✳flâneur — a map of the web's best reading

AI Safety via Debate

openai.com · saved by 1 readers

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