
Boris Dayma — The Story Behind DALL·E mini, the Viral Phenomenon
Gradient Dissent: Conversations on AI
00:00
How Do You Train a Tody Image?
The training, the way talks, is like a the image, yo will fist and conditon to a sequence numbers. And then your your text will be also sequence of numbers,. And it's de coder that goes from one to one. Ander the attention looks onle the previous, the previous spatches ane an, it will, it will predict it. And then you have the grand truth, because you know, what is the real image. So then you are going to calculate crosantropy or gone calculate the loss, the same that you would do with another proem to see how wrong You were.
Transcript
Play full episode