AI 导读
Ethan Mollick 转发 Anthropic 关于自复制 AI 的研究，认为这是对关注 AI 能力阈值而非平稳爬升的有价值且令人不安的提醒。该研究讨论的阈值是在杂牌电视等 IoT 设备上实现自复制 AI 蠕虫所需的能力，配图显示可在电视集群上运行的小型开源模型能力指数持续上升，距阈值线还差约 2.4 ECI。
正文
A worthwhile (and unnerving) reminder from Anthropic that we should be paying attention to AI ability thresholds, not just steady climbs
In this case, the threshold is the capability needed for a self-replicating AI worm on IoT devices like off-brand TVs https://www.anthropic.com/research/self-replicating-ai
来源：Ethan Mollick · x.com