An alternative hypothesis: - model performance is plateauing - compute is getting much more expensive - AI data centers are massively unpopular - slowing down AI development is a way to explain slowing progress, reduce spending, and try to regain some goodwill. - This is an effort to save the IPO not humanity.