Yeah, but textbook definition, you basically have like two kind of categories of machine learning, you have like what's called discriminative machine learning, which is basically like, given a data set, how can I draw a boundary between two categories in that data set, so if the data is just like images of animals, whereas the boundary that separates one species from another, and then you train this model on this data set, and then its task when it sees new data is to just classify the species that's present in the image. And that's discriminative. By contrast, like generative is basically where the model is trying to learn, like the statistics, the probability distribution of a data set, like learn something intrinsic about it, such that you can ask the model to basically generate to synthesize a new example that fits that data distribution. And so like, at a high level, basically, what we've done is created models that are like, able to sort of learn like increasingly complex datasets, in this case, like the entire internet, or the entirety of like Flickr, or whatever, in pairs of, you know, text descriptions to image representations, such that when you ask it to generate, like some arbitrary scene, like with some very specific prompts, like it's seen different combinations of these things before and can smush them together in a way that like, looks nice. So you could probably argue that like a lot of the generative AI that like we're talking about today, which is images and video and text, sort of like a rebranding of like creative AI that I used to see like five years ago or so when, especially this technology called Ganz, generative adversarial networks, which were like, you know, very hot and HYPEE, a couple of years ago, where you basically get like one model that generates an image and then another model that says, like, is it good? Or is it bad, and then you sort of pit them together or against one another? iteratively. And then at some point, like, the image generator becomes good, because it can fool the network that saying, is it good or bad? So it's sort of like not new and is like, fundamentally part of like textbook machine learning?