An interesting quandry: In a world where all ML models are derivative works of the works included in their training sets, said models will still exist because the mere production of a derivative work is not restricted by copyright law. However, the only ones that exist will be those that run in the cloud, because it will be illegal to distribute all of the large models For all that I have concerns about the things happening in the GenAI field (and dislike for a lot of the things being done with them), I can't help but feel thats the worst possible outcome. (Making it illegal to train a model is basically incompatible with the fundamentals of present day copyright law)