AXRP - the AI X-risk Research Podcast cover image

1 - Adversarial Policies with Adam Gleave

AXRP - the AI X-risk Research Podcast

00:00

Training Adversarial Policies

Researchers have developed a way to train an adversarial policy. The method is similar to self play, but it can be used against normal opponents. It's possible that many gradient steps being taken by a self play are really just kind of exploiting some silly bug in your opponent. And so iing it's gong to be very hard to get full confidence that you'r robust to be espens ife attacks because you're never sure that you've tested every possible attack.

Transcript
Play full episode

The AI-powered Podcast Player

Save insights by tapping your headphones, chat with episodes, discover the best highlights - and more!
App store bannerPlay store banner
Get the app