AI safety via debate

7 years ago 10
We’re proposing an AI safety technique which trains agents to debate topics with one another, using a human to judge who wins.
Read Entire Article