Others will disagree — especially “national security” jingoists — but I think Ben has a decent perspective: what to pay attention to vs: what to ignore. I’m particularly taken by his observation that DeepSeek’s relative success is a consequence, rather than in spite of, export controls on China; e.g.
This is something which elsewhere is framed as “the internet perceives censorship as damage and routes around it” but more generally: if there is a lack of resource, the geek community tend to view it as a challenge rather than a limitation — this e.g. is why BitTorrent exists, because upstream bandwidth from ADSL is poor for individuals, but huge if cleverly aggregated, and development of such solutions sometimes leads to new evolutions.
Thus, Ben:
I noted above that if DeepSeek had access to H100s they probably would have used a larger cluster to train their model, simply because that would have been the easier option; the fact they didn’t, and were bandwidth constrained, drove a lot of their decisions in terms of both model architecture and their training infrastructure. Just look at the U.S. labs: they haven’t spent much time on optimization because Nvidia has been aggressively shipping ever more capable systems that accommodate their needs. The route of least resistance has simply been to pay Nvidia. DeepSeek, however, just demonstrated that another route is available: heavy optimization can produce remarkable results on weaker hardware and with lower memory bandwidth; simply paying Nvidia more isn’t the only way to make better models.
Leave a Reply