> You can't simultaneously project a hypothetical Chinese capability that it hasn't proven it can do (develop the leading models)
China does lead the frontier. If you're a cybersecurity researcher, a biologist or work in any other censored field, there's nothing that the American models can do for you. The frontier effectively begins where abliterated models can be distributed, and America releases none of them at-scale.
You can't just claim that cybersecurity and biology are non-factors in this race. There are objectively areas where Chinese AI is the only frontier; American regulation is entirely to blame.
> It's like, the thing is so valuable you can't even buy it.
We have a word for this in software actually, it's called "vaporware" in modern parlance. Good luck convincing people that Xenix and Duke Nukem Forever are too valuable for the public to own.
America's stance towards legitimate cybersecurity research is why American businesses like HuggingFace do their incident responses with GLM-5.2. If American AI is "so valuable" that people can't buy the frontier to fix legitimate problems, then China wins by default. It's already happening in several industry-leading American firms, they're not paying Anthropic or OpenAI to fix their real issues.
> Ok we'll just copy their stuff then who cares because apparently bleeding edge model development doesn't matter.
Yes, do this. OpenAI and Anthropic will inevitably turn to the US government for funding, which means that taxpayers will be financially supporting model training for weights that they won't own. If the purpose of AI research is to entrench the monopoly on power, then I would genuinely advocate for all taxpayer funds to be withheld from OpenAI and Anthropic since their products aren't contributing to the public good. There is literally no reason to copy China's state-driven investment strategy if we're not also going to copy the public ownership model. You might as well let China do the frontier research, and let America focus on hand-wringing over the consequences and writing censorship laws to prevent citizens from accessing uncensored Chinese models.
After all, it's where we're headed regardless of what China does. Anthropic, OpenAI and the current US administration have made that clear.
> Why would I trust your comment over Google's business decisions?
My comment doesn't contradict Google's business decisions. It's not an oxymoron to highlight a bad decision as part of a grander AI strategy; but so far you've failed to engage on the actual grounds of FAANG's strategies itself. You're just insisting that American businesses are right because they spend a lot of money, while failing to acknowledge that "spending lots of money" is how most financial crises start. Therefore, it seems appropriate that we acknowledge wasted investments if we want American businesses to stop digging a hole they can't get out of. The onus falls on you to argue that things like OpenCL should have died, or that the Neural Engine was the right choice for millions of iPhone owners.
The point I'm making in this regard isn't complex, you're just offended by the notion of these businesses making wrong decisions. Metal Compute Shaders is not an industry-leading inference solution. It just isn't, the free market already decided. Apple could have increased their addressable market during the datacenter boom if they supported an agnostic GPGPU compute layer like OpenCL in Apple Silicon; or, they could have freed the PCIe lanes on the rackmount Mac Pro to make it a legitimate CUDA host in a world spending trillions on ARM datacenter hardware. Instead they abandoned Khronos group, insisted on standardizing around Metal, and discontinued the Mac Pro.
Apple made some right decisions here. They also burned billions of dollars in R&D and abandoned a trillion-dollar market segment by discontinuing the Mac Pro. Despite being the most advanced ARM core designer in the world, Apple was enamored by B2C service profits and actively avoided any genuine competition that would pressure Nvidia or Apple themselves to invest in frontier research. As the richest American tech company, Apple completely failed to pivot their stack into AI inference even 10 years after CoreML was reluctantly introduced.
Compare this to how, for instance, Huawei works, and you'll see pretty quickly why ego-poisoned businesses like Apple are the antithesis of free market competition. Apple doesn't want to compete, they want to make money. They love ignoring genuinely innovative technology
> I'm not sure why you and others seem to assume that China [...] can take some action and that we can't also take some action in response
Well, China has a centrally planned economy for one. There are objectively things that their government can finance that the United States cannot reciprocally copy.