AI safety via debate

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