研究人员开发了 SIEVES，一种用于提高多模态大语言模型（MLLMs）在分布外场景下可靠性的新方法。SIEVES 通过学习估计推理模型提供的视觉证据质量来实现选择性预测。这种方法显著提高了模型覆盖率，在具有挑战性的基准测试中最高可提高三倍。值得注意的是，SIEVES 可以应用于 Gemini-3-Pro 等专有模型，而无需访问其内部权重或 logits。 AI
影响 通过改进选择性预测和泛化到未见数据，提高 MLLMs 在真实世界场景下的可靠性。
排序理由 介绍多模态大模型泛化新方法的学术论文。
AI 生成摘要 · Google Gemini · 来自 3 个来源。 我们如何撰写摘要 →
完整方法见我们的编辑标准。
Multimodal large language models (MLLMs) achieve ever-stronger performance on visual-language tasks. Even as traditional visual question answering benchmarks approach saturation, reliable deployment requires satisfying low error tolerances in real-world out-of-distribution (OOD) …
arXiv:2604.25855v1 Announce Type: new Abstract: Multimodal large language models (MLLMs) achieve ever-stronger performance on visual-language tasks. Even as traditional visual question answering benchmarks approach saturation, reliable deployment requires satisfying low error tol…
Multimodal large language models (MLLMs) achieve ever-stronger performance on visual-language tasks. Even as traditional visual question answering benchmarks approach saturation, reliable deployment requires satisfying low error tolerances in real-world out-of-distribution (OOD) …