The Gradient: Perspectives on AI cover image

Miles Brundage on AI Misuse and Trustworthy AI

The Gradient: Perspectives on AI

00:00

GP2: A Learning Experience

I think GP2 was kind of a first case study in thinking more about what to do with publication norms as far as dual use technology. We ended up releasing, you know, all the versions of GPT-2. And I think there ended up being a ton of creative and commercial and other kinds of applications of language models. But I think the risks are starting to get, get significant. You know, if you look at some of the scaling graphs in the GPT-3 paper, I am more concerned about G PT-3 scale models in GPT- 2. There is a statistical relationship between size and propensity to deceive humans.

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