STOP using small models! just buy 8xH100 and inference your own GPT-4 instance

I feel like the main consequence of barring China from high end compute accelerators is ensuring China massively bolsters the research of efficient, smaller LLMs (and ends up developing domestic chip fabrication in the end too).