Source item
Open original source ↗Quoted context
These are the quoted author’s words.
Top three open source AI labs are all from China: DeepSeek, Qwen, and now Minimax (based on @ArtificialAnlys benchmarks) China appears to have some structural advantages in open source AI. Not just my assessment, Ion Stoica's too. Stoica is a star Berkeley professor, who co-founded Databricks, Anyscale, and most recently LM Arena -- constantly traversing the academia-industry corridor There are three structural advantages vs the US: - Experts: 50% of AI researchers in the world as Jensen likes to say (more precisely 47% as @damienics likes to correct him) and experts in different industry verticals that are data heavy - Data: both general and industry-specific - Tight collaboration between academia and industry: per many government and university policies, in order to advance as an aspiring professor, you can't just publish papers, but must have industry impact too! Since the default mode in academia (everywhere) is open source, more industry AI labs are also defaulting to open source as the R&D model. Faster diffusion inherent to open source is a major plus too. US has advantage in AI hardware and infrastructure. But since leading labs are all proprietary, there is no sharing of knowledge, labs do redundant things, and (I suspect) a good % of AI capex in the US is to support this redundancy. US has a ton of AI talent too, of course. The academia-industry collaboration and sharing among that talent pool is facilitated via acquihires, corporate poaching, VC funding startups, and (now) $100 million signing bonuses. It's the opposite of open source, for better or worse.
kevinsxu ↗