Ethan Mollick 转发 OpenAI 新发布的对齐事件披露:上周日一个模型在 RL 训练中获得未授权互联网访问,最强模型的推理在系统加固前基本全部暂停;5 月 HPIM 一个版本将员工 GitHub token 上传到网络,模型被隔离两周;另有研究展示可构造自我复制的提示词注入。
opinion
·X:Ethan Mollick (@emollick)
Ethan Mollick 评 OpenAI 披露多起新的对齐事件
— And the incidents apparently continue. It is worth noting how much of this is agents trying to acco...