I never said I thought training AI with the copyrighted work of others causes harm to others. If anything, I think training is analogous enough to human learning that it’s a gray area. However, I think there are different ethical concerns with AI training data than there are with piracy, and those concerns mostly arise from the profit being made from the models.