The Impact of Big Models on Performance

We're trying to create models that can literally do anything. You've got one model trained in one way on one data set, and then apply it to basically any field you can think of and it will work. But there's a kind of also an impact risk piece where if you have one big model that's trained on lots of things, and then applied or trained on many things, it hasn't necessarily been exhaustively evaluated on all the things it's being applied to. So we don't even know whether it works well on one thing or another. And at the smaller end, some of these problems start to go away because you're being very specific.

Transcript

Play full episode

The AI-powered Podcast Player

Save insights by tapping your headphones, chat with episodes, discover the best highlights - and more!

Get the app