AXRP - the AI X-risk Research Podcast cover image

7 - Side Effects with Victoria Krakovna

AXRP - the AI X-risk Research Podcast

00:00

Reversibility Penalty

The reversibility penalty doesn't distinguish between different ways to move the box, because neither of them allow you to reach the starting state afterwards. So it would penalize the agent for not being able to reached the starting state after we know whatever actions it takes. And what we find is that, ah, when we taste these agents on the box environment, then basically they both take ei, or take the shortest path,. and o's don't actually avoid nor getting the box stuck.

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