The Gradient: Perspectives on AI cover image

Joel Lehman: Open-Endedness and Evolution through Large Models

The Gradient: Perspectives on AI

00:00

Using a Measure Post Hoc to Optimize for the Thing You Really Care About

I think it's a good point that you still might have an objective function that is measuring kind of the goodness of something post hoc but that there could be a fundamental difference between applying a measure post hoc to select as opposed to optimizing that measure directly. I'm like hey to my process just go out there explore this do different things and then as you said afterward maybe in line with fitness function measurements things like that I can measure whether this diversity of behaviors that have been produced actually lead towards a goal I had in mind. It sounds like a really interesting analog to some of the broader processes you brought up things like evolution where it's not entirely volitionless but in some sense still looking at let

Play episode from 22:46
Transcript

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