观点
38AI 评分(两次独立评分的平均)

METR 论 AI 自主能力与灾难风险

原标题:Time horizons used to give us a simple way to argue catastrophic risk was low: models’ autonomous capabilities had broad limitations.

METR@METR_Evals发布时间 (北京时间)

时间跨度曾让我们能简单地论证灾难性风险很低:模型的自主能力存在广泛的局限。 如今这个论证越来越难成立。例如,如果 AI 的优势足以大幅加速 AI 研发呢? https://x.com/METR_Evals/status/1953525150374150654?s=20

中文译文尚未完成,先为你展示原文。

Time horizons used to give us a simple way to argue catastrophic risk was low: models’ autonomous capabilities had broad limitations.

That argument is now harder to make. For example, what if AI’s strengths are sufficient to drastically accelerate AI R&D?
https://x.com/METR_Evals/status/1953525150374150654?s=20

In a new report, we evaluate whether GPT-5 poses significant catastrophic risks via AI R&D acceleration, rogue replication, or sabotage of AI labs. We conclude that this seems unlikely. However, capability trends continue rapidly, and models display increasing eval awareness.
查看原帖
OpenAI / ChatGPT安全对齐评测基准大佬观点#大佬观点#安全/对齐#评测/基准#OpenAI

同一件事的其他报道1 篇