China's strategy of open source frontier models seems bizarre. Releasing the model and telling everyone (the inference providers) that they can just host it on their own infrastructure seems like a guaranteed loss for China. So why do they do it?
My guess is that their goal is to heavily disrupt the US AI market. Anthropic & OpenAI run private models at their own set prices. If they have competing models at even 90% of the quality, but at half the price, going into profitability is going to be extremely hard. They're fighting against inference providers who specialize in owning the hardware, with little to no spend on R&D. The inference providers just take the open Chinese models and battle against Anthropic and OpenAI on their own turf.
The more the Chinese can weaken the power of these gigantic American powerhouses, the more the US stock market & investments suffer. Without a gigantic amount of cash flowing into these companies, they will be unable to keep subsidizing their products & pushing R&D forward. In the end, China pays the cost of R&D (which, when they distill the models, isn't really all that bad) while they wipe out billions, possibly trillions of dollars out of the US stock market. It's a genius idea.
On US nationalization of AI, it's very clear why we did it. Right now, literally NO other country other than the US and China have actively developed the frontier. Japan launched a fusion model which just combines the best US models, Koreans haven't put out anything good, Taiwan is busy making the chips, and so on. It's only the US vs China. Mistral is irrelevant (sadly).
So what's the point of trying to carry our allies with us? From every possible angle there is absolutely no use to it. A better idea is to just take the top talent from our allies and bring them to our AI research companies. Spoiler alert, by paying these top talents egregious amounts of money and giving them Visas, it's already taken place.
Is the US in the right here? It's messy, as you can see. It makes no sense to try and funnel cash into a joint R&D program with other countries. We already take their top talent because of the nature of cash and talent in the US.
Do we want China to win? In the short term, I'd say you'd have to look past an extreme amount of risk to say yes. I'd like to remind everyone that the US is the greatest success of political philosophy and freedom in the modern era. I say this as a person of Chinese descent.
China is brutal. Tourism there may present it as a great country, but it's really just a horrifying place to live if you ever cross the government. There is little crime, but that's because there are cameras in every little crevice. You're unable to meet in groups past a curfew. On every anniversary of Tiananmen Square, police round up around the block and you're unable to even enter the area. If you're suspected of being a "rebel" against the government, you can't travel between the provinces. And don't even get me started on the schools. When I visited China, every single student I met told me that it was their dream to leave China and go to the United States. It's insane what the internet has made China out to be.
The US may be "burning bridges" between its allies, but I say (with much respect) that our potential allies are completely useless at the moment.
I love using Chinese models, they're great and are a needed source of competition to the US models. But do not forget that the US needs to win. If China wins and the US fails, it is really over.
TLDR: China open-sources its models to weaken the US stock market and investment in AI. They eat the cost of R&D in order to attack US investments. The US has no reason to drag its allies with it in some sort of AI coalition. The US and China are the only relevant powers in the AI battle. You do not want China to win the AI battle, even if they provide a needed source of competition right now. China is not some morally superior country in the AI space. If they could, they'd close off their models, but it's just a better strategy to open source them for now.