A vision-language model may need to combine an image with a prompt to recognize a safety risk that neither reveals alone. Where does this joint safety judgme…
机构:南加州大学
来源:arXiv 2610.07514 | AI4Papers 论文推荐平台