Mind-blowing in a scary way, thank you. Elon Musk cannot transform our society or bring in this innovation and attitude culture of the Chinese...he can bulldoze his way to so-called efficiency, but you are spot on: "DeepSeek’s greatest advantage over US labs has always been efficiency. Their engineers are talented, hard-working, smart, and knowledgeable but their defining strength is knowing how to get more done with less."
Alberto, really interesting observations! China has always been good at mandating certain initiatives on a state or province level, as we also saw with consumption incentives during COVID or with attracting foreign capital into the provinces. However, the overall adoption in the private sector still remains to be seen. E.g. in 2024 China had 6x less funding for startups developing AI applications, than the US. Probably it will change after Xi's meeting with TOP entrepreneurs.
DeepSeek's raising external funding are still rumours, as the company itself denies it. However, I personally think it will happen sooner or later because despite DeepSeek's efficiency the AGI enterprise will require MUCH MORE resources than their parent fund has. I wrote an article about it a few weeks back: https://www.deeptech.asia/p/would-you-invest-in-deepseek-if-you
If they don't already, DeepSeek will quickly have access to a much larger GPU cluster--all the GPUs the Chinese government and Chinese companies have stockpiled--to train their next AI model.
Key quotes:
1) DeepSeek "is now seeking external funding for the first time, and alongside other Chinese tech leaders, has recently met with President Xi Jinping."
2) DeepSeek "produced a high-performing model—on par with Grok 3, perhaps slightly worse on popular benchmarks—using the smallest cluster among the world’s top AI labs (none is "small" in absolute terms, though)."
Fascinating. Thanks for sharing. Its really simple - short term greed over long term desire for absolute power… To me personally, the splash of endless power will make sense when individuals will be able to “speak with decentralized AI”
China actually is quite concerned with existential risk and sent out another feeler recently to discuss with the US. And as someone who does speak Chinese, they take it very seriously and its in their draft regulation.
Unlike the US though, they went ahead to work on regulations and testing mechanisms to try to reduce it, as well as efforts for coordination, rather than just give up as in the US.
If anything, it shows regulation doesnt stop innovation and indeed makes society more comfortable.
That's interesting. DeepSeek - at least the CEO - doesn't seem fazed by existential risks. He's very clearly focused on achieving AGI. Probably the landscape there is much more heterogeneous than I've portrayed here. I'd be interested in knowing more about this. Thanks for the link to ChinaTalk, missed this one.
And yes, he is not fazed. But this isnt the entire government which tries to thread between innovation and safety. They do not care for "woke", tho. Its focused around state security and loss of control risks.
Hinton is rationalist-coded. He got his ideas about AI existential risk from them. I really doubt Chinese people got to AI doom independently. They were well behind the US in AI, both in terms of R&D/products and also philosophically. Yudkowsky is the first one who really devoted a lot of time to the question. I'd bet in the entire world.
As far as I know, the term is always "loss of control" with less focus on instrumental convergence. Fudan university recently released a paper on risks of self-replication.
And I disagree, even by 2020, China had a paper on existential risks from AI.
Both terms have historically been used in xenophobic narratives that portray China and its people as an unthinking mass, often to justify fear, discrimination, or political hostility.
“Monolith” implies that all Chinese people think, act, and believe the same things, ignoring the country’s vast ethnic, cultural, linguistic, and ideological diversity. It reinforces stereotypes that dehumanise individuals by erasing differences among them.
“Hive” carries additional negative connotations, often used in a derogatory way to suggest that Chinese people function as mindless, obedient drones without independent thought. This plays into longstanding racist tropes about authoritarianism, collectivism, and the supposed lack of free will among non-Western societies.
These linguistic choices matter precisely because when you reduce a fifth of humanity to an undifferentiated mass, you need not grapple with the complexity of, or even acknowledge the very diversity of thinking that gave us insights like Native Sparse Attention.
It is prejudice masquerading as perspective, and all the more dangerous for wearing the garb of AI punditry.
Well, as a skeptic I am still waiting for America to prove it knows what to do with AI, because you have to agree so far it has not.
Glad the expectations were lowered from ‘AI will replace people’ to ‘give everyone an AI assistant‘.
My problem with the assistant is that one has to be capable of verifying its output, otherwise it’s less useful. It may prove more useful than not having one, but is that really something to ‘make the world look radically different’?
I have been trying to keep an open mind for years, but based on the minor usefulness of LLMs (I will forgo the customary ˋso far’, at some point you have to stop hype-ing) it just looks to me like US and China are just fighting to see who gets nowhere first. If I were a CIO my recommendation would be to ignore LLMs and focus on good software engineering instead.
Europe may not have an LLM but why would they? Europe is probably happy to watch the US (mostly) and China waste their time and money and instead just try to prevent reckless misuse via regulation. We do know that is one of the things LLMs can do.
Does no one else see this fear of being left behind as part of the hype?
So you think the US and China, as countries, know nothing of how to understand, create, and apply technology. That's... quite the belief. Here's my crazy belief: you didn't have the mind as open as you think. You have the LLM Skeptic Syndrome. I wrote about this in my latest weekly review, if you're interested.
Great article and perspectives. This is a great example of how I'm entirely not sure about what will "win at GenAI" or what the future holds.
I can look at what China/CCP is doing and think any number of things: "wow they are making the best of this new tech" | or | "looks like a bunch of pilots across the country" | or | "taking unproven tech and unleashing it on citizens sounds like a lot of fun" | or | "this will accelerate AI development in China" | or | "this won't help at all because simply feeding the model more data doesn't help that much" | or | "what will be the stories about GenAI tools messing up someone's records" ...
I'm not entirely convinced either way that the 'Western' stereotypical model of decentralised, private sector-driven innovation (which isn't actually true given how active central, government funding has been to innovation), or the 'Chinese' stereotypical model of central planning and mandates with provincial autonomy on implementation - will win out or result in better development outcomes. I also have a suspicion that the truth is somewhere in-between.
I don't see dumping further chips and LLMs buckets of money into this direction as the way to achieve autonomous, self-reasoning, automated systems. But maybe this is one direction/step in that journey of forking paths...
What it does feel like is the whole space is clouded by how big, expensive and promising the technology seems to be. That makes everything seem appealing and puts momentum behind anyone adopting aggressively. But that hype holds me back. The bigger and more world-changing the tech seems to be, the more I wonder "is it, really?"
Even if our government is composed of crusty centigenarians who wouldn’t be able to figure out ChatGPT even if it WAS mandated, a parallel exists in corporate culture, where AI is a hot buzz word and everyone wants to integrate it into employee workflows and customer products asap (often whether that’s a good thing or not.)
That’s about what I’d expect. Especially with Musk gutting the government, in our neoliberal and soon to be technolibertarian / technofeudalist society, Corporations are our equivalent to the CCP.
"from having once led the world in living standards (no longer the case)" - Many, myself included, consider the quality of life in Europe to be higher than in the US (and, of course, in China). Since nobody ever introduced "AI" in order to give their workforce a better work/life balance, I don't see how exactly things would get better on that front? Sure it'll help with mass surveillance and crowd control, but that's hardly improving the standard of living (except of those in control, of course).
Sure, you can have a very high standard of living if you've been reaping the fruits of colonies for centuries. Europe won't be the best place to live by 2050 (in some cases, it isn't already; people who've lived in China can attest to the changes there, even with the authoritarian government).
I've been playing around with deepseek for two weeks and it constantly crashes with a server busy excuse (even for my basic demands). I think it's got a long way to go until it rivals chatgpt
This is not "the CCP" doing this, this is China's technology sector....Bureaucrats in Beijing were just as surprised as those in DC and as investors and technologists in the Valley by the success of DeepSeek. This is not/not a top down initiative, it is the Chinese technology sector responding to the success of a very capable model. Lets get the causality right here....
Mind-blowing in a scary way, thank you. Elon Musk cannot transform our society or bring in this innovation and attitude culture of the Chinese...he can bulldoze his way to so-called efficiency, but you are spot on: "DeepSeek’s greatest advantage over US labs has always been efficiency. Their engineers are talented, hard-working, smart, and knowledgeable but their defining strength is knowing how to get more done with less."
Hey FYI you left some editing unfinished at the end of section II.
Oops thanks James. Hate when that happens
That was extremely well written. Thank you!
If you use AI all the time it’ll make you dumb. If you never use AI, it’ll make you even dumber.
Alberto, really interesting observations! China has always been good at mandating certain initiatives on a state or province level, as we also saw with consumption incentives during COVID or with attracting foreign capital into the provinces. However, the overall adoption in the private sector still remains to be seen. E.g. in 2024 China had 6x less funding for startups developing AI applications, than the US. Probably it will change after Xi's meeting with TOP entrepreneurs.
DeepSeek's raising external funding are still rumours, as the company itself denies it. However, I personally think it will happen sooner or later because despite DeepSeek's efficiency the AGI enterprise will require MUCH MORE resources than their parent fund has. I wrote an article about it a few weeks back: https://www.deeptech.asia/p/would-you-invest-in-deepseek-if-you
If they don't already, DeepSeek will quickly have access to a much larger GPU cluster--all the GPUs the Chinese government and Chinese companies have stockpiled--to train their next AI model.
Key quotes:
1) DeepSeek "is now seeking external funding for the first time, and alongside other Chinese tech leaders, has recently met with President Xi Jinping."
2) DeepSeek "produced a high-performing model—on par with Grok 3, perhaps slightly worse on popular benchmarks—using the smallest cluster among the world’s top AI labs (none is "small" in absolute terms, though)."
Fascinating. Thanks for sharing. Its really simple - short term greed over long term desire for absolute power… To me personally, the splash of endless power will make sense when individuals will be able to “speak with decentralized AI”
China actually is quite concerned with existential risk and sent out another feeler recently to discuss with the US. And as someone who does speak Chinese, they take it very seriously and its in their draft regulation.
Unlike the US though, they went ahead to work on regulations and testing mechanisms to try to reduce it, as well as efforts for coordination, rather than just give up as in the US.
If anything, it shows regulation doesnt stop innovation and indeed makes society more comfortable.
https://www.chinatalk.media/p/sb-1047-with-socialist-characteristics
That's interesting. DeepSeek - at least the CEO - doesn't seem fazed by existential risks. He's very clearly focused on achieving AGI. Probably the landscape there is much more heterogeneous than I've portrayed here. I'd be interested in knowing more about this. Thanks for the link to ChinaTalk, missed this one.
Did you also see this?
And yes, he is not fazed. But this isnt the entire government which tries to thread between innovation and safety. They do not care for "woke", tho. Its focused around state security and loss of control risks.
https://www.chinausfocus.com/peace-security/fu-ying-as-long-as-the-us-and-china-can-cooperate-and-work-together-with-the-rest-of-humanity-well-find-a-way-to-keep-ai-under-control
That makes sense. Are their concerns rationalist-coded?
Not that I know of: basic logic that making something smarter than humans is dangerous. If anything, its Hinton-coded and independently arrived on.
https://www.economist.com/china/2024/08/25/is-xi-jinping-an-ai-doomer
Hinton is rationalist-coded. He got his ideas about AI existential risk from them. I really doubt Chinese people got to AI doom independently. They were well behind the US in AI, both in terms of R&D/products and also philosophically. Yudkowsky is the first one who really devoted a lot of time to the question. I'd bet in the entire world.
As far as I know, the term is always "loss of control" with less focus on instrumental convergence. Fudan university recently released a paper on risks of self-replication.
And I disagree, even by 2020, China had a paper on existential risks from AI.
https://www.engineering.org.cn/sscae/EN/10.15302/J-SSCAE-2021.03.005
China as “monolith”, a “hive”? Really?
Both terms have historically been used in xenophobic narratives that portray China and its people as an unthinking mass, often to justify fear, discrimination, or political hostility.
“Monolith” implies that all Chinese people think, act, and believe the same things, ignoring the country’s vast ethnic, cultural, linguistic, and ideological diversity. It reinforces stereotypes that dehumanise individuals by erasing differences among them.
“Hive” carries additional negative connotations, often used in a derogatory way to suggest that Chinese people function as mindless, obedient drones without independent thought. This plays into longstanding racist tropes about authoritarianism, collectivism, and the supposed lack of free will among non-Western societies.
These linguistic choices matter precisely because when you reduce a fifth of humanity to an undifferentiated mass, you need not grapple with the complexity of, or even acknowledge the very diversity of thinking that gave us insights like Native Sparse Attention.
It is prejudice masquerading as perspective, and all the more dangerous for wearing the garb of AI punditry.
Not using them in a derogatory way. I consider that a strength in this case.
Well, as a skeptic I am still waiting for America to prove it knows what to do with AI, because you have to agree so far it has not.
Glad the expectations were lowered from ‘AI will replace people’ to ‘give everyone an AI assistant‘.
My problem with the assistant is that one has to be capable of verifying its output, otherwise it’s less useful. It may prove more useful than not having one, but is that really something to ‘make the world look radically different’?
I have been trying to keep an open mind for years, but based on the minor usefulness of LLMs (I will forgo the customary ˋso far’, at some point you have to stop hype-ing) it just looks to me like US and China are just fighting to see who gets nowhere first. If I were a CIO my recommendation would be to ignore LLMs and focus on good software engineering instead.
Europe may not have an LLM but why would they? Europe is probably happy to watch the US (mostly) and China waste their time and money and instead just try to prevent reckless misuse via regulation. We do know that is one of the things LLMs can do.
Does no one else see this fear of being left behind as part of the hype?
So you think the US and China, as countries, know nothing of how to understand, create, and apply technology. That's... quite the belief. Here's my crazy belief: you didn't have the mind as open as you think. You have the LLM Skeptic Syndrome. I wrote about this in my latest weekly review, if you're interested.
Great article and perspectives. This is a great example of how I'm entirely not sure about what will "win at GenAI" or what the future holds.
I can look at what China/CCP is doing and think any number of things: "wow they are making the best of this new tech" | or | "looks like a bunch of pilots across the country" | or | "taking unproven tech and unleashing it on citizens sounds like a lot of fun" | or | "this will accelerate AI development in China" | or | "this won't help at all because simply feeding the model more data doesn't help that much" | or | "what will be the stories about GenAI tools messing up someone's records" ...
I'm not entirely convinced either way that the 'Western' stereotypical model of decentralised, private sector-driven innovation (which isn't actually true given how active central, government funding has been to innovation), or the 'Chinese' stereotypical model of central planning and mandates with provincial autonomy on implementation - will win out or result in better development outcomes. I also have a suspicion that the truth is somewhere in-between.
I don't see dumping further chips and LLMs buckets of money into this direction as the way to achieve autonomous, self-reasoning, automated systems. But maybe this is one direction/step in that journey of forking paths...
What it does feel like is the whole space is clouded by how big, expensive and promising the technology seems to be. That makes everything seem appealing and puts momentum behind anyone adopting aggressively. But that hype holds me back. The bigger and more world-changing the tech seems to be, the more I wonder "is it, really?"
Don’t worry too much about the implications here.
Even if our government is composed of crusty centigenarians who wouldn’t be able to figure out ChatGPT even if it WAS mandated, a parallel exists in corporate culture, where AI is a hot buzz word and everyone wants to integrate it into employee workflows and customer products asap (often whether that’s a good thing or not.)
That’s about what I’d expect. Especially with Musk gutting the government, in our neoliberal and soon to be technolibertarian / technofeudalist society, Corporations are our equivalent to the CCP.
"from having once led the world in living standards (no longer the case)" - Many, myself included, consider the quality of life in Europe to be higher than in the US (and, of course, in China). Since nobody ever introduced "AI" in order to give their workforce a better work/life balance, I don't see how exactly things would get better on that front? Sure it'll help with mass surveillance and crowd control, but that's hardly improving the standard of living (except of those in control, of course).
Sure, you can have a very high standard of living if you've been reaping the fruits of colonies for centuries. Europe won't be the best place to live by 2050 (in some cases, it isn't already; people who've lived in China can attest to the changes there, even with the authoritarian government).
I guess we'll see - but for sure this will depend on much more than who has the better glorified auto-complete in their bureaucracy.
If Europeans still can't see beyond "auto-complete" then we're already lost.
Well, I didn't need an AI to see that coming :)
I've been playing around with deepseek for two weeks and it constantly crashes with a server busy excuse (even for my basic demands). I think it's got a long way to go until it rivals chatgpt
That's happening because they're probably cutting access to people outside of China while they don't have enough GPUs to cover demand
Ah ok. And I'm in India right now so perhaps that makes a difference.
You can use a US-hosted version of DeepSeek via Perplexity
This is not "the CCP" doing this, this is China's technology sector....Bureaucrats in Beijing were just as surprised as those in DC and as investors and technologists in the Valley by the success of DeepSeek. This is not/not a top down initiative, it is the Chinese technology sector responding to the success of a very capable model. Lets get the causality right here....
I got it right lol. I never said DeepSeek's success was a CCP success