AI公司内部红队承诺引质疑
AI summary
AI公司同意设立内部红队。👀 👍 希望能看到他们发的帖子，发现能直接送到董事会，而不是经过层层过滤的报告。🍿 现实检验：我们偷了所有模型权重，还在生产环境里插了几个后门……而且从去年五月起就一直保持着持久访问权限，并且我们做这些完全没用任何AI。😂 不过，这大概不够格写一篇花哨的营销博客文章吧。🤔
Full text · Original
AI companies agreed to have internal red teams. 👀 👍
Hope we'll see posts from them and findings go directly to board of directors rather than through layers of filtered reporting. 🍿
Reality check: We stole all the model weights, and also inserted a couple of backdoors in prod... and still have persistent access since May of last year AND we didn't use any AI to do all of that. 😂
Probably not material for a fancy marketing blog post, though. 🤔
Joint Commitment on Frontier ResponsibilitiesView the quoted post on X