Mitigating Perceptual Judgment Bias in Multimodal LLM-as-a-Judge via Perceptual Perturbation and Reward Modeling
This paper identifies and analyzes 'Perceptual Judgment Bias' in multimodal LLM judges, where models anchor on response text rather than visual evidence when the two conflict. The authors introduce a Perceptually Perturbed Judgment Dataset using counterfactual responses to isolate perceptual errors, and a training framework combining GRPO-based reward modeling with batch-ranking objectives. Experiments on MLLM-as-a-Judge benchmarks show improved perceptual fidelity, ranking coherence, and alignment with human evaluation.
Related guides (4)
Related events (8)
Failure Modes of Multi-Objective Prompt Optimization for LLM Judges
This paper investigates multi-objective prompt optimization for LLM-as-judge systems, testing five decomposition modes of textual gradient optimizers across varying levels of cross-task information sharing. In 6 of 10 configurations, optimization fails to improve over the initial prompt, with gradient specificity dropping 59% when multiple criteria are processed jointly. The authors identify two separable failure modes: gradient dilution at optimization time and instruction interference at inference time. These findings constrain the design space for customizing LLM judges via textual feedback across multiple evaluation criteria simultaneously.
StylisticBias benchmark reveals a small set of visual cues drives most social bias in MLLMs
Researchers introduce StylisticBias, a controlled benchmark of ~25K photorealistic face images with single-attribute variations designed to isolate how specific visual cues shift social judgments in multimodal LLMs. Evaluating six MLLMs across 25 binary social judgment scenarios, they find that age and body type dominate identity-level effects, while fashion style drives the largest attribute-level shifts, with ~15 attributes accounting for ~80% of total bias variation. The benchmark is released publicly on GitHub and Hugging Face, enabling fine-grained bias auditing of multimodal models.
Towards Reliable Multilingual LLMs-as-a-Judge: An Empirical Study
This paper systematically investigates strategies for extending LLM-based automatic evaluation (LLMs-as-a-Judge) to multilingual settings, covering high-, mid-, and low-resource languages (English, Spanish, Basque). The authors compare instruction translation, monolingual vs. multilingual supervision, and model size, finding that fine-tuned smaller models can match proprietary models when in-domain data is available, while zero-shot larger models are preferable out-of-domain. Two meta-evaluation datasets are extended to Spanish and Basque, and all data and code are publicly released.
Judge Arena: Benchmarking LLMs as Evaluators
Hugging Face and Atla have launched Judge Arena, a platform for benchmarking large language models in their role as automated evaluators. The initiative uses an Elo-based ranking system to compare how well different LLMs judge the quality of model outputs, addressing the growing reliance on LLM-as-judge paradigms in evaluation pipelines. This fills a meta-evaluation gap: as LLM judges become standard practice, understanding their relative reliability and biases becomes critical infrastructure for the field.
LLMs fail to consistently simulate demographic perspective-taking in hate speech annotation
A new arXiv paper evaluates whether persona-conditioned LLMs can replicate how different demographic groups perceive hate speech, testing three dimensions: inter-group disagreement, in-group sensitivity, and vicarious prediction. No model consistently captures all three dimensions, and performance is highly model-dependent rather than emerging reliably from identity prompts alone. Vicarious prompting with Llama 3.1 provides the closest approximation to human disagreement patterns across demographic axes. The findings have implications for using LLMs as proxies for diverse human annotators in content moderation tasks.
Real Images, Worse Judgments: Evaluating VLMs on Concreteness and Imagery
This paper evaluates whether vision-language models (VLMs) benefit from real image context when making lexical judgments about word concreteness and imagery. The authors find that real-image contexts frequently hurt alignment with human ratings, especially when visual evidence is least relevant to the word being judged. Probing and canonical correlation analysis reveal that real images cause representational shifts and increased sensitivity to spurious visual cues. Instructing models to focus on text-only content at inference time partially mitigates this degradation.
IFLLM dataset uses mouse and eye-tracking signals to improve LLM alignment via implicit feedback
Researchers introduce IFLLM, a dataset of 1,336 multi-turn interactions from 59 Mechanical Turk workers capturing mouse trajectories and webcam-derived eye gaze to study implicit user feedback for LLM alignment. A reward model trained on this implicit feedback improves text-based reward model accuracy from 55% to 64% and nearly triples relative response quality improvements when combined with DPO across eight LLMs. The work addresses the scarcity and cost of explicit preference annotations by mining behavioral signals already present in user interactions.
PARL: Preference-Aware Rubric Learning for Personalized LLM Evaluation
This paper introduces PARL (Preference-Aware Rubric Learning), a framework that reframes personalized LLM evaluation as a learning problem rather than static judgment. PARL induces preference-aware evaluation rubrics from raw user interaction histories and uses a discriminative reinforcement learning objective to contrast user-authored responses against model outputs, capturing user-specific decision boundaries. Experiments on personalized text generation tasks show PARL produces high-fidelity rubrics that generalize across users and tasks, outperforming existing LLM-as-a-judge and automatic metric approaches.