- Joined
- Sep 13, 2023
- Messages
- 1,767
- Points
- 83
Instead of trying all means to cripple China's rise, the US should collaborate with China and synergise their innovative abilities to benefit mankind. But this would be a pipe dream under the Trump administration.
China’s unstoppable STEM army
A graduation ceremony at the Chinese University of Hong Kong
Sam Dunning
Donald Trump called it a “wake-up call”; Marc Andreessen called it a “Sputnik moment”. By all accounts, the release of DeepSeek’s R1 model was a bombshell. Not only is the Chinese AI model highly intelligent, but it was supposedly far cheaper to build than its American rivals. It may well be a harbinger of many more Chinese tech breakthroughs to come.
When news of DeepSeek’s triumph hit America, the markets descended into pandemonium. In a day, hundreds of millions of dollars were wiped off American technology stocks, with Nvidia leading the fall. Washington had restricted the sale of highly complex chips like Nvidia’s to Chinese companies in the hope of stunting their AI advances — and yet Chinese developers had succeeded even without them.
The naysayers soon piled in. American AI executives, aware that they had been brutally shown up, belittled DeepSeek’s advances. Cold warriors accused DeepSeek of ripping off Western research, pointing to the fact that its model appeared to “think” it was ChatGPT (for which there is a perfectly reasonable explanation), as well as lambasting R1’s reluctance to discuss political subjects that are taboo in China, such as the Tiananmen Square massacre.
They were all missing the point. This was not a fluke, nor the result of foul play, but a show of China’s engineering ingenuity. The real revolution lay in DeepSeek’s willingness to go “open source”. Unlike OpenAI’s products, any designer can look under the hood of a DeepSeek AI model to see how it works. DeepSeek has even released a detailed scientific paper explaining how it designed R1. Western coders are now busy remaking its model, independently and from scratch, using information released to the public.
Another extraordinary aspect of DeepSeek is that its models, once trained, require very little computing power to use. In the industry, this is called “inference”. Up until now, AI companies have tended to rely on vast data centres — energy-hungry buildings humming with interconnected computers — to run their AI models. A company will license out its AI model to other companies — a bank, let’s say, that wants to create a chatbot to field customer questions. For a fee, that bank can then send questions to the data centre, where the AI model will process the questions and send back answers. That’s why you need internet connection to use ChatGPT.
“DeepSeek is really only the tip of the iceberg — one which the US is about to crash straight into.”
DeepSeek works differently. Its free, open-source model can be installed wholesale onto a (slightly specialised) personal computer. A “distilled” or simplified (that is, still as adept at reasoning if not as knowledgeable) version of DeepSeek’s model can then be used and edited offline, for no extra cost. This is a colossal development. It means that any Chinese user wishing to discuss Tiananmen Square would simply need to download DeepSeek onto an offline computer, without fear of a CCP hack, and edit the model. Voilà: you’ll have permanent access to a high-performance AI that will talk to you about any heretical subject you like.
Why would a Chinese company share its secrets in the middle of a global tech race? From DeepSeek’s perspective, there may be a great advantage to remaining open source. Open-source technologies create communities of mutual benefit and learning that benefit everyone involved. In the world of cybersecurity, for example, there is a vigorous and partly open community of cyber-defence analysts all trying to locate, publicise and eventually “patch” up weak points in hardware and software. In such a scenario, there is advantage in collaboration.
Like any community, an open-source community is ruled by hierarchies. A company at the leading edge of such a network is likely to attract collaborators, and gain trust and credit. Since DeepSeek published its models, it has been contacted by dozens of leading AI researchers and companies in the US and China. It will now be busy making powerful new friends, and discussing new techniques. If it can build quickly on its successes, DeepSeek will gain much from sharing its research.
Having cottoned on to this approach, American companies are now scrambling to design their own open-source models. It looks like a new paradigm and has been compared to the release of the first open-source internet browser.
American tech leaders are still right to be alarmed. DeepSeek didn’t come out of nowhere, but has emerged from a vast ecosystem of Chinese tech talent. How many Westerners have heard of CATL, Hesai, DJI, SMIC, Oppo, Pony.ai, Zhiyuan? — among the world’s best companies in robotics, autonomous driving, drones, lidar, AI and battery technology. DeepSeek is really only the tip of the iceberg — one which the US is about to crash straight into.
For one thing, Huawei designs chips now — and it has been on the warpath ever since the US had one of its leading executives arrested in 2018 and then tried to destroy its phones and network technology business. Given DeepSeek’s very low inference costs, it doesn’t depend on sanctioned Nvidia ultra-advanced chips to run. It can work with cheaper Huawei chips — which are anyway growing more sophisticated every year.
Nor is DeepSeek the only up-and-coming Chinese AI company. Other impressive Chinese models have been released recently, such as that of sanctioned firm iFlytek, or Alibaba. All of this has taken place in spite of America’s best efforts to prevent or slow down the rise of Chinese technology. Some have even suggested that American sanctions helped Chinese chipmakers — since sanctions made it profitable to innovate domestic chips instead of just buying American ones.
https://unherd.com/2024/05/why-china-is-winning-the-weapons-war/?=refinnar
A clear narrative is emerging here. The Chinese, quite apart from IP theft, are bloody good at technology. They are not going to stop innovating. America can’t stop them, and they are well past merely “catching up”.
How did it all go wrong for the West? Steve Hsu, an American physicist, argues that China is eclipsing America when it comes to cultivating human capital in the sciences, maths and engineering. Within a couple of generations, malnutrition (which stunts brain development) has been all but eliminated in China; the majority of children now graduate from high school, and roughly 60% go on to university. Nearly a third of students study STEM subjects at university, compared to just 3% a generation ago. And China’s best universities are now ranked as highly as those in the West.
Due to the size of China’s population, these reforms have left the country with a much larger, and still growing, pool of very smart and highly STEM-educated citizens. This pool is much larger than in India, whose comparable population continues to be blighted by illiteracy, malnutrition, and a lack of educational opportunity. And the long era of Chinese brain drain is finally over: more and more, Chinese STEM graduates remain in or return to China rather than trying their luck in the West, their prospects at home bolstered by a dynamic economy and improved standards of living. In fact, the humiliating trend has begun to reverse, with the annual number of scientists leaving the US for China increasing by a factor of five or so in the last 15 years.
Critically, and unlike in the West, China’s STEM graduates gravitate towards jobs in engineering, design, technology, and basic research rather than in finance. DeepSeek itself illustrates the precarity of China’s finance industry — it grew out of a quantitative investment fund that was hit eight months ago by a government crackdown on computer-driven high-speed trading.
Might an aversion to finance be a blessing for China? Steve Hsu likes to quote what billionaire investor Charlie Munger said of America’s elite culture: “I regard the amount of brainpower going into money management as a national scandal. […] We have armies of people with advanced degrees in physics and math in various hedge funds and private-equity funds trying to outsmart the market. […] It’s crazy to have incentives that drive your most intelligent people into a very sophisticated gaming system.”
Not only are Chinese keeping more AI-ready engineers and designers, but theirs have a massive industrial ecosystem to play with — a contrast with deindustrialised America. In April 2022, Sun Ninghai, the director of the Institute of Computing Technology at China’ Academy of Sciences, gave a lecture to the 200-odd strong standing committee of China’s National People’s Congress, which comprises some very senior leaders. He warned that China must not go down America’s path of “contempt for the real economy” and diversion of IT talent into the “virtual economy” — consisting of virtual reality, the meta-universe, blockchain, and the like.
Sun’s argument is a persuasive one. America has neglected making useful physical goods, he says, with the percentage of jobs in manufacturing falling dramatically since 2000. He argues that far too much intellectual energy has gone into the false economy of financial speculation or virtual tech that doesn’t produce tangible goods. Google and Facebook’s revenue comes mostly from advertisements (77% and 98% respectively), while Amazon’s revenue comes from selling Chinese goods, which make up 71% of their products.
By contrast, Sun urged China to apply its AI innovation to the “real economy”. This means integrating AI algorithms into physical mechanisms across industry. The idea is to build AI innovation into China’s successful manufacturing and export-based model, bolstering the government’s existing efforts to surpass America technologically in as many industrial categories and supply chains as possible.
For China’s innovations are already coming in thick and fast. First there was Huawei’s superior 5G, then the PLA’s world-leading hypersonic missiles, which stunned the US military establishment. The story some media missed last week was the discovery of a vast and apparently secret experimental fusion research centre in Sichuan province. If China could invent cold fusion, a means of generating energy by joining atoms together, that would be the ultimate Sputnik moment.
None of this is going to stop. We should expect more stunning advances from Chinese science and technology in the coming months and years. And unless the West begins to nurture and direct STEM talent more effectively, it risks falling behind. We must ensure more incentives for ultra-bright maths graduates; the science faculties at top institutions must be insulated from campus culture wars; policymakers must listen more to scientists and technologists. But, above all, the West must start making things again, fostering the physical and intellectual proximity between industry and research that breeds real advances. We need to get our heads out of the cloud, and into the real world. For the Chinese have entered their era of innovation.
https://unherd.com/2025/02/chinas-stem-army-is-coming/
Last edited:
