May 2, 2018
Researchers propose using debate between AI systems to help human judges evaluate difficult questions.
A proposed route toward scalable oversight, supported by an initial experiment rather than a complete safety solution.
Explore on Humanity Unfolds
