JudgeStealer: Extracting LLM Judging Capabilities across Evaluation Protocols

Large language model (LLM) judges are increasingly used across various evaluation scenarios, making their judgment capabilities valuable intellectual property. However, black-box access exposes these capabilities to model extraction attacks. Existing extraction methods do not specifically target LLM judges and provide limited support for multiple evaluation protocols under restricted query budgets. In this study, we propose JUDGESTEALER, the first query-efficient model extraction framework for replicating judging capabilities across pointwise scoring, pairwise comparison, and listwise ranking protocols. JUDGESTEALER exploits the strong cross-protocol agreement to acquire pointwise scores and transform them into pairwise and listwise supervisions without additional victim queries. To capture informative judge patterns and improve query efficiency, JUDGESTEALER dynamically selects pointwise inputs based on semantic diversity, predictive uncertainty, and potential judge biases. It further applies score smoothing and multi-protocol review to preserve the ordinal structure of scores and mitigate catastrophic forgetting during surrogate adaptation. Extensive experiments on state-of-the-art LLM-as-a-judge and reward models show that JUDGESTEALER consistently outperforms existing extraction baselines, achieving up to 73.3%, 87.0%, and 71.6% accuracy for pointwise, pairwise, and listwise evaluation, respectively. JUDGESTEALER also remains effective across different sur- rogate model scales, adaptation strategies, and reasoning settings. Moreover, JUDGESTEALER demonstrates robustness against representative extraction defenses.

Data Leakage Inflates Generalizability of Power Outage Prediction Models

Power outage prediction models are increasingly used in assessments of climate-driven infrastructure risk, yet current evaluation practices obscure whether these models generalize to the novel conditions such applications require. We identify three common methodological choices in power outage prediction models that influence their ability to generalize across spatial, temporal, and event-based settings. We compare the predictive performance impacts of different methodological decisions using publicly available data for the U.S. East Coast from 2018 to 2023 and feature sets derived from weather reanalysis and land-cover data, and embeddings from a GeoAI foundation model (Prithvi WxC). Specifically, we assess model performance under multiple test selection strategies, including unfiltered random splits, leave-one-state-out, and leave-one-event-out designs, which increasingly approximate real-world deployment conditions. While random train-test splits yield strong performance, we show that these results are inflated by spatial and temporal autocorrelation. Under spatial and temporal holdout experiments, predictive accuracy degrades substantially, with models often failing to outperform a simple null baseline. Incorporating GeoAI foundation model embeddings yields limited and inconsistent improvements, primarily for spatial generalization, and does not resolve poor event-level transferability. These findings suggest that, given current data availability and evaluation practices, publicly trained outage prediction models offer limited and uncertain operational value. Progress will likely require improved data coverage, more realistic evaluation protocols, and a shift in focus from marginal modeling advances toward addressing structural data constraints.

Evaluation Protocols and Cross-Subject Generalization in EEG Emotion Recognition

Reported accuracy in electroencephalography (EEG) emotion recognition depends on the complete evaluation procedure, not only the classifier. We separate the target quantity, development procedure, and reporting rule, then use one archived dynamical graph convolutional neural network (DGCNN) pathway on SEED and SEED-IV as an illustrative case. In a protocol-matched subject-dependent check, the SEED result was within 1.47 percentage points of the public reference value; the 3.40-point SEED-IV difference remained unresolved. Across 30 matched SEED subject-session trajectories, checkpoint selection based on repeated test-set evaluation increased mean window accuracy from 0.7855 at epoch 80 to 0.8892. Under five-fold subject-disjoint evaluation, validation-selected checkpoints achieved training-participant trial accuracies of 0.9990 on SEED and 0.9920 on SEED-IV. Accuracy for entirely held-out participants was 0.5348 (95% conditional subject-level bias-corrected and accelerated [BCa] interval [0.4667, 0.5985]) on SEED. The SEED-IV estimate was 0.3954 ([0.3343, 0.4648]) and is reported only as secondary sensitivity evidence because its protocol-matched compatibility check remained unresolved. The observed train-to-held-out-subject gaps are inconsistent with simple optimization underfitting, but they do not isolate subject identity from implementation, preprocessing, representation, or distributional factors. Supporting analyses further showed that participant rankings depended on representation and time scale, while a development-selected tail-risk ensemble did not establish a positive gain in a separate final evaluation. Subject-dependent, subject-disjoint, and cross-session results should therefore be reported as answers to different questions.

Computational Humor with Multimodal LLMs: Methods, Datasets, Evaluation, and Challenges

Multimodal humor in memes, cartoons, and comics remains difficult for AI systems because intended meaning depends on non-literal mechanisms, shared cultural knowledge, and communicative intent rather than literal scene description. This survey focuses on visual humor understanding in single-image and multi-panel artifacts, while treating humor generation as an emerging downstream frontier. We position the literature against prior humor, sarcasm, and general MLLM surveys and organize it using a capability-centric hierarchy spanning recognition, interpretation and reasoning, and generation. Under this lens, we synthesize benchmark design, evaluation protocols, and modeling paradigms, tracing the field's shift from task-specific fusion models to large-model approaches based on multimodal alignment, evidence-grounded reasoning, and controlled generation. We conclude by highlighting the main barriers to progress: shortcut-prone evaluation, limited cultural and narrative coverage, weak evidence grounding, and unresolved safety and ownership concerns.

Speculative Refinement: A Hybrid Autoregressive Diffusion Decoding Strategy and Its Behavior Across Benchmarks

How should we evaluate generation systems that combine autoregressive (AR) and diffusion decoding? We study this question through Speculative Refinement (SpecRef), a training-free hybrid method that warm-starts a masked diffusion language model from an AR draft using entropy-guided selective masking. Evaluating SpecRef across six benchmarks (HumanEval, MBPP, GSM8K, BBH, ARC-Challenge, HellaSwag) with three distinct evaluation protocols (execution-based pass@1, exact-match, log-likelihood scoring), we surface several findings relevant beyond our specific system: (1) code benchmarks conflate structural discovery with logical correctness: providing a syntactic scaffold lifts accuracy from near zero to over 20% without changing the model, indicating that much of the baseline failure is structural; (2) a refinement tension phenomenon where multi-stage correction degrades already-correct tokens, exposing benchmark saturation ceilings invisible to single-model evaluation; (3) log-likelihood and generative evaluation produce different model rankings for the same model pair, suggesting they measure different capabilities; (4) standard Python post-processing silently breaks code evaluation for non-AR generators. These observations apply to any multi-stage or non-autoregressive generation pipeline and point toward more diagnostic evaluation practices.