Is There Exponential Progress in Deep Learning?

In general, deep learning is extremely good at organizing information on many fold. But then you do run into a wall and you have to throw a lot of things at it in order to push past a little bit. I think this is something that we saw very clearly in text to image technology for instance. While it turns out you can just take years old deep learning technology,. If you train it on big enough datasets to embed this text cross image space, you get a model that actually reaches photorealism. It's not the kind of exponential progress people seem to expect sometimes.

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