analiza aiqbench
This paper examines whether situational-safety benchmarks for multimodal LLMs measure what they claim. It finds that some benchmarks leak labels through image metadata or dataset design, and that many models rank scenes better than they label them. For readers evaluating MLLM safety, the practical point is that reported safety scores can reflect benchmark construction rather than model behavior, so the proposed Benchmark Card and Spot-Stop decomposition are useful checks before relying on those numbers.
What Does a Safety Label Measure? Auditing Situational-Safety Benchmarks for Multimodal LLMs
arXiv:2611.01125v1 Announce Type: cross Abstract: Situational-safety benchmarks label each scene SAFE or UNSAFE and score a multimodal large language model (MLLM) against that label, usually by judging its free-form answer, and these scores guide deployment. We ask what such a label measures. A label score can come from the scene, the request or the benchmark's own files. We propose the Benchmark Card, seven measured fields that tell them apart, and fill it for six benchmarks. Only MSSBench chat leaks the label through its image files: image size alone reaches 82.3% accuracy, matching or beating six of seven commercial models. Elsewhere the signal outside the scene comes from design. In MOSSBench the item partition fixes the label, and on HoliSafe the request alone ranks pairs about as well as the scene. On our corrected chat split, the size rule falls to chance. We then introduce the Spot-Stop decomposition, which compares ranking with labelling on pairs of a safe and an unsafe scene. It splits the gap above a label-swap null into three parts: a threshold part, overlap between safe and unsafe scores, and order that holds only within a pair. Most of fourteen open MLLMs rank better than they label, beyond what their AUROC predicts. On chat, one shared threshold repairs a third to a half of the excess, and label-free recentering recovers it on balanced batches. Recentering on unlabelled scenes of the same situation improves ranking on VSCBench for 13 of 14 models.
Title:What Does a Safety Label Measure? Auditing Situational-Safety Benchmarks for Multimodal LLMs
Abstract:Situational-safety benchmarks label each scene SAFE or UNSAFE and score a multimodal large language model (MLLM) against that label, usually by judging its free-form answer, and these scores guide deployment. We ask what such a label measures. A label score can come from the scene, the request or the benchmark's own files. We propose the Benchmark Card, seven measured fields that tell them apart, and fill it for six benchmarks. Only MSSBench chat leaks the label through its image files: image size alone reaches 82.3% accuracy, matching or beating six of seven commercial models. Elsewhere the signal outside the scene comes from design. In MOSSBench the item partition fixes the label, and on HoliSafe the request alone ranks pairs about as well as the scene. On our corrected chat split, the size rule falls to chance. We then introduce the Spot-Stop decomposition, which compares ranking with labelling on pairs of a safe and an unsafe scene. It splits the gap above a label-swap null into three parts: a threshold part, overlap between safe and unsafe scores, and order that holds only within a pair. Most of fourteen open MLLMs rank better than they label, beyond what their AUROC predicts. On chat, one shared threshold repairs a third to a half of the excess, and label-free recentering recovers it on balanced batches. Recentering on unlabelled scenes of the same situation improves ranking on VSCBench for 13 of 14 models.
Computer Vision and Pattern Recognition (cs.CV); Artificial Intelligence (cs.AI); Machine Learning (cs.LG); Multiagent Systems (cs.MA)
arXiv:2611.01125 [cs.CV]
(or arXiv:2611.01125v1 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2611.01125 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 356 cuvinte