5. Charlie Snell on DALL-E and CLIP
The Inside View
00:00
How to Like De-Neuralize a Noise Generator?
The de-noising process looks like you're trying to learn the inverse of the noiseing process. So there's some neural network which takes in the noisy image and outputs an, an, a noisy version of the image. And so the way it's trained is about like by the thousandth time step, it converges to like the final version. But if after it doesn't step, it doesn't come true to anything, it's still pretty much garbage.
Transcript
Play full episode
