Is the upshot of this that algorithmic progress needs to stall in order for OpenAI to become profitable? Even if cost of GPUs/electricity goes down, as long as new techniques matter, there will always be the competitive necessity to spend as much you can training models, lest your models fall behind