Boris Dayma — The Story Behind DALL·E mini, the Viral Phenomenon
Gradient Dissent: Conversations on AI
00:00
How Long Does It Take to Train on 400 Million Images?
When you have so much data, maybe more than what your model size can handle. You need to be very r ful of having a good validation lus. So i an cuts the images also in a higher resolution than me, so they have less of those artifacts. But sometimes you realize that you will take a description and twill show an image that it has seen before. There's a lot of memory in that e.
Transcript
Play full episode
