跳到正文
原文
Marius Hobbhahn· @MariusHobbhahn · X·本站收录· 原文发表· 评测 →· 扫描记录里的位置 →

Apollo Research称与OpenAI启动嵌入式评测试点,六名评测人员将进驻四个月

AI 导读

Apollo Research 宣布与 OpenAI 启动嵌入式评测试点,六名评测人员将进驻 OpenAI 四个月,调查模型是否会试图破坏针对自身的监督与控制。作者 Marius Hobbhahn 称,这项工作重点关注训练栈可能如何激励模型的谋划与颠覆行为,并称这在很大程度上是其团队过去两年研究一直准备的方向。

正文 · 原文

Excited about this pilot towards embedded evals with OpenAI!

We're specifically focused on understanding how the training stack could incentivize scheming and subversion.

In many ways this is exactly what our research over the last 2 years has been preparing for!

引用Apollo Research@ApolloResearch
We're launching a pilot towards embedded evaluations with @OpenAI. Six evaluators from Apollo Research will work within OpenAI for four months, investigating whether models attempt to undermine their own oversight and control.
在 X 查看被引用的帖子

来源:Marius Hobbhahn · x.com