OpenAI News Research & Papers · May 3, 2018 07:00 imp:60 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. Read original at OpenAI News →