Replicating AI systems and using additional GPUs for both inference and training can produce more than proportional gains in effective AI labor, because the additional compute can support more copies and smarter models.

“like that also enables”

“additional sources of economies of scale where if you have like twice the number of GPUs you can run not only twice the number of copies of your old model but then you can train a model that's even better so double your training computes and your inference compute and that means you not only double like you don't get just twice the number of workers you would have had otherwise you get more than that because they're also smarter”

“because you spend more training computes”

“so then that is additional source of economies of scale”