Raw LLM Responses

A) Stability AI's founder admitted to there being copyrighted images in Open AI's Clip dataset. That was why the "in the style of _artist's name_"-prompt worked so well a few weeks back. The companies only started to care when that was found out and now they claim to have retrained their LAION dataset without Clip. I guess we have to take their word for it. Do we know that Instagram actually sold people's art to be part of the training datasets? Is there any record of that? Also, I think the impact of open source AI art models should warrant an update in the terms of service. The impact of these image generators was unknow less than a year ago. This will affect so many people on such a scale that it changes the whole environment around art. No matter how it was done, the datasets were collected in a sneaky way. I tried the "have I been trained" website, and there definitely was nudity in there. And I just search my name. (fortunately for me, unfortunately for someone else, it was not my nudity) B) So just in terms of processing power we can't compare the human brain to the AI. A human kan throw a bullet at someone and things won't be that bad, if a machine gun throws a bunch of bullets at you the potential damage can be devestating. And you wrote this, stating another major difference: "It can find pattern in a way that humans simply can't. It can look at all the pixels of your art and learn the pattern. Learn What makes your style. And I don't think it's possible to copy right a style." The focus shouldn't be on the word "style" but on "fair use" connected to that style. One way to violate fair use is to take job opportunities or income away from someone by offering a product too close to their "style". Since an AI trained on a specific artist can generate a whole new portfolio resembling that artist's work in just a couple of days(hours?), that could be infringing on fair use. If the people behind that AI model are competing on the same market and taking profit away from the artist the model was trained on. But since the datasets are collected en masse, I guess "fair use" could be a hard thing to figure out. At least SD saw the problem with this and took away the option to prompt "in the style of _artist's name_" but that doesn't really stop smaller open source AI developers from training their own models on copyrighted material. We simply need new laws for these situations. I agree with you on AI art and human art being separate in some way. I think that could be the way forward. AI art is art, I agree. Though, comparing it to a chess engine seems a bit misguided. Chess is a game with a clear goal. If the engine shows you the optimal move order to win the game, you can understand the strategy and motivation behind the moves. Art doesn't have a clear goal. And to truly study and learn something you must understand the fundamental aspects of what you are studying. An artist is part of the full process of creation, coming up with the idea, placement of objects/people, compsition, colors, lighting, mood, symbolism... every step that leads up to the finished art piece. Generating an AI image is mostly a shortcut to the end result.