Most of this law actually sounds very reasonable. It explicitly bans a bunch of ways companies and states have been looking into using AI to abuse people, and c

Most of this law actually sounds very reasonable. It explicitly bans a bunch of ways companies and states have been looking into using AI to abuse people, and categorizes a bunch more as high-risk to ensure greater scrutiny. I'm not quite sure how to interpret the mention of copyright compliance. Is it imlying the EU has decided that training on protected works constitutes a copyright violation, or is it merely affirming that if copyright law is ruled to apply then it will apply? Besides that it's the "general intelligence existential risk" rules that are concerning to me. They set a hard arbitrary cutoff on how much compute can be used to train a model before it is subject to a bunch of extra requirements, and reserve the right to declare any general AI a risk and demand changes. Note that the definition of a "general AI" is incredibly broad, and by my reading covers all LLMs.