Vision Language Models (VLMs) excel on visual benchmarks but fail systematically on tasks requiring abstract reasoning. Existing benchmarks document this fai…
机构:Google DeepMind
来源:arXiv 2610.07646 | AI4Papers 论文推荐平台