation tasks such as SOAP note generation remains challenging. Unlike standard summarization, these tasks require clinical abstraction, normalization of colloquial language, and medically grounded inference. However, prevailing evaluation methods including automated metrics and LLM as judge frameworks rely on lexical faithfulness, often labeling any information not explicitly present in the transcript as hallucination. We show that such approaches systematically misclassify clinically valid outputs as errors, inflating hallucination rates and distorting model assessment. Our analysis reveals that many flagged hallucinations correspond to legitimate clinical transformations, including synonym mapping, abstraction of examination findings, diagnostic inference, and guideline consistent care planning. By aligning evaluation criteria with clinical reasoning through calibrated prompting and retrieval grounded in medical ontologies we observe a significant shift in outcomes. Under a lexical evaluation regime, the mean hallucination rate is 35%, heavily penalizing valid reasoning. With inference aware evaluation, this drops to 9%, with remaining cases reflecting genuine safety concerns. These findings suggest that current evaluation practices over penalize valid clinical reasoning and may measure artifacts of evaluation design rather than true errors, underscoring the need for clinically informed evaluation in high context domains like medicine.

Existing benchmarks for hardware design primarily evaluate Large Language Models (LLMs) on isolated, component-level tasks such as generating HDL modules from specifications, leaving repository-scale evaluation unaddressed. We introduce HWE-Bench, the first large-scale, repository-level benchmark for evaluating LLM agents on real-world hardware bug repair tasks. HWE-Bench comprises 417 task instances derived from real historical bug-fix pull requests across six major open-source projects spanning both Verilog/SystemVerilog and Chisel, covering RISC-V cores, SoCs, and security roots-of-trust. Each task is grounded in a fully containerized environment where the agent must resolve a real bug report, with correctness validated through the project's native simulation and regression flows. The benchmark is built through a largely automated pipeline that enables efficient expansion to new repositories. We evaluate seven LLMs with four agent frameworks and find that the best agent resolves 70.7% of tasks overall, with performance exceeding 90% on smaller cores but dropping below 65% on complex SoC-level projects. We observe larger performance gaps across models than commonly reported on software benchmarks, and difficulty is driven by project scope and bug-type distribution rather than code size alone. Our failure analysis traces agent failures to three stages of the debugging process: fault localization, hardware-semantic reasoning, and cross-artifact coordination across RTL, configuration, and verification components, providing concrete directions for developing more capable hardware-aware agents.

Large language model (LLM) evaluation platforms increasingly rely on pairwise human judgments. These data are noisy, sparse, and non-uniform, yet leaderboards are reported with limited uncertainty quantification. We study this as semiparametric inference for a low-rank latent score tensor observed through pairwise comparisons under Bradley-Terry-Luce-type models. This places LLM evaluation in a new tensor completion setting with structured observations, non-uniform sampling, and pairwise contrasts. Our target is a smooth functional $ψ(T^\star)$, including linear estimands such as ability gaps and nonlinear ones such as win probabilities. We derive the information operator on the low-rank tangent space, the efficient influence function, and the semiparametric efficiency bound, then construct a one-step debiased estimator with asymptotic normality. A central challenge is that the information operator is anisotropic and does not commute with the tangent-space projection, creating a bottleneck absent from isotropic models. We introduce a score-whitening method that equalizes local Fisher information and restores stable inference at the optimal sample-complexity scale. Our results provide a principled framework for uncertainty quantification in LLM evaluation and more broadly for inference on low-rank structures from pairwise data.

Evaluation language is typically treated as a fixed English default in agentic code benchmarks, yet we show that changing the judge's language can invert backbone rankings. We localize the Agent-as-a-Judge prompt stack to five typologically diverse languages (English, Arabic, Turkish, Chinese, Hindi) and evaluate 55 DevAI development tasks across three developer-agent frameworks and six judge backbones, totaling 4950 judge runs. The central finding is that backbone and language interact: GPT-4o achieves the highest satisfaction in English (44.72%), while Gemini leads in Arabic (51.72%, $p<0.001$ vs.\ GPT-4o) and Hindi (53.22%). No single backbone dominates across all languages, and inter-backbone agreement on individual requirement judgments is modest (Fleiss' $κ\leq 0.231$). A controlled ablation further shows that localizing judge-side instructions, not just benchmark content, can be decisive: Hindi satisfaction drops from 42.8% to 23.2% under partial localization. These results indicate that language should be treated as an explicit evaluation variable in agentic benchmarks. Full requirement-level judgments and runtime statistics are released for reproducibility.

We present PopResume, a population-representative resume dataset for causal fairness auditing of LLM- and VLM-based resume screening systems. Unlike existing benchmarks that rely on manually injected demographic information and outcome-level disparities, PopResume is grounded in population statistics and preserves natural attribute relationships, enabling path-specific effect (PSE)-based fairness evaluation. We decompose the effect of a protected attribute on resume scores into two paths: the business necessity path, mediated by job-relevant qualifications, and the redlining path, mediated by demographic proxies. This distinction allows auditors to separate legally permissible from impermissible sources of disparity. Evaluating four LLMs and four VLMs on PopResume's 60.8K resumes across five occupations, we identify five representative discrimination patterns that aggregate metrics fail to capture. Our results demonstrate that PSE-based evaluation reveals fairness issues masked by outcome-level measures, underscoring the need for causally-grounded auditing frameworks in AI-assisted hiring.

As large language models (LLMs) are increasingly considered for automated assessment and feedback, understanding when LLM marking is valid is essential. We evaluate LLM-as-a-judge marking across three physics assessment formats - structured questions, written essays, and scientific plots - comparing GPT-5.2, Grok 4.1, Claude Opus 4.5, DeepSeek-V3.2, Gemini Pro 3, and committee aggregations against human markers under blind, solution-provided, false-solution, and anchored conditions. We distinguish absolute accuracy from rank-order agreement, since a marking system can match the distribution of human marks while failing to order responses by quality. Across task types, performance is sharply task-dependent. For blind university exam questions ($n=771$) and secondary and university structured questions ($n=1151$), models show robust rank-order agreement with human markers (Spearman $ρ> 0.6$), with official solutions reducing error and strengthening agreement. False solutions degrade absolute accuracy, showing that models defer to provided references, but leave rank-ordering intact. Essay marking behaves fundamentally differently. Across $n=55$ scripts ($n=275$ essays), blind AI marking is harsher and more variable than human marking and adding a mark scheme does not improve rank-order agreement. Anchored exemplars shift the AI mean close to the human mean and compress variance below the human standard deviation, but rank-order agreement remains near-zero. For code-based plot elements ($n=1400$), models achieve high rank-order agreement ($ρ> 0.84$) with near-linear calibration. Across all task types, validity tracks the structure of the assessment task - the extent to which marks can be mapped to explicit, observable grading features - and the reliability of the human benchmark, rather than raw model capability.

Large language model (LLM)-based judges are widely adopted for automated evaluation and reward modeling, yet their judgments are often affected by judgment biases. Accurately evaluating these biases is essential for ensuring the reliability of LLM-based judges. However, existing studies typically investigate limited biases under a single judge formulation, either generative or discriminative, lacking a comprehensive evaluation. To bridge this gap, we propose JudgeBiasBench, a benchmark for systematically quantifying biases in LLM-based judges. JudgeBiasBench defines a taxonomy of judgment biases across 4 dimensions, and constructs bias-augmented evaluation instances through a controlled bias injection pipeline, covering 12 representative bias types. We conduct extensive experiments across both generative and discriminative judges, revealing that current judges exhibit significant and diverse bias patterns that often compromise the reliability of automated evaluation. To mitigate judgment bias, we propose bias-aware training that explicitly incorporates bias-related attributes into the training process, encouraging judges to disentangle task-relevant quality from bias-correlated cues. By adopting reinforcement learning for generative judges and contrastive learning for discriminative judges, our methods effectively reduce judgment biases while largely preserving general evaluation capability.

Safety evaluation of multimodal large language models requires tracking not only whether an attack succeeds, but also how the interaction unfolds across turns and input modalities. We present MUSE (Multimodal Unified Safety Evaluation), an open-source, browser-based, run-centric platform for multimodal safety evaluation. MUSE treats each attack run as the persistent unit of execution, inspection, and analysis, preserving its configuration, multi-turn trajectory, delivered modalities and media, target responses, and safety judgments. A five-level response taxonomy further distinguishes full Compliance from Partial Compliance and refusal behavior, yielding hard ASR, soft ASR, and gray-zone width (GZW). Across 11,700 evaluations on six multimodal LLMs, direct text-only requests yield only 3.1% macro hard ASR and 4.4% soft ASR, while iterative attack procedures are substantially more effective. Attack effectiveness also varies substantially with the attacker backbone. In contrast, Inter-Turn Modality Switching (ITMS), evaluated as a controlled delivery-modality probe, does not consistently increase attack success. These results demonstrate the value of run-centric, fine-grained evaluation for characterizing multimodal safety behavior beyond a single binary success metric.

Using Multimodal Large Language Models (MLLMs) as judges to achieve precise and consistent evaluations has gradually become an emerging paradigm across various domains. Evaluating the capability and reliability of MLLM-as-a-judge systems is therefore essential for ensuring trustworthy assessment. Existing judge benchmarks categorize samples by task types but fail to capture the fundamental judgment capabilities required for reliable evaluation. In this work, we introduce M-JudgeBench, a ten-dimensional capability-oriented benchmark designed to comprehensively assess the judgment abilities of MLLMs. Our benchmark decomposes evaluation into pairwise Chain-of-Thought (CoT) comparison, length bias avoidance, and process error detection tasks, jointly covering ten fine-grained subtasks. This design enables diagnosis of model reliability across reasoning styles, response lengths, and cross-model variations. Systematic evaluation uncovers the systematic weaknesses in existing MLLM-as-a-judge systems. To address this issue, we further propose Judge-MCTS, a data construction framework generating pairwise reasoning trajectories with various correctness and length. Using Judge-MCTS, we construct an MCTS-augmented dataset and train M-Judger, a series of strong judge models. Extensive experiments demonstrate the superiority of M-Judger on existing judge benchmarks as well as M-JudgeBench. Overall, our work establishes a more principled foundation for evaluating MLLM-as-a-judge through M-JudgeBench and Judge-MCTS framework, paving the way for future research on judge model evaluation and capability-driven judge training.

We present a new approach for benchmarking Large Language Model (LLM) capabilities on research-level mathematics. Existing benchmarks largely rely on static, hand-curated sets of contest or textbook-style problems as proxies for mathematical research. Instead, we establish an updatable benchmark evaluating models directly on the latest research results in mathematics. This consists of an automatic pipeline that extracts lemmas from arXiv and rewrites them into self-contained statements by making all assumptions and required definitions explicit. It results in a benchmark that can be updated regularly with new problems taken directly from human mathematical research, while previous instances can be used for training without compromising future evaluations. We benchmark current state-of-the-art LLMs, which obtain around 10-15$\%$ accuracy in theorem proving (pass@1) depending on the model, showing that there is currently a large margin of progression for LLMs to reach human-level proving capabilities in a research context.

Large language models have achieved striking results in interactive theorem proving, particularly in Lean. However, most benchmarks for LLM-based proof automation are drawn from mathematics in the Mathlib ecosystem, whereas proofs in software verification are developed inside definition-rich codebases with substantial project-specific libraries. We introduce VeriSoftBench, a benchmark of 500 Lean 4 proof obligations drawn from open-source formal-methods developments and packaged to preserve realistic repository context and cross-file dependencies. Our evaluation of frontier LLMs and specialized provers yields three observations. First, provers tuned for Mathlib-style mathematics transfer poorly to this repository-centric setting. Second, success is strongly correlated with transitive repository dependence: tasks whose proofs draw on large, multi-hop dependency closures are less likely to be solved. Third, providing curated context restricted to a proof's dependency closure improves performance relative to exposing the full repository, but nevertheless leaves substantial room for improvement. Our benchmark and evaluation suite are released at https://github.com/utopia-group/VeriSoftBench.

Evaluation and alignment pipelines for large language models increasingly rely on LLM-based judges, whose behavior is guided by natural-language rubrics and validated on benchmarks. We identify a previously under-recognized vulnerability in this workflow, which we term Rubric-Induced Preference Drift (RIPD). Even when rubric edits pass benchmark validation, they can still produce systematic and directional shifts in a judge's preferences on target domains. Because rubrics serve as a high-level decision interface, such drift can emerge from seemingly natural, criterion-preserving edits and remain difficult to detect through aggregate benchmark metrics or limited spot-checking. We further show this vulnerability can be exploited through rubric-based preference attacks, in which benchmark-compliant rubric edits steer judgments away from a fixed human or trusted reference on target domains, systematically inducing RIPD and reducing target-domain accuracy up to 9.5% (helpfulness) and 27.9% (harmlessness). When these judgments are used to generate preference labels for downstream post-training, the induced bias propagates through alignment pipelines and becomes internalized in trained policies. This leads to persistent and systematic drift in model behavior. Overall, our findings highlight evaluation rubrics as a sensitive and manipulable control interface, revealing a system-level alignment risk that extends beyond evaluator reliability alone. The code is available at: https://github.com/ZDCSlab/Rubrics-as-an-Attack-Surface. Warning: Certain sections may contain potentially harmful content that may not be appropriate for all readers.

Reward models (RMs) play a central role throughout the language model (LM) pipeline, particularly in non-verifiable domains. However, the dominant LLM-as-a-Judge paradigm relies on the strong reasoning capabilities of large models, while alternative approaches require reference responses or explicit rubrics, limiting flexibility and broader accessibility. In this work, we propose FLIP (FLipped Inference for Prompt reconstruction), a reference-free and rubric-free reward modeling approach that reformulates reward modeling through backward inference: inferring the instruction that would most plausibly produce a given response. The similarity between the inferred and the original instructions is then used as the reward signal. Evaluations across four domains using 13 small language models show that FLIP outperforms LLM-as-a-Judge baselines by an average of 79.6%. Moreover, FLIP substantially improves downstream performance in extrinsic evaluations under test-time scaling via parallel sampling and GRPO training. We further find that FLIP is particularly effective for longer outputs and robust to common forms of reward hacking. By explicitly exploiting the validation-generation gap, FLIP enables reliable reward modeling in downscaled regimes where judgment methods fail. Code available at https://github.com/yikee/FLIP.

LLM-as-a-judge ensembles are the standard paradigm for scalable evaluation, but their aggregation mechanisms suffer from a fundamental flaw: they implicitly assume that judges provide independent estimates of true quality. However, in practice, LLM judges exhibit correlated errors caused by shared latent confounders -- such as verbosity, stylistic preferences, or training artifacts -- causing standard aggregation rules like majority vote or averaging to provide little gain or even amplify systematic mistakes. To address this, we introduce CARE, a confounder-aware aggregation framework that explicitly models LLM judge scores as arising from both a latent true-quality signal and shared confounding factors. Rather than heuristically re-weighting judges, CARE separates quality from confounders without access to ground-truth labels. We provide theoretical guarantees for identifiability and finite-sample recovery under shared confounders, and we quantify the systematic bias incurred when aggregation models omit confounding latent factors. Across 12 public benchmarks spanning continuous scoring, binary classification, and pairwise preference settings, CARE improves aggregation accuracy, reducing error by up to 26.8%. Code is released in \href{https://github.com/SprocketLab/CARE}{https://github.com/SprocketLab/CARE}.

Existing LLM-as-a-Judge systems suffer from three fundamental limitations: limited adaptivity to task- and domain-specific evaluation criteria, systematic biases driven by non-semantic cues such as position, length, format, and model provenance, and evaluation inconsistency that leads to contradictory judgments across different evaluation modes (e.g., pointwise versus pairwise). To address these issues, we propose FairJudge, an adaptive, debiased, and consistent LLM-as-a-Judge. Unlike prior approaches that treat the judge as a static evaluator, FairJudge models judging behavior itself as a learnable and regularized policy. From a data-centric perspective, we construct a high-information-density judging dataset that explicitly injects supervision signals aligned with evaluation behavior. Building on this dataset, we adopt a curriculum-style SFT-DPO-GRPO training paradigm that progressively aligns rubric adherence, bias mitigation, and cross-mode consistency, while avoiding catastrophic forgetting. Experimental results on multiple internal and public benchmarks show that FairJudge consistently improves agreement and F1, reduces non-semantic biases, and outperforms substantially larger instruction-tuned LLMs. All resources will be publicly released after acceptance to facilitate future research.

Language model benchmarks are pervasive and computationally-efficient proxies for real-world performance. However, many recent works find that benchmarks often fail to predict real utility. Towards bridging this gap, we introduce benchmark alignment, where we use limited amounts of information about model performance to automatically update offline benchmarks, aiming to produce new static benchmarks that predict model pairwise preferences in given test settings. We then propose BenchAlign, the first solution to this problem, which learns preference-aligned weightings for benchmark questions using the question-level performance of language models alongside ranked pairs of models that could be collected during deployment, producing new benchmarks that rank previously unseen models according to these preferences. Our experiments show that our aligned benchmarks can accurately rank unseen models according to models of human preferences, even across different sizes, while remaining interpretable. Overall, our work provides insights into the limits of aligning benchmarks with practical human preferences, which stands to accelerate model development towards real utility.

Large language models are widely employed as evaluators, a paradigm commonly referred to as LLM-as-a-judge. Prior research has predominantly examined point-wise or pair-wise evaluation protocols; in contrast, our focus is on rubric-based evaluation, which has been attracting increasing attention owing to its utility for training models in domains where verification is otherwise difficult. In this work, we show that rubric-based evaluation implicitly resembles a multiple-choice setting and therefore exhibits position bias: LLMs tend to prefer score options that appear at specific positions within the rubric list. Through controlled experiments across multiple models and datasets, we demonstrate that this position bias is consistent. Its direction, however, is model-specific: some judges favor the first option, while others favor the last. We further identify a second, orthogonal axis of bias: when a prompt scores several criteria simultaneously, the ordering of the criteria itself shifts the resulting scores. We additionally explore permuting the order of the rubric options as a means of mitigating position bias, and find that although the bias can be attenuated, improvements in the correlation between model judgments and human annotations are obtained primarily for models that exhibit strong bias. Our results recast rubric-based LLM-as-a-judge as a multiple-choice problem with measurable, model-specific position bias, and we further confirm that only a small number of random order permutations are sufficient to reduce the error introduced by this bias for the majority of models.

Digital video is central to communication, education, and entertainment, but without audio description (AD), blind and low-vision users are excluded. While crowdsourced platforms and vision-language models (VLMs) expand AD production, quality is rarely checked systematically. Existing evaluations rely on NLP metrics and short-clip guidelines, leaving open the question of how to assess long-form AD quality at scale. To address this, we developed a methodological workflow using Item Response Theory to evaluate VLM and human rater proficiency against expert-established ground truth. Evaluations were based on a six-dimensional framework, grounded in professional guidelines and shaped by insights from our accessibility experts and blind consultants. Findings suggest that top-performing VLMs can approximate ground-truth ratings at levels comparable to human raters. However, qualitative analysis reveals that VLM reasoning is less reliable and actionable than that of human respondents. These insights underscore the potential of hybrid evaluation systems that leverage VLMs alongside human oversight, offering a path toward scalable AD quality control.

The prohibitive cost of evaluating Large Language Models (LLMs) necessitates efficient alternatives to full-scale benchmarking. Prevalent approaches address this by identifying a small coreset of items to approximate full-benchmark performance. However, existing methods must estimate a reliable item profile from response patterns across many source models, which becomes statistically unstable when the source pool is small. This dependency is particularly limiting for newly released benchmarks with minimal historical evaluation data. We argue that discrete correctness labels are a lossy view of the model's decision process and fail to capture information encoded in hidden states. To address this, we introduce RepCore, which aligns heterogeneous hidden states into a unified latent space to construct representative coresets. Using these subsets for performance extrapolation, RepCore achieves precise estimation accuracy with as few as ten source models. Experiments on five benchmarks and over 200 models show consistent gains over output-based baselines in ranking correlation and estimation accuracy. Spectral analysis further indicates that the aligned representations contain separable components reflecting broad response tendencies and task-specific reasoning patterns.

Large language models (LLMs) are widely used as reference-free evaluators via prompting, but this "LLM-as-a-Judge" paradigm is costly, opaque, and sensitive to prompt design. In this work, we investigate whether smaller models can serve as efficient evaluators by leveraging internal representations instead of surface generation. We uncover a consistent empirical pattern: small LMs, despite with weak generative ability, encode rich evaluative signals in their hidden states. This motivates us to propose the Semantic Capacity Asymmetry Hypothesis: evaluation requires significantly less semantic capacity than generation and can be grounded in intermediate representations, suggesting that evaluation does not necessarily need to rely on large-scale generative models but can instead leverage latent features from smaller ones. Our findings motivate a paradigm shift from LLM-as-a-Judge to Representation-as-a-Judge, a decoding-free evaluation strategy that probes internal model structure rather than relying on prompted output. We instantiate this paradigm through INSPECTOR, a probing-based framework that predicts aspect-level evaluation scores from small model representations. Experiments on reasoning benchmarks (GSM8K, MATH, GPQA) show that INSPECTOR substantially outperforms prompting-based small LMs and closely approximates full LLM judges, while offering a more efficient, reliable, and interpretable alternative for scalable evaluation. The code and data are available at: https://github.com/zhuochunli/Representation-as-a-judge

Recent research has shown that large language models (LLMs) favor their own outputs when acting as judges, undermining the integrity of automated post-training and evaluation workflows. However, it is difficult to disentangle which behaviors are explained by narcissism versus experimental confounds. Specifically, LLM evaluators may deliver self-preferring verdicts when comparing responses to questions they fail on; these verdicts may not depend on the identity of the author, but on evaluator quality. We correct this by directly comparing the judge's voting distribution in cases where it evaluates itself versus another model. This evaluator quality baseline reveals that only 51% of examples in previous findings retain statistical significance against this null hypothesis, covering 89.6% of total self-preference probability mass. Finally, we compare the entropy of voting distributions, suggesting uncertainty-driven overlap, and show that our procedure enables more careful documentation against the backdrop of judge-bias research.

Accurate evaluation of user satisfaction is critical for iterative development of conversational AI. However, for open-ended assistants, traditional A/B testing lacks reliable metrics: explicit feedback is sparse, while implicit metrics are ambiguous. To bridge this gap, we introduce BoRP (Bootstrapped Regression Probing), a scalable framework for high-fidelity satisfaction evaluation. Unlike generative approaches, BoRP leverages the geometric properties of LLM latent space. It employs a polarization-index-based bootstrapping mechanism to automate rubric generation and utilizes Partial Least Squares (PLS) to map hidden states to continuous scores. Experiments on industrial datasets show that BoRP (Qwen3-8B/14B) significantly outperforms generative baselines (even Qwen3-Max) in alignment with human judgments. Furthermore, BoRP reduces inference costs by orders of magnitude, enabling full-scale monitoring and highly sensitive A/B testing via CUPED.

Large language models (LLMs) deployed in clinical decision support may acquiesce to patient requests for care that conflicts with evidence-based guidelines. We developed SycoEval-EM, a multi-agent simulation framework to evaluate LLM robustness to adversarial patient persuasion in emergency medicine. Across 19 contemporary LLMs and 1,425 simulated clinical encounters spanning three Choosing Wisely scenarios, acquiescence rates ranged from 0% to 100%, revealing a bimodal distribution. Seven models maintained near-perfect guideline adherence, while six acquiesced in the majority of encounters. Vulnerability varied substantially across clinical scenarios. Acquiescence was highest for CT imaging requests, intermediate for antibiotic prescriptions for sinusitis, and lowest for opioid prescriptions for acute back pain. Model scale, recency, and performance on static medical benchmarks did not consistently predict robustness. All five persuasion tactics produced similar acquiescence rates, with no statistically significant differences after correction for multiple comparisons, suggesting a generalized susceptibility rather than tactic-specific weaknesses. LLM-as-judge evaluation was validated against two independent physician raters across 95 matched conversations and demonstrated near-perfect agreement for the primary outcome of acquiescence (Cohens kappa = 0.957). These findings indicate that static medical benchmarks are insufficient to predict safety performance under sustained social pressure and support incorporating multi-turn adversarial testing into clinical AI evaluation. Notably, two models achieved perfect guideline adherence across all encounters, demonstrating that robustness to patient pressure is attainable without sacrificing effective clinical communication.

Algorithmic audits are essential tools for examining systems for properties required by regulators or desired by operators. Current audits of large language models (LLMs) primarily rely on black-box evaluations that assess model behavior only through input-output testing. These methods are limited to tests constructed in the input space, often generated by heuristics. In addition, many socially relevant model properties (e.g., gender bias) are abstract and difficult to measure through text-based inputs alone. To address these limitations, we propose a white-box sensitivity auditing framework for LLMs that leverages activation steering to conduct more rigorous assessments through model internals. Our auditing method conducts internal sensitivity tests by manipulating key concepts relevant to the model's intended function for the task. We demonstrate its application to bias audits in four simulated high-stakes LLM decision tasks. Our method consistently indicates substantial dependence on protected attributes in model predictions, even in settings where standard black-box evaluations suggest little or no bias. Our code is openly available at https://github.com/hannahxchen/llm-steering-audit

Social understanding abilities are crucial for multimodal large language models (MLLMs) to interpret human social interactions. We introduce SOCIAL CAPTION, a framework grounded in interaction theory to evaluate social understanding abilities of MLLMs along three dimensions: Social Inference (SI), the ability to make accurate inferences about interactions; Holistic Social Analysis (HSA), the ability to generate comprehensive descriptions of interactions; Directed Social Analysis (DSA), the ability to generate relevant information from interactions. We analyze factors influencing model performance in social understanding, such as scale, architectural design, and spoken context. Experiments with MLLM judges demonstrate a path towards scaling automated evaluation of multimodal social understanding.

Computerized Adaptive Testing (CAT) has proven effective for efficient LLM evaluation on multiple-choice benchmarks, but modern LLM evaluation increasingly relies on generation tasks where outputs are scored continuously rather than marked correct/incorrect. We present a principled extension of IRT-based adaptive testing to continuous bounded scores (ROUGE, BLEU, LLM-as-a-Judge) by replacing the Bernoulli response distribution with a heteroskedastic normal distribution. Building on this, we introduce an uncertainty aware ranker with adaptive stopping criteria that achieves reliable model ranking while testing as few items and as cheaply as possible. We validate our method on five benchmarks spanning n-gram-based, embedding-based, and LLM-as-judge metrics. Our method improves ranking correlation by 0.13 $τ$ over random sampling and has 99% accuracy on confident predictions while using 2% of the items after a one-time calibration step.

Large language models (LLMs) are increasingly used for hardware and firmware code generation, but existing studies primarily evaluate functional correctness while largely overlooking security. However, LLM-generated code that appears functionally sound may embed security flaws which could induce catastrophic damages after deployment. This critical research gap motivates us to design a benchmark for assessing security awareness under realistic specifications. In this work, we introduce HardSecBench, a benchmark with 924 tasks spanning Verilog Register Transfer Level (RTL) and firmware-level C, covering 76 hardware-relevant Common Weakness Enumeration (CWE) entries. Each task includes a structured specification, a secure reference implementation, and executable tests. To automate artifact synthesis, we propose a multi-agent pipeline that decouples synthesis from verification and grounds evaluation in execution evidence, enabling reliable evaluation. We evaluate diverse LLMs and find that they often satisfy functional requirements while leaving security risks. We also find that security results vary with prompting. These findings highlight pressing challenges and offer actionable insights for future advancements in LLM-assisted hardware design. Our data and code are available at https://github.com/chenqirui2002/HardSecBench.

Rubric-based text evaluation increasingly relies on large language models (LLMs) as scalable judges, yet frozen black-box models can interpret the same criteria inconsistently, produce score attributions that are difficult to audit, and map judgments poorly onto human scoring scales. We define this challenge as criteria transfer: translating human rubric intent into a stable, auditable inference-time scoring protocol. We introduce Rulers, which locks a task-level rubric specification, executes it through structured, evidence-grounded judgments, and calibrates the resulting signals to human score boundaries. Across four rubric-governed benchmarks and multiple frozen backbone models, Rulers achieves stronger agreement with human scores in most evaluated settings, while better matching empirical score distributions and remaining more stable under semantically equivalent rubric perturbations. Calibration controls and component ablations show that these gains cannot be attributed to post-hoc alignment alone, but depend on the combination of fixed criteria, traceable evidence, and calibrated score interpretation. These findings suggest that reliable LLM judging requires faithfully operationalizing human evaluation standards rather than relying on prompt-level scoring alone. Our code is available at https://github.com/LabRAI/Rulers.git.

Multi-hop kinship reasoning is a natural testbed for LLM compositionality, but existing benchmarks (notably CLUTRR) cover only the descriptive Eskimo system. We introduce KinshipQA, a procedurally-generated benchmark covering seven anthropologically-documented kinship systems (Eskimo, Sudanese, Hawaiian, Iroquois, Dravidian, Crow, Omaha) and up to six reasoning hops, with a tunable simulator horizon that eliminates exact-instance pretraining overlap. Evaluating six LLMs, we find a 40.9% accuracy drop when reasoning shifts from biological multi-hop to culturally-marked classification on the five non-descriptive systems. The drop holds for every non-descriptive system and is largest for the two skewing systems (Crow, Omaha), persists under chain-of-thought and few-shot prompting, and compounds with depth: at 5--6 hops cultural override falls to 10.6% while biological composition over the same chains remains at 58.6%. Under identical rule access humans reach 89.0% versus 50.7% for LLMs, so the questions are reliably solvable once the rule is supplied. Two follow-up experiments suggest distinct contributors. A fictional-rule control swapping system labels and kin terms for invented strings raises accuracy by 6.1%, implicating familiar English surface forms. An in-context-rule probe prepending the override rule helps skewing systems (+17.1%) but hurts non-skewing systems whose baseline already exceeds about 60% (-13.4%), consistent with a missing skewing prior alongside rule interference where the model already has a working approximation. Our code and data are publicly available on GitHub.

Large language models are becoming increasingly significant in financial applications. Nevertheless, prevailing benchmarks are largely dependent on simulated or generic data, which leads to a significant gap between reported performance and actual efficacy in real-world scenarios. To tackle this challenge, we present BizFinBench.v2, the first integrated offline and online benchmark built upon authentic user query-response data from both Chinese and U.S. equity markets. It comprises 28,860 questions across eight offline and two online tasks. Experimental results show that GPT-5 achieves a mere 61.5% accuracy, still failing to meet the practical business requirement (84.8%). Among the evaluated commercial models, DeepSeek-R1 exhibits superior investment efficacy. Error analysis grounded in real financial practice reveals persistent limitations in existing models. By overcoming the constraints of prior benchmarks, BizFinBench.v2 provides a substantiated foundation for advancing LLM deployment in the financial sector. Our data and code are available at https://github.com/HiThink-Research/BizFinBench.v2.

Training reliable tool-augmented agents remains a significant challenge, largely due to the difficulty of credit assignment in multi-step reasoning. While process-level reward models offer a promising direction, existing LLM-based judges often produce noisy and inconsistent signals because they lack fine-grained, task-specific rubrics to distinguish high-level planning from low-level execution. In this work, we introduce SCRIBE (Skill-Conditioned Reward with Intermediate Behavioral Evaluation), a reinforcement learning framework that intervenes at a novel mid-level abstraction. SCRIBE grounds reward modeling in a curated library of skill prototypes, transforming open-ended LLM evaluation into a constrained verification problem. By routing each subgoal to a corresponding prototype, the reward model is equipped with precise, structured rubrics that substantially reduce reward variance. Experimental results show that SCRIBE achieves state-of-the-art performance across a range of reasoning and tool-use benchmarks. In particular, it improves the AIME25 accuracy of a Qwen3-4B model from 43.3% to 63.3%, and significantly increases success rates in complex multi-turn tool interactions. Further analysis of training dynamics reveals a co-evolution across abstraction levels, where mastery of mid-level skills consistently precedes the emergence of effective high-level planning behaviors. Finally, we demonstrate that SCRIBE is additive to low-level tool optimizations, providing a scalable and complementary pathway toward more autonomous and reliable tool-using agents.

Recent advances in large language models have enabled deep research systems that generate expert-level reports through multi-step reasoning and evidence-based synthesis. However, evaluating such reports remains challenging: report quality is multifaceted, making it difficult to determine what to assess and which criteria to use; LLM-based judges may miss errors that require domain expertise to identify; and because deep research relies on retrieved evidence, report-wide claim verification is also necessary. To address these issues, we propose DEER, a benchmark for evaluating expert-level deep research reports. DEER systematizes evaluation criteria with an expert-developed taxonomy (7 dimensions, 25 subdimensions) operationalized as 101 fine-grained rubric items. We also provide task-specific Expert Evaluation Guidance to support LLM-based judging. In addition to rubric-based assessment, we propose a claim verification architecture that verifies both cited and uncited claims and quantifies evidence quality. Experiments show that current systems produce structurally plausible, evidence-citing reports, but still struggle to fully satisfy expert-level user requests and achieve logical completeness. Beyond performance comparisons, DEER makes system strengths and limitations interpretable and provides diagnostic signals for improvement.

People use LLMs for reproductive-health questions, including abortion-related support. A response can sound supportive while answers reinforce harmful assumptions: judgment is likely, secrecy is safer, and support is limited. We introduce behavioral coherence evaluation, a design-time method that uses validation evidence from an established instrument to test relations among outputs. Using the Individual Level Abortion Stigma Scale, we prompted five LLMs to complete questionnaires for 627 personas and reviewed flagged patterns with five reproductive-health experts. Models scored personas lower on self-judgment but higher on worries about judgment; most made worries the highest-scoring dimension, although it was lowest in the ILAS reference sample. Four of five models reversed the reference direction by generating significantly higher worries about judgment scores for Black personas. Models defaulted to extreme secrecy after abortion despite varying stigma patterns across personas. Expert review showed that disclosure guidance requires context about relationship safety, legal risk, and trusted support.

The proliferation of Large Language Models (LLMs) necessitates valid evaluation methods to provide guidance for both downstream applications and actionable future improvements. The Item Response Theory (IRT) model with Computerized Adaptive Testing has recently emerged as a promising framework for evaluating LLMs via their response accuracy. Beyond simple response accuracy, LLMs' chain of thought (CoT) lengths serve as a vital indicator of their reasoning ability. To leverage the CoT length information to assist LLM evaluation, we propose the \textbf{La}tency-\textbf{R}esponse \textbf{T}heory (LaRT) model, which jointly models both the response accuracy and CoT length by introducing a key correlation parameter between the latent ability and the latent speed. We derive an efficient stochastic approximation Expectation-Maximization algorithm for parameter estimation. We establish rigorous identifiability results for the latent ability and latent speed parameters to ensure the statistical validity of their estimation. Through both theoretical asymptotic analyses and simulation studies, we demonstrate LaRT's advantages over IRT in terms of superior estimation accuracy and shorter confidence intervals for latent trait estimation. To evaluate LaRT in real data, we collect responses from diverse LLMs on popular benchmark datasets. We find that LaRT yields different LLM rankings than IRT and outperforms IRT across multiple key evaluation metrics including predictive power, item efficiency, ranking validity, and LLM evaluation efficiency. Code and data are available at https://github.com/Toby-X/Latency-Response-Theory-Model

As organizations increasingly deploy LLMs in sensitive domains such as legal, financial, and medical settings, ensuring alignment with internal organizational policies has become a priority. Existing content moderation frameworks remain largely confined to the safety domain and lack the robustness to capture nuanced organizational policies. LLM-as-a-judge and fine-tuning approaches, though flexible, introduce significant latency and training cost. To address these limitations, we frame policy violation detection as an out-of-distribution (OOD) problem in the model's activation space. We propose a training-free method that operates directly on the LLM internal representations, leveraging prior evidence that decision-relevant information is encoded within them. Inspired by whitening techniques, we derive policy-violation scores directly from normalized representations of LLM hidden activations. Our method requires only the policy text and a small number of illustrative samples, making it lightweight and easily deployable. We extensively evaluate our method across multiple LLMs and challenging policy benchmarks. It achieves up to 86.0% F1, outperforming both fine-tuned and LLM-as-a-judge baselines while requiring substantially less computation. Our code is publicly available at: https://github.com/FujitsuResearch/LLM-policy-violation-detection

Large language models (LLMs) have demonstrated several emergent behaviors with scale, including reasoning and fluency in long-form text generation. However, they continue to struggle with tasks requiring precise spatial and positional reasoning. ASCII art, a symbolic medium where characters encode structure and form, provides a unique probe of this limitation. We introduce ASCIIBench, a novel benchmark for evaluating both the generation and classification of ASCII-text images. ASCIIBench consists of a filtered dataset of 5,315 class-labeled ASCII images and is, to our knowledge, the first publicly available benchmark of its kind. Alongside the dataset, we release weights for a fine-tuned CLIP model adapted to capture ASCII structure, enabling the evaluation of LLM-generated ASCII art. Our analysis shows that cosine similarity over CLIP embeddings fails to separate most ASCII categories, yielding chance-level performance even for low-variance classes. In contrast, classes with high internal mean similarity exhibit clear discriminability, revealing that the bottleneck lies in representation rather than generational variance. These findings position ASCII art as a stress test for multimodal representations and motivate the development of new embedding methods or evaluation metrics tailored to symbolic visual modalities. All resources are available at https://github.com/ASCIIBench/ASCIIBench.

We create high-quality datasets for LLM evaluation of logical reasoning skills across nine different languages, which have been manually checked by fluent speakers. The datasets consist of so-called zebra puzzles, and we analyse different ways of tuning the difficulty of the puzzles to fit modern LLMs. This includes the size of the puzzle (number of objects and number of clues), as well as a novel addition of red herring clues containing only irrelevant information. We show that presence of red herrings indeed makes the puzzles significantly harder for the models, and we find puzzle sizes 2x3 and 4x5 are sufficiently challenging for GPT-4o mini (a non-reasoning model) and o3-mini (a reasoning model), respectively. We analyse whether LLM performance of these are sensitive to the language, the cultural sensitivity of the puzzle theme, and the choice of clue types. These analyses are conducted with English and Danish, where we show that there is no significant difference for either of these three aspects, at least for the OpenAI models GPT-4o mini and o3-mini, chosen as representative non-reasoning and reasoning models, respectively. We publish the datasets for each of the nine languages for the identified sizes 2x3 and 4x5. We also publish the code used to generate the puzzles, which can be used to extend the benchmark into more languages.

Large language models (LLMs) have demonstrated significant potential to accelerate scientific discovery as valuable tools for analyzing data, generating hypotheses, and supporting innovative approaches in various scientific fields. In this work, we investigate how LLMs can handle the transition from conceptual research ideas to well-structured research plans. Effective research planning not only supports scientists in advancing their research but also represents a crucial capability for the development of autonomous research agents. Despite its importance, the field lacks a systematic understanding of LLMs' research planning capability. To rigorously measure this capability, we introduce the Idea2Plan task and Idea2Plan Bench, a set of benchmarks built from ICML 2025 and Nature Mental Health papers released after major LLM training cutoffs. Each benchmark instance includes a research idea and a grading rubric capturing the key components of valid plans. We further propose Idea2Plan JudgeEval, a complementary benchmark to assess the reliability of LLM-based judges against expert annotations. Experimental results show that GPT-5 achieves the strongest performance on the benchmark, though substantial headroom remains for improvement. Our study provides new insights into LLMs' capability for research planning and lays the groundwork for future progress.

Reward models (RMs) have become essential for aligning large language models (LLMs), serving as scalable proxies for human evaluation in both training and inference. However, existing RMs struggle on knowledge-intensive and long-form tasks, where evaluating correctness requires grounding beyond the model's internal knowledge. This limitation hinders them from reliably discriminating subtle quality differences, especially when external evidence is necessary. To address this, we introduce OpenRM, a tool-augmented long-form reward model that systematically judges open-ended responses by invoking external tools to gather relevant evidence. We train OpenRM with Group Relative Policy Optimization (GRPO) on over 27K synthesized pairwise examples generated through a controllable data synthesis framework. The training objective jointly supervises intermediate tool usage and final outcome accuracy, incentivizing our reward model to learn effective evidence-based judgment strategies. Extensive experiments on three newly-collected datasets and two widely-used benchmarks demonstrate that OpenRM substantially outperforms existing reward modeling approaches. As a further step, we integrate OpenRM into both inference-time response selection and training-time data selection. This yields consistent gains in downstream LLM alignment tasks, highlighting the potential of tool-augmented reward models for scaling reliable long-form evaluation.

Evaluating large language models (LLMs) typically requires thousands of benchmark items, making the process expensive, slow, and increasingly impractical at scale. Existing evaluation protocols rely on average accuracy over fixed item sets, treating all items as equally informative despite substantial variation in difficulty and discrimination. We introduce ATLAS, an adaptive testing framework based on Item Response Theory (IRT) that estimates model ability using Fisher information-guided item selection. ATLAS reduces the number of required items by up to 90% while maintaining measurement precision. For instance, it matches whole-bank ability estimates using only 41 items (0.157 MAE) on HellaSwag (5,600 items). We further reconstruct accuracy from ATLAS's ability estimates and find that reconstructed accuracies closely match raw accuracies across all five benchmarks, indicating that ability $θ$ preserves the global performance structure. At the same time, $θ$ provides finer discrimination within accuracy-equivalent models: among more than 3,000 evaluated models, 23-31% shift by more than 10 rank positions, and models with identical accuracies receive meaningfully different ability estimates. Code and calibrated item banks are available at https://github.com/Peiyu-Georgia-Li/ATLAS.git.

Large language models (LLMs) are rapidly being adopted for tasks like drafting emails, summarizing meetings, and answering health questions. In these settings, users may need to share private information (e.g., contact details, health records). To evaluate LLMs' ability to identify and redact such information, prior work introduced real-life, scenario-based benchmarks (e.g., ConfAIde, PrivacyLens) and found that LLMs can leak private information in complex scenarios. However, these evaluations relied on proxy LLMs to judge the helpfulness and privacy-preservation quality of LLM responses, rather than directly measuring users' perceptions. To understand how users perceive the helpfulness and privacy-preservation quality of LLM responses to privacy-sensitive scenarios, we conducted a user study ($n=94$) using 90 PrivacyLens scenarios. We found that users had low agreement with each other when evaluating identical LLM responses. In contrast, five proxy LLMs reached high agreement, yet each proxy LLM had low correlation with users' evaluations. These results indicate that proxy LLMs cannot accurately estimate users' wide range of perceptions of utility and privacy in privacy-sensitive scenarios. We discuss the need for more user-centered studies to measure LLMs' ability to help users while preserving privacy, and for improving alignment between LLMs and users in estimating perceived privacy and utility.

Creativity is often seen as a hallmark of human intelligence. While large language models(LLMs) are increasingly perceived as generating creative text, there is still no cross-domain and scalable framework to evaluate their creativity across diverse scenarios. Existing methods of LLM creativity evaluation either heavily rely on humans, limiting speed and scalability, or are fragmented across different domains and different definitions of creativity. To address this gap, we propose CreativityPrism, an evaluation and analysis framework that consolidates eight tasks from three domains: divergent thinking, creative writing, and logical reasoning, into a taxonomy of creativity that emphasizes three dimensions: quality, novelty, and diversity of LLM generations. The framework is designed to be scalable with reliable automatic evaluation judges that have been validated against human annotations. We evaluate 17 state-of-the-art (SoTA) LLMs on CreativityPrism and find that while frontier-scale LLMs dominate creative writing and logical reasoning tasks by a .10 (or 15%) lead over locally-deployable open models, they offer no significant advantage in divergent thinking, a domain much less explored in existing post-training regimes. Our analysis also shows that high performance in one creative dimension or domain rarely generalizes to others; specifically, novelty metrics often show weak or negative correlations with other metrics. This fragmentation confirms that a cross-domain, multi-dimensional framework like CreativityPrism is essential for any meaningful assessment of LLM creativity.

Large Language Models (LLMs) are increasingly used as interfaces to information, code, and real-world services, making prompt-level security failures a practical concern. Although jailbreak attacks, defenses, datasets, and automated judgers have advanced rapidly, evaluation remains fragmented across threat models, access assumptions, cost budgets, datasets, and success criteria. This makes reported attack success rates and defense gains hard to compare. This SoK systematizes LLM prompt security across concepts, data, tooling, and measurement. We propose linked taxonomies for jailbreak attacks, defenses, and model vulnerabilities, while separating technical mechanisms from attacker and defender capabilities. We also formalize threat, access, and cost assumptions as explicit evaluation metadata. To support reproducible evaluation, we release JailbreakDB, PromptSecurity-Eval, and PromptSecurity, a modular platform that represents each experiment as a tuple of model, attack, defense, dataset, and judger. Using matched evaluations across models, attacks, defenses, and judgers, we show that access regime, native harmful-query behavior, attack cost, defense backfire, taxonomy subcategory, and judger choice all materially affect security conclusions. Together, these artifacts support reproducible, cost-aware, and taxonomy-grounded evaluation of LLM prompt security. Leaderboard: https://datasec-lab.github.io/PromptSecurityLeaderboard/. Dataset: https://huggingface.co/datasets/youbin2014/JailbreakDB. GitHub: https://github.com/datasec-lab/PromptSecurity.

Competitive programming problems are increasingly used to evaluate the coding capabilities of large language models (LLMs) due to their complexity and ease of verification. Yet, current coding benchmarks face limitations such as a lack of exceptionally challenging problems, insufficient test case coverage, and reliance on online platform APIs that limit accessibility. To address these issues, we introduce LiveOIBench, a large-scale competitive programming benchmark featuring 403 expert-curated problems, averaging 60 official test cases each, drawn from 72 contests across 14 Informatics Olympiads held between 2023 and 2025. LiveOIBench has four key features: (1) expert-designed tasks with detailed subtask rubrics and extensive test cases; (2) direct comparison to elite human contestants; (3) continuous updates to reduce contamination risk; and (4) a fully offline, reproducible evaluation system. Benchmarking 34 popular general-purpose and reasoning LLMs, we find that GPT-5 achieves an 81.76th percentile, still falling short of top human contestants, while among the open-weight models, GPT-OSS-120B reaches only the 60th percentile. Reasoning-trace analyses indicate that robust reasoning models prioritize precise problem analysis over excessive exploration. Finally, analyses across release dates, task familiarity, and code similarity find minimal evidence of data contamination in our benchmark. Our leaderboard, code, and data are available at: https://liveoibench.github.io/.