← Todas las noticias

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.
Leer el original en OpenAI News →