How the tables have turned. It's okay for Anthropic to train their models on copyrighted data, but it's wrong to steal the stolen data from Anthropic models.