Some insist that Chinese models are a few generations behind, how many probably depends more on patriotism rather than fact.
Those people typically also insist that Chinese models are just distillations and often neglect to see how many of these companies contribute to the theory of designing efficient and capable models. It is somehow thought that they will always trail US models.
Well. i would say look at recent history. China worked up the ladder of manufacturing from simple, bad stuff to highly complex things - exactly what westerners then claimed they’d never be able to. Then as that was conquered, westerners comforted themselves by insisting that China could copy, but trail-blazing would always still be our thing. Well, Baidu and Alibaba face scaling issues few western companies do and BYD seems to match Tesla or VW just fine.
I am unsure why anyone would think US models are destined to remain in the lead forever.
At “best”, I see a fragmented world where each major region (yes also Europe) will eventually have their own models - exactly because no one wants to give any competitive power a chokehold over their society. But beyond that, models will largely be so good that this “generation”/universal superiority idea becomes completely obsolete.
PRC pureplay AI only companies has same problems as openAI, that's not the same as huge tech companies like Baidu or Alibaba or Tencent (i.e. Google/Microsoft tier) who can afford to lose money on AI. And ultimately they are also not sinking 100s of billions in capex, they can't even if they tried due to sanctions. Their financial exposure is magnitude less, i.e. it matters if you're losing 500m a year vs 5 billion per year especially as systemic economic contagion risk - PRC and US bubble sizes as % of economy not the same.
US models might not be "destined" to stay in the lead, but I see no reason to believe that won't at the moment.
Betting on the trainwreck is quite easy, you got nothing to lose in the analogy, while shorting companies will cost you something, most times a lot if the bet has the wrong timing.
The particular bet I linked to is probably a bad idea though, because there is a causal link between OpenAI doing well and deciding to go public. So this is not the way to bet on it crashing and burning.
If OpenAI is worth $5B, 4% of MSFT market Cap is Open AI.
ARK Venture Fund (ARKVX) holding is 7.2% of its total but also has xAI, Anthropic and lots of other AI
https://www.ark-funds.com/funds/arkvx#hold
OpenAI going bust might be a shock to shareprices of publicly traded companies like Oracle, CoreWeave, Softbank and the like
EDIT: obviously if OpenAI is worth $500B, not 5
The media has been "social media'd", with everything being driven by algorithms, everything being about capturing attention at the cost of everything else. Negativity sells. FUD sells.
> Remember all the "no more data" craze? Despite no actual researcher worth their salt saying it or even hinting at it?
We ran out of fresh interesting data. A large chunk of training needs to generate its own now. Synthetic data training became a huge thing over the last year.
> Remember the "hitting walls" rhetoric?
Since then the basic training slowed down a lot and improvements are more in the agentic and thinking solutions, with lots more reinforcement training than in the past.
The fact we worked around those problems doesn't mean they weren't real. It's like people say Y2K wasn't a problem... ignoring all the work that went into preventing issues.
No, we didn't. Hassabis has been saying this for a while now, and Gemini3 is proof of that. The data is there, there are still plenty of untapped resources.
> Synthetic data training became a huge thing over the last year.
No, people "heard" about it over the last year. Synthetic data training has been a thing in model training for ~2 years already. L3 was post-trained on synthetic-only data, and was released in apr24. Research only was even earlier with the phi family of models. Again, if you're only reading the mainstream media you won't get an accurate picture of these things, as you'd get from actually working in this field, or even following good sources, read the key papers and so on.
> The fact we worked around those problems doesn't mean they weren't real.
The way the media (and some influencers in this space) have framed it over the last year is not accurate. I get that people don't trust CEOs (and for good reasons), but even amodei was saying there is no data problem in early interviews in 25.
- Opus 4.5 for software development. Works faster, and tends to write cleaner code.
- GPT 5.2 xHigh for mathematical analysis, and analysis in general (e.g. code review, planning, double checks), it's very meticulous.
- Gemini 3.0 Pro for image understanding, though this one I haven't played around with much.
Will it be in time to recoop capex.