No the AI race hasn’t been reset. Leading American companies maintain 6-12 month advantages and China releasing these models just raises the floor for everyone and provides no real advantage to China from a security perspective but American companies maintain an important lead.

One could make an argument that they are attempting to “flood the market” with lower priced models to attack American companies that have been investing so much in data center and other buildout but it’s a bad strategy. If it was a good strategy then American companies could just let China get ahead and then do the same copy/paste strategy back to them. But American companies don’t do that because the 6-12 month lead is a meaningful and worthwhile advantage.

I don’t intend to disparage or discount China or the hard work Chinese engineers (why don’t they let in more immigrants? Wouldn’t their models be better by harvesting the world’s talent? Interesting. Anyway.) put into making these advancements but I find the general analysis, pearl-clutching, and fear mongering to be a bit annoying.

Think about it this way. If these open-weight models and such were the threat that others say they are, Americans could just stop training models, wait and let China spend all the money, and then release their own open weight models and “reset the AI race”. Why do we spend money when we don’t have to? Well it’s because the premise is bullshit and it does matter that we have a lead and it does matter that we continue to train models as we are doing.