The Copyright Office's report on generative AI training is superb - thoughtful, thorough, and clear in rejecting the idea that all gen AI training is fair use. A few things jumped out: 1. A use is less transformative if it ultimately serves the same purpose as the original. 'Transformative', in the context of fair use decisions, doesn't just mean 'something is transformed'. A use is less transformative if it serves essentially the same purpose as the original. They give the example of training audio models on sound recordings: if you do this to create new sound recordings that compete in the same market as the originals, this is clearly less transformative. 2. Being 'transformative' is not enough anyway. Even when AI training is transformative, the market effects on the original can outweigh this in the fair use evaluation. 3. Market dilution by gen AI weighs against fair use. This is critical. They say: "The speed and scale at which AI systems generate content pose a serious risk of diluting markets for works of the same kind as in their training data. That means more competition for sales of an author’s works and more difficulty for audiences in finding them." They argue that this weighs against a fair use finding, and they remind us that this fourth factor of fair use - the effect on the potential market for or value of the work that's copied - is "undoubtedly the single most important element of fair use", according to the Supreme Court. Many of us have been saying this for a long time, shouted down by a chorus of AI boosters. It's great to see the US Copyright Office - the government body responsible for providing guidance on copyright law - agreeing. copyright.gov/ai/Copyright-a…