AI 导读
Bengio 谈递归自我改进：“它应该是违法的。” > 他认为有 ***五分之一*** 的概率所有人即将死亡。 > “我们会允许某人做一个可能摧毁大气层的实验吗，即使我们不确定它是否会成功？不会！” （他是全世界被引用次数最多的科学家——横跨*所有*领域。 他放弃了 AI 能力方面的工作，转而专注于 AI 安全，并向世界警告 AI 灭绝风险。）
正文
Bengio on recursive-self improvement: "it should be illegal."
> He thinks there is a ***1 in 5*** chance everyone is about to die.
> "Would we allow someone to do an experiment that could destroy the atmosphere, even if we aren't sure if it will work? No!"
(He's the #1 most cited scientist in the world - across *all* fields.
He quit working on AI capabilities to focus on AI safety and warning the world about AI extinction risk.)
来源：@AISafetyMemes · x.com