AI safety via debate 事件

AI safety via debate We’re proposing an AI safety technique which trains agents to debate topics with one another, using a human to judge who wins.