Multimodal large language models, or MLLMs, perform well at visual understanding and structured generation, yet these capabilities do not establish whether a…
机构:复旦
来源:arXiv 2609.21493 | AI4Papers 论文推荐平台