Complex Objects: Why AI Safety Can’t Just Think in Posts
中文摘要
AI安全不能仅识别有害帖子，必须转向处理更复杂的对象（如用户个人资料），因为深层危害往往隐藏其中。
English Summary
AI safety must evolve beyond detecting harmful posts to addressing complex objects like user profiles, where deeper harm is often embedded.
原文节选
We got good at catching bad posts. Then we realized the harm was living in the profile. Continue reading on Medium »