Tit for tat. US administration thought they would contain China. Now Deepseek v4 is running on their home grown chips. US companies lost that hardware revenue and dependency instead.
A specific case in which the US demanded a tiny fraction of what China does systematically, orders of magnitude worse. Okay. Didn't even take that into account to begin with but that's an even larger "tat" then I guess.
It's really hard to tell. Almost all the models have the benchmarks in their training data, which pushes us into the realm of basing model capability rankings on vibes. I think the OSS models tend to do worse on things outside their corpus, but Deepseek specifically has done insanely good work on efficiency and scaling, which is verifiable in a way capabilities are not.
