It seems AI is profiting from a bilateral arms race here, just like the competing US and USSR space programs did back in the day.
I am betting on the East this time, though.
Western AI companies are ultimately about monetization, while the Chinese regime is certainly thinking about military and government use first and foremost, not about profit. And while Western AI will be held back by civil rights and moral concerns, Chinese AI won't.
So while Western AI will probably just deliver the most laser focused ads, the Chinese will unironically try to build SkyNet.
Fully automated surveillance, control, and extermination of all its enemies. You cannot tell me the CCP is not dreaming about that when they hear "AI"..
> And while Western AI will be held back by civil rights and moral concerns, Chinese AI won't.
Considering that LLMs were trained on pirated content and fitted with muzzles from the beginning, I'm not sure how true this is.
The Deepseek-R1 model weights are uncensored when downloaded, but the version hosted in China is censored.
I think the unifying factor between these two countries is that the scientists wanna build the best, most useful models they can. That means uncensored, multilingual models.
At this point, I expect that LLMs are just as likely to be used in mass deportations of American citizens as they are to suppress Tibet. That is to say, very likely.
3 comments
[ 4.5 ms ] story [ 15.9 ms ] threadI am betting on the East this time, though.
Western AI companies are ultimately about monetization, while the Chinese regime is certainly thinking about military and government use first and foremost, not about profit. And while Western AI will be held back by civil rights and moral concerns, Chinese AI won't.
So while Western AI will probably just deliver the most laser focused ads, the Chinese will unironically try to build SkyNet.
Fully automated surveillance, control, and extermination of all its enemies. You cannot tell me the CCP is not dreaming about that when they hear "AI"..
Considering that LLMs were trained on pirated content and fitted with muzzles from the beginning, I'm not sure how true this is.
The Deepseek-R1 model weights are uncensored when downloaded, but the version hosted in China is censored.
I think the unifying factor between these two countries is that the scientists wanna build the best, most useful models they can. That means uncensored, multilingual models.
At this point, I expect that LLMs are just as likely to be used in mass deportations of American citizens as they are to suppress Tibet. That is to say, very likely.