I think he got it wrong or maybe even backwards, here is a follow up to Dario's Machine of Loving Grace...
Artificial Intelligence thread
- Thread starter 9dashline
- Start date
More likely AI bubble pops causing a recession will do the job instead.
That's some pretty strong claim especially when top Chinese models still lag behind the US in most metrics while training of those aforementioned models still mostly rely on US chips.
I hate when people jump into conclusions as if their victory is a fait accompli meanwhile there is no reason to believe that’s the case. I really don’t know why people are so quick to claim they are guaranteed victory way before it happens. Some sort of caution is always warranted. I don’t see a reason to claim China has won the AI race at all. That’s a bold premature claim
History is never finished, but if you use AI on a daily basis at the enterprise level, you'd come to the same conclusion as Microsoft, Uber, OpenAI, Anthropic, and USG: US AI is f*ed.
But of course most people don't actually do any coding or software development, especially not at enterprise level, so hyping it up with news articles, bench-maxing or just buying out benchmarks is still enough to maintain the fake impression and keep the bubble pumped for another day.
Still, at end of the day people and companies have budgets and you can only brute force your pretense for so long. Chinese AI's ROI has such a massive lead right now, and that's before fully transitioning to Huawei servers, that businesses basically are forced to choose between using Chinese AI or getting wrecked by those who do.
There is no AI race, nobody's getting to AGI anytime soon, but it's also a simple objective fact that as of right now Chinese models are simply superior at every cost level
People who regurgitate Anthropic's distillation talking point fundamentally don't get the basic principle at work: anyone can distill everyone else, but not all models has the same capability or efficiency, the real differentiator is clearly not the data but the achitecture and techniques, and Chinese researchers overwhelming dominate fronteer AI research and publications.
The only reason Anthropic likes to talk about distillation is because data and brute force compute is the only advantage they have
so my one week of testing GLM is that it's much better than latest Kimi on coding. I've been doing all the hard stuff on GLM-5.2 & keeping Kimi for easier coding + research.
anyway, here is Ulanqab aiming to be the Token capital of China.
It's aiming for 200 EFLOPS by the end of this year. Seems not that hard to reach
Deepseek is making a major plea for hiring. I think it made a mistake by not doing the fund raising and hiring sooner.
anyway, here is Ulanqab aiming to be the Token capital of China.
It's aiming for 200 EFLOPS by the end of this year. Seems not that hard to reach
Deepseek is making a major plea for hiring. I think it made a mistake by not doing the fund raising and hiring sooner.