Impressive achievement for OS models.<p>The more time passes, the more it appears that China is likely to eclipse the US in AI development. At the knowledge frontier,Massive parallel sharing of information is going to beat deep vertical organizations almost every time.<p>OpenAI should have stayed open. Looks like greed may cost the US the “AI arms race”.<p>Efforts to throttle China with chip shortages are going to fail in the long term as they will build in larger architectures what they can’t in smaller, more efficient nodes. RISC-v and copies of earlier NVIDIA architectures will be built at massive scale on less efficient nodes, and solar/nuclear/coal power infrastructure will be rolled out to compensate for the difference in efficiency.<p>In the end this will motivate them to create efficiencies algorithmicly where they run into thermal and power limitations, propelling them ahead of countries that have better access too 2-3 nm process nodes<p>They will burn coal and spin up new nuclear plants overnight to power massive infrastructure for training , because the government is getting behind this in a big way. I think it’s shaping up to be the space race of the 21 st century, and China is much more like 1950s USA in focus and drive than the USA of 2025 is.<p>They will brute force what they cannot finesse, and they will find ways to leverage disadvantage to make their efforts more scalable with process improvements.<p>With each new fab generation they can roll out, their backing logistics (power, data centers) will be massive because of their technology lag, and their scaling with each process node implementation will be exponential because the support infrastructure will already be in place.<p>It’s not the same China it was a decade ago.