AI safety via debate
来自 OpenAI News
· 2018-05-03
精选
We’re proposing an AI safety technique which trains agents to debate topics with one another, using a human to judge who wins.