Papers for
automated content moderation teams
Papers whose findings have a practical use for this group, as judged from the abstract. Open a paper to read what it means in practice.
Combining human and AI feedback to better rank language models
Prediction-Powered LLM Ranking with Adaptively Collected Human Feedback
Abstract: Ranking large language models (LLMs) through human evaluations is essential for downstream decision-making. Because human feedback is expensive, evaluation platforms often collect it adaptively, prioritizing informative model comparisons using previously observed feedback. To further improve precision with limited human feedback, evaluators can draw on abundant unrated responses and relatively inexpensive LLM judges; however, judge bias and adaptive dependence challenge standard i.i.d. inference frameworks. We propose Prediction-Powered Ranking (PP-Rank), a framework that combines LLM-judge predictions with human feedback to improve precision while providing valid inference under adaptive sampling. Data-driven prediction weights minimize the asymptotic covariance trace within the proposed class. In theory, we establish asymptotic normality for general evaluation targets and simultaneous ranking inference. A joint martingale analysis handles dependence between adaptive human feedback and auxiliary data. We further derive a rule for reducing exploration that prioritizes lower-cost comparisons while matching a conservative variance bound, and characterize allocation limits for regular $\sqrt{T}$-rate estimation of pairwise means. Semi-synthetic experiments on UltraFeedback-Extended show improvements over human-only baselines in score estimation, confidence interval width, and ranking certification. Synthetic experiments demonstrate robustness to judge quality and adaptive allocation, while showing that PP-Rank can substantially reduce evaluation costs without sacrificing precision.
Agora method improves aggregating opinions of large language model judges
AGORA: Reliability-Aware Aggregation of LLM Judges via Item-Response Theory
Abstract: LLM-as-a-judge is a common substitute for costly human evaluation, and aggregating several judges is the standard remedy for any single judge's bias. Judge pools mix judges of very different reliability, while the human-labeled sets that validate them hold a few hundred items or fewer. In this regime the per-item mean weights weak judges as heavily as strong ones, and confusion-table models such as Dawid-Skene must estimate a per-judge error table from too few items. We propose AGORA (Aggregation Of Reliability-weighted Assessments), an item-response-theory aggregator that places every item on one latent quality scale and describes each judge by a sensitivity parameter and ordered rating thresholds, so judge reliability is learned from all items jointly and without human labels. Across three benchmarks, AGORA matches or exceeds Dawid-Skene on large evaluation sets, and its advantage widens as the set shrinks toward a few hundred items. It remains accurate as pools grow more uneven, a label-free dispersion statistic indicates in advance when trust weighting helps, and its learned trust weights track per-judge human validity.
Vision language model ranking improves by detecting cue reliance shifts
Shift-Reliance Alignment: Ranking Vision-Language Models by the Cues They Trust
Abstract: Vision-language models (VLMs) are usually chosen from leaderboards, which give one ranking for every application. On questions about objects that a scene suggests but does not contain, however, the model with the lowest error on the POPE hallucination benchmark is not the one with the lowest error on such questions. For yes/no questions about object presence, we show that the best model for a target domain can be predicted without running any candidate on that domain, given 650 to 1,200 labeled target items that calibrate a detector. We measure each model's reliance on a cue, a feature such as scene context or text in the image, by editing 100 labeled source images so that the cue suggests the wrong answer, and comparing with a matched control edit. Reliance predicts which VLMs are misled by physical sticker attacks (Spearman .93 on nine models not used to design the score) and which hallucinate on unedited COCO photographs, beyond what their tendency to answer yes explains. Shift-Reliance Alignment (SRA) scores a model by its source error on the labeled source questions plus its reliance multiplied by the shift. The shift is the proportion of target questions on which the cue misleads, and a detector estimates it on the unlabeled target images. SRA selects the model with the lowest score. In pre-registered tests on new images where most questions mislead, it lowered the error of the selected model from 30.4% under selection by source error to 18.8% on Objects365 and OpenImages but, against its prediction, not on COCO. On CueDial, a benchmark we release that varies how often a cue misleads, selection by model confidence falls more than 40 percentage points behind the best model when overlaid text denies a visible object, and SRA stays within 3 points.
Dualview improves multipoint text scoring with fewer parameters
DUALVIEW: Leveraging Broad and Output-Specific Evidence for LLM-based Multi-Output Text Regression
Abstract: While work on LLM-based regression has largely focused on scalar prediction, multi-output text regression remains less explored. We present DualView, a parameter-efficient approach to leverage LLMs for multi-output text regression. \method{} first extracts broad and output-specific representations from a frozen LLM, then uses a dual-branch architecture to produce the full score vector. Across five heterogeneous regression tasks and eight LLMs (up to 122B), DualView outperforms seven supervised regression baselines while requiring 82--1842 times fewer fitted parameters than the three strongest baselines. These improvements extend across individual output dimensions and persist even with limited training data. Notably, DualView using a smaller open-weight LLM can outperform unsupervised prediction with Gemini 3.1 Pro. Controlled ablations show that broad and output-specific evidence provide complementary predictive benefits, while lightweight shared predictors achieve strong performance with fewer trainable parameters.
Label collapse undermines reliability of diffusion language model predictions
Label Collapse in Diffusion Language Model In-Context Learning
Abstract: In-context learning (ICL) applies large language models to new tasks by prepending a few labeled demonstrations to a query. Prior work has studied how demonstration ordering affects ICL, but the interaction between label arrangement patterns and model biases has not been systematically examined. We define label collapse as the failure mode in which the model generates query-independent labels, and show that it is architecturally universal at the tested 8-9B scale: five models spanning autoregressive (Llama-3, Qwen3.5) and discrete diffusion (LLaDA, iLLaDA, LLaDA-Instruct) families all exhibit label collapse even under the class-balanced alternating arrangement, with some label word pairs collapsing in virtually every tested condition and others flipping their collapse direction with arrangement. Critically, accuracy alone cannot detect this collapse: models can achieve near-chance accuracy while generating query-independent labels, a score indistinguishable from balanced performance, so the failure can persist undetected in deployment. We measure separation, the log-odds of the two label words at the query position (the quantity that label-bias calibration operates on), as a complementary diagnostic that exposes collapse hidden beneath surface accuracy, together with a measurement protocol that guards label extraction against tokenizer and formatting artifacts. For autoregressive models, collapse persists where the majority-label-bias account does not apply, and arrangement patterns can reverse the collapse direction; for diffusion models, we present a systematic evaluation of ICL failure modes, complementing contemporaneous studies of dLLM context comprehension and demonstration-order sensitivity, and identify open questions about how iterative denoising interacts with label biases.
VisExMEM verifies image captions with claim-level visual evidence
VisExMEM: Visual Evidence-Based Explainable Multimodal Evaluation Metric
Abstract: Current image-text evaluation often relies on a global score, which gives limited insight into which semantic details are supported by the image. Explainable metrics provide rationales, but the visual evidence supporting those rationales can still be difficult to verify. We introduce VisExMEM, a reference-free evaluator that decomposes captions into atomic claims, localizes visual evidence, verifies each claim, and averages claim-level support into a global score. Explanations use the same evidence and decisions. We train Qwen3.5-based 4B and 9B models on 100K samples with multi-task SFT, counterfactual refinement, and evidence-aware GDPO. Across eight benchmarks, VisExMEM-9B achieves the highest descriptive composite among open-source evaluators in our comparison, exceeding VQAScore by 11.7 points while remaining competitive with GPT-5.6 Sol. We also introduce AGITA-Bench, containing 600 human-annotated image-text pairs and 8.7K atomic claims for global and claim-level evaluation. For untuned Qwen3.5 and GPT-5.6 Sol controls, decomposed prompting raises the same composite by 3.6-5.1 points relative to holistic prompting. Human evaluation shows that VisExMEM explanations outperform prior explainable metrics and approach GPT-5.6 Sol.
Stateless service gives repeatable text changes preserving original meaning
perturber: A Stateless Service for Semantics-Preserving Text Perturbations
Abstract: Large language model (LLM) evaluations on clean benchmarks overstate model robustness: real inputs arrive perturbed, and these meaning-preserving variants routinely move model answers. Studying this systematically requires applying perturbations one at a time, reproducibly, and identically wherever they run, yet existing tools couple them to a benchmarking stack rather than packaging them as a reusable transform layer. We present perturber, a stateless service and Python library that applies exactly one semantics-preserving perturbation per call, drawn from a first-party catalog of 68 perturbations across 11 families. It turns on a single primitive, with composition left to the caller chaining calls, so the service and the library produce identical output by construction. We evaluate the tool itself, its correctness, reproducibility, and cost, rather than any model's robustness, measuring meaning-preservation per perturbation rather than assuming it. The demonstration adds a live playground, an introspectable catalog for benchmark selection, and a one-command reproducible evaluation.
Controlled fusion of audio video and text improves decision making with limited data
Controlled Acquisition and Abstention in Three-Channel Score Conflicts
Abstract: When audio, video, and text disagree, accuracy alone does not show whether to acquire another source or abstain. We study these choices in a controlled three-score benchmark: a policy observes two signed scores, may request the third at a cost, and can abstain. The primary reward is mechanism-specific: abstention is correct only for one designated ambiguity mechanism and is penalized under mixed corruption. Matched controls show that a threshold policy matches always-request decisions with fewer requests; its advantage over always-answer fusion depends on the reward assigned to that ambiguity. On a partially held-out synthetic split, the threshold policy reaches 0.789 +/- 0.006 targeted decision accuracy and 0.481 +/- 0.014 utility across 83 seeds. A three-score majority reference reaches 0.626 +/- 0.008 and 0.252 +/- 0.016, but uses more information. In a matched-budget test, a train-only value selector improves utility over no-query and matched-random policies at 10% and 25% budgets, while pair uncertainty has higher utility at every budget. At 50% and 63.7% budgets, the selector lowers utility despite slightly higher non-ambiguous accuracy. If all abstentions are scored incorrect, majority outranks the threshold policy in utility. At a central temporal setting, full-trace controls match the neural models while position perturbations separate them. On held-out-actor emotion clips, eight-frame fusion has opposite-signed accuracy differences for two encoder pairs, with both actor intervals containing zero; matched-request routing gains are small and uncertain. These results separate full-modality accuracy from pre-request selection value and show that selection value depends on budget and the observed-pair ranking.
Vision language models reveal new clues to spot adversarial images
Detecting Adversarial Images through Response Profiles of Vision-Language Models
Abstract: Adversarial perturbations can alter the predictions of frozen vision-language models (VLMs) while leaving their confidence and image--text similarity patterns seemingly plausible. We investigate whether we can identify adversarial inputs based on the broader way an image interacts with a collection of general semantic prompts. Our detector summarizes these responses using category-level statistics, relationships among prompts, deviations from clean reference distributions, and stability under weak image transformations, producing a compact response profile that is classified by a lightweight model while the VLM remains fixed. We evaluate the approach on multiple public image datasets, several CLIP-style visual backbones, and a range of gradient-based, optimization-based, automated, and spatial attacks. The detector achieves strong discrimination in attack-specific settings and retains substantial performance when evaluated on attacks not seen during training. Under a controlled detector-specific protocol, the response-profile representation outperforms the evaluated embedding-geometry baselines. Additional analyses show that the feature groups provide complementary information and that the method remains effective under variations in the prompt configuration. We also examine inference cost and performance against detector-aware adaptive attacks. Overall, the results indicate that response patterns across semantic prompts provide a useful complementary signal for adversarial image detection in frozen VLMs.
Llm judges adapt to human preferences despite position bias
DIAL: Position-Debiased LLM Judges with Adaptive Human Preference Calibration
Abstract: Large language models (LLMs) as a judge enable scalable evaluation, but their judgments can be sensitive to response order and, even after removing such position effects, can still diverge systematically from human preferences.We introduce DIAL, a unified framework that combines abundant LLM comparisons with limited human comparisons to separate judge-specific position effects, learn shared structure in position-debiased LLM preferences, and adaptively calibrate that structure toward the human preference target. Theoretically, we study three aspects of DIAL: (i) identification of latent LLM preferences, position effects, and human calibration; (ii) adaptive estimation that balances LLM anchoring against limited human evidence; and (iii) fixed-weight uncertainty quantification for the calibrated human preference. Empirically, we evaluate position debiasing and human alignment separately in controlled simulations and on three human-preference benchmarks, showing that DIAL remains robust to unbalanced response order, achieves strong human-aligned rankings with limited labels, and adapts toward human evidence when LLM information is imperfect. Our real-data study collects over 410K judgments from 21 LLM judges in both display orders, providing a resource for future studies of LLM-judge bias, heterogeneity, and human alignment.
Visual question answering improves safety using evidence based agent collaboration
A$^2$Safe: Counterfactual Evidence-Aligned Adaptive Agent Collaboration for Safe and Effective Visual Question Answering
Abstract: Visual Question Answering (VQA) with Multimodal Large Language Models (MLLMs) requires not only producing safe and effective responses, but also grounding safety decisions in the multimodal evidence that determines risk. Recent safety-alignment methods improve refusal behavior and contextual risk awareness, yet correct safety outcomes may still rely on superficial textual or visual correlations, particularly when risk emerges from interactions between individually benign image and question content. To address this issue, we propose A$^2$Safe, a counterfactual evidence-aligned adaptive agent collaboration framework for safe and effective VQA. A$^2$Safe organizes localized visual observations, textual intent, and cross-modal risk relations through a Grounded Safety Evidence Board, making the basis of safety decisions explicit. Counterfactual safety evidence alignment enforces invariance to safety-irrelevant changes while requiring appropriate safety-state and response-mode transitions when risk-critical evidence is minimally altered. The resulting evidence state further supports adaptive collaboration, enabling direct answering when grounded evidence is sufficient and invoking policy critique and response revision when evidence is risky, uncertain, or conflicting. Under complementary safety-critical and general VQA protocols, A$^2$Safe achieves a 95.72 SIUO safety score, reduces the benign refusal rate on MOSSBench to 14.67%, and maintains an average general VQA score of 78.34 with 27.8% token overhead. These results support counterfactual evidence-aligned adaptive collaboration for safe and effective multimodal question answering.
Flow matching stability improved by budgeting per-step variance in image generation
$λ$-Controlled GRPO: Turning Flow-Matching Ratio Instability into a Budgeted Resource
Abstract: Reinforcement learning is increasingly used to align image generators with reward signals, and Flow-GRPO recently extended this paradigm to flow-matching models by treating the denoising sampler as a stochastic policy that can be optimized from reward feedback. Training in this setting is unstable in a way specific to multi-step denoising: the policy update changes systematically across denoising steps, with importance ratios drifting below one, becoming increasingly dispersed, clipping at different rates, and leaving fewer usable samples late in training. Prior work treats these effects as separate failure modes and addresses each with a hand-tuned stabilizer. We show instead that they arise from a single per-step quantity, which we call path variance. This quantity is determined exactly by the sampler's Gaussian transition kernel and can be estimated cheaply during training. This reframes instability as a resource that can be measured and budgeted rather than a collection of symptoms to repair. Our method, $λ$-Controlled GRPO, calibrates importance-ratio behavior from this predicted law rather than from noisy empirical statistics, and allocates gradient effort across denoising steps according to their predicted cost. The two scales governing the update are fixed by standard policy choices rather than introduced as free tuning parameters. On a text-to-image model under two reward settings, rendering difficult target text scored by optical character recognition and matching human preferences scored by a preference model, $λ$-Controlled GRPO improves both text accuracy and preference reward over the strongest empirical stabilizer. It also keeps late-step path variance within its intended budget, precisely where the baseline systematically overshoots. The result is a Flow-GRPO update calibrated by its own transition law rather than stabilized after instability appears.