GANs, forged data, and building a brain simulator

36:07But then now GANs are kind of back up. We don't really know which architecture is going to win in the end. But here's the idea, okay? So, you've got something called a forger Mhm. which is trying to forge real-life data. For example, in this case, what we're trying to do is create real faces. So, we're going to have a data set of real faces, and then we're going to have a forger, that's the generator over there, and that's going to generate faces. Okay? And then we're going to have a discriminator on the other side that is going to take the real data and going to take the forged data and try to figure out which one is forged and which one is real. Now, in the beginning, it's going to be obvious. >> In the beginning. Yeah. In the beginning

36:47because the the forger doesn't have any training. He's just going to be making up random nonsense. So, the discriminator is going to be able to look and be like, oh yeah, this is real, this is fake. But as the forger gets better and better because that output of, hey, this is fake, you keep making fake stuff, I want the real stuff, that output is getting back back propagated through to the forger, and the forger is thinking, okay, how do I get better? How do I get better? >> Yes. Pretty soon after multiple rounds of training, the forger is going to get good at making the real data such that it fools your discriminator. Okay? That's the generative part, and the adversarial part is you've got two networks that are adversaries. There's a detective that's trying to detect the

37:28fakes, and then there's a forger that's trying to create even better fakes. That's the adversarial part, right? For anyone who's ever used image generation, particularly Midjourney, this structure is exactly the reason why you'll see a blurry weird thing when you first initiate the image generation, and then it's going through the cycle you just talked about it. The fidelity gets closer and closer and closer until it gives you the final like produced like image. The this is like this the whole stable diffusion kind of came in and changed the game. >> well, and and that one's that one >> slightly different slightly different because so, what you're talking about is very very close to generative adversarial networks because at the end