Was it ethical for Anthropic/OpenAI to train their models by gobbling a treasure trove of copyrighted material?