“For nearly a decade, the training compute for the largest AI systems has consistently grown by approximately half an order of magnitude (0.5 OOMs) per year.”