← Back to context

Comment by gypsy_boots

7 hours ago

It's funny how you can look at obscenely overvalued Western AI companies, with capex's that are unsustainable, and along comes a system which pokes a gaping hole in this model, and the response is "Ya, but what do those sneaky Chinese have up their sleeve here?!"

Perhaps the question should instead be "Why are these Western AI companies getting insane valuations on dubious ROI, and how can these Chinese models run on a fraction of the infrastructure?"

Not saying you're doing this specifically, but I'd be careful with thinking that "company" in China means the same as "company" in America (or in the West more generally).

One thing is that many of these Chinese companies have had to do more with less due to technology export restrictions. Doing more with less is not something that the big US AI companies think about much. The Chinese models are apparently much more compute and energy efficient.

>"Ya, but what do those sneaky Chinese have up their sleeve here?!"

I feel like that is way over simplified. I do not trust the American government. I do not trust the Chinese government. As an American, I believe the Chinese government has a longer and broader history of stealing intellectual property and far less checks and balances than the American government. The current American administration will be gone soon and maybe we can get some sanity back at some point. Overall, I trust my data in America more than China and I think that's reasonable. I am not naive. I'm aware of all the problems with my country and am quite vocal about it. In fact, I think it would be naive to think that the Chinese government won't have complete access to everything that goes through a Chinese server.

We are in full agreement on your second point.

  • > a longer and broader history

    From what perspective? The American colonies repeatedly and flagrantly ignored foreign property and intellectual rights, e.g. via laws to protect domestic but not foreign authors. Samuel Slater was called Slater the Traitor in Britain for a reason.

  • stealing intellectual property == producing higher quality goods for cheaper, because the IP owners are not paid for it

The main cost is training the first version of model. It is very easy to just train the copy cat model based on the output of another model.

  • That doesn't make any sense. If the cost is training, the cost is compute, not data. Distilling another model means paying for that model, which isn't obviously less expensive than crawling the web and curating a dataset. Regardless of where your data comes from, the training process costs the same.

  • this is a justification for why one frontier lab would have extremely high spend rates. There is >1 frontier lab though.

    More explicitly: if the Chinese can get near-leading performance models at a fraction of the cost by stealing from Anthropic, why doesn't OpenAI? Lord knows they could use the extra cash.

>It's funny how you can look at obscenely overvalued Western AI companies, with capex's that are unsustainable, and along comes a system which pokes a gaping hole in this model, and the response is "Ya, but what do those sneaky Chinese have up their sleeve here?!"

The only way this is funny is if you are completely oblivious to how China usually operates.