What’s interesting in that statement is that modern llms are largely built off… reinforcement learning from low paid humans.