DeepSeek Revelation

Source: Qianglitan

China has dropped a depth bomb in the AI ​​world. The emergence of DeepSeek has not only shocked the global AI technology field, but also once again set off a revolution that touches all aspects of social economy. While feeling the shock and inspiration of technology again, several "old questions" asked a few years ago have once again surfaced. But this time, finding answers seems to be more urgent: Under this stress test of the impact of intelligent civilization on human economic society, how should we understand and respond? What are the implications of China's innovative industry? How to make the dividends of technology benefit every group?

"Pinduoduo of AI": a technological miracle with low cost and high capability

The rise of DeepSeek is by no means accidental. In a way that is almost a "technological miracle", it has shown the world another possibility of artificial intelligence: low cost and high performance.

Take DeepSeek V3 as an example, its training cost is only 5.57 million US dollars - this figure is enough to make Silicon Valley's technology giants gasp - the annual salary of many senior executives in Meta is far higher than this figure. DeepSeek can create a large model with excellent performance in such a frugal way, which is undoubtedly a huge challenge to the traditional AI R&D model, and it also makes the entire industry begin to reflect: Does the input and output of AI R&D really have to follow the law of "great effort produces miracles"?

And just in the past few days, Fei-Fei Li's team used distillation technology to train an AI reasoning model called s1 at a mere $50 in cloud computing fees. Its performance in math and coding ability tests is comparable to OpenAI's o1 and DeepSeek's R1, pushing "cost control" to a new extreme.

The rise of DeepSeek once again reveals to people the underlying logic of successful large-scale and socialized application of technological innovation: lower cost and greater utility.

There are different angles to evaluate a technology, such as the complexity and difficulty of the technology itself; the difficulty of achieving a certain performance; the pure pioneering nature, etc., which are what technicians are more concerned about. But whether a technology can be commercialized, socialized, or applied on a large scale depends entirely on cost and utility. Throughout history, all successful technological inventions ultimately win by being the first to achieve cost economy, that is, affordable applications. Those technological inventions that are eliminated halfway are often not lost in the technical ability to achieve performance itself, but are surpassed by others in cost control. Historically, Ford Motor is a good example; currently, new energy batteries are also a good example; now, DeeSeek will also show this logic.

Therefore, "great effort to create miracles" is often effective in the short-term breakthrough of technical bottlenecks, but it is very dangerous as a long-term competitive strategy. It is naive to simply regard high cost investment and computing power control as a moat or high barrier to the competition of artificial intelligence.

The shock caused by DeepSeek: The butterfly flapped its wings

The wings first "flashed" the capital market. The sense of smell of capital is still the most sensitive, even too sensitive. Hardware and chip giants such as Nvidia first plummeted, and then the "Jevons Theory" narrative made the market feel that it seemed to be a huge long-term benefit, and computing power would always be scarce - DeepSeek lowered the threshold for AI applications, which is bound to inspire more companies to increase their investment in computing power, thereby promoting the growth of hardware demand.

Regarding computing power, the current narrative of the capital market is superficial. Human beings have unlimited demand for computing power. First, the greater the centralized computing power, the more possibilities it can provide for artificial intelligence. Although DeepSeep's algorithm does not require high computing power, it does not deny the utility of large computing power, so the demand for hardware and chips with high computing power performance is still unlimited. Second, the total demand for computing power for social development and the application of artificial intelligence is unlimited, but the computing power requirements for a single hardware device are limited.

But no matter what type of computing power hardware and chips, while achieving performance (utility), they will face a cost reduction problem in the future, which is what the capital market should really pay attention to.

Then it "fanned" the open source and closed source battle in AI. DeepSeek brought unprecedented pressure to closed source models such as OpenAI. OpenAI had to adjust its strategy and open more permissions to free users; while companies such as Meta, which originally focused on the open source route, could use the experience of DeepSeek to accelerate the research and development of their own models and further consolidate their market position. This struggle between open source and closed source may profoundly change the future development pattern of AI. Perhaps, we can look at this struggle between open source and closed source from a different perspective. Before this, AI was still in the research and development stage, and now it has truly entered the industrialization and commercialization stage. This struggle will determine the development pattern of human AI in the next stage.

In a certain sense, the thinking logic behind the struggle between open source and closed source is the same, that is, both want to dominate the world with a technical model or business model. In this regard, we can draw the following conclusions: 1. If an enterprise, or even a country, wants to dominate the world in a certain field with a technology model or business model, it is impossible. 2. Open source or closed source, as a strategic choice of a business model or technology model, both have the possibility of success. The future development of AI must be a situation where open source and closed source models coexist, but the market share will rise and fall, and the advantages and disadvantages of each enterprise lie in its own capabilities. 3. The breadth and depth of application of a certain technology (including algorithm models) in the market does not necessarily reflect the economic benefits of the enterprise that invented the technology.

But the impact of butterfly wings goes far beyond this: how can we infer its impact on social and economic structures?

More important than the above technological innovations, DeepSeek has greatly reduced the hardware threshold for AI reasoning, making powerful AI services within reach, and truly promoting the "popularization of AI". In the future, lightweight and powerful small models will be everywhere. This is an unavoidable challenge for every social person. The popularity of AI tools has made efficiency, vision and reaction speed the key to competition. Individuals or companies that continue to refuse to use AI tools face a dimensionality reduction attack. The key words here are "efficiency, vision and reaction speed". Don't simply regard AI as a traditional machine or tool that replaces manual labor.

Three years ago, the emergence of ChatGPT caused economists to worry that AI might bring permanent deflation.

The worry is that AI improves productivity but fails to create new demand. In other words, the growth rate of demand is far less than the speed of technological iteration. As a result, companies are accelerating the reduction of manpower and equipment investment. Now, DeepSeek comes with a small model, and the impact is even greater. Especially in the Chinese market, DeepSeek fills the gap of ChatGPT. "Technology" is running all the way, but "demand" is far behind. The gap is widening. Are the previous concerns still groundless?

It’s still the same old question: Are we really ready?

AIand the era of deflation: Anxiety of white-collar workers

It has to be said that in the past few years, due to the rise of generative AI, many workplace ecosystems have changed.

The author has personally experienced subtle changes within a small fund in Hong Kong. In the past, traders and programmers were often at loggerheads because programmers were tired of dealing with code writing and system upgrades, while traders kept urging them because of some seemingly trivial but urgent needs (such as automatic reporting of trading data, functional testing, etc.). After all, time is money for traders. The emergence of AI tools is like a lubricant. Traders can now use AI to solve these problems by themselves.

Now, under the guidance of ChatGPT, traders can write simple programs to handle these problems by themselves, without having to urge programmers every day, and the relationship between the two sides has improved. But this directly led to the company stopping its plan to recruit junior programmers and junior traders, and young people lost their opportunities.

It seems that employment opportunities, at least in similar industries and positions, have been "intercepted" by AI.

Later, some "plot twists" occurred in this matter: programmers jumped to large companies that are vigorously developing AI automatic trading strategies, while traders scoffed at the possibility of AI replacing manual trading, believing that large companies are wasting resources. This example makes people wonder: Did AI "intercept" job opportunities? Create new jobs? Or just shift employment demand?

The relationship between AI and human social economy is very complicated. At present, the industry's impact on it can be summarized into three views: One view compares it to the manufacturing automation revolution in the 1990s, believing that it can greatly improve the productivity of white-collar jobs such as lawyers, accountants, economists, etc. Many people will benefit from it. The second view regards it as a "parlor trick" or a "flash in the pan", believing that the transformation path from laboratory to market is full of challenges, unable to fully play the role of improving productivity, and difficult to become a force for changing the rules of the game.

The third view is the most pessimistic, warning that AI may repeat the blue-collar unemployment wave and put a large number of white-collar workers in trouble - AI can quickly replace or reduce white-collar jobs, and companies will further reduce the need to hire or even purchase equipment. White-collar workers may face the same dilemma as blue-collar workers in the 1990s and early 21st century: large-scale unemployment and reduced income.

As early as three years ago, when ChatGPT was born and caused an uproar, OpenAI and the University of Pennsylvania jointly published a paper in the same year, titled: "GPTs are GPTs: An Early Look at the Labor Market Impact, Potential of Large Language Models" (The potential for early impact of large language models on the labor market)". By classifying job data in the U.S. Occupational Information Network (O*NET) into "routine tasks" and "non-routine tasks", "manual tasks" and "cognitive tasks", the data set is then labeled as "automatable" or "non-automatable" by human experts, and then the data set is trained. Finally, a machine learning model is used to predict whether a given task can be automated by GPT. The conclusion is: the impact is very "universal", covering almost all industries, and all types of work and wage levels from "low-paying services" to "high-skilled professional jobs" - about 80% of workers will have 10% of their jobs "taken away" by AI; large About 19% of the workers will even have more than 50% of their work tasks "taken away".

History tells us that although technological innovation will eventually improve overall well-being, it often exacerbates inequality in the short term. Groups replaced by technology often become the price of development. At present, although AI has shown signs of increased productivity at the micro level, macro data has not yet reflected this change.

Especially at a time when "new demands" for economic development have not been "created" or "stimulated", the losses may outweigh the gains.

Yuval Harari once warned in a speech: AI does not need to have consciousness to manipulate human society. In "Human In the book, Compatible, author Russell mentioned a metaphor: if humans receive an email from an alien civilization saying that it will come to Earth within a month, they may be afraid. However, if they hear that an incomprehensible artificial intelligence will come to Earth within a year, people will not care too much. Both thinkers are implying that people are afraid of alien civilizations, but take AI lightly - its impact will subtly catch human society "off guard".

But in the end, the application of AI is ultimately decided by people. Which jobs are replaced is also a human choice. Trump's re-election was partly due to concerns that American blue-collar jobs would be replaced by foreign workers and machines, and these concerns also occurred under the decision-making of American companies and factory owners. Driven by competitive pressure, internal circulation and FOMO psychology, companies may also accelerate the process of layoffs in the new round of efficiency revolution brought by DeepSeek - but this time the target may be white-collar workers.

If AI leads to an imbalance in wealth distribution and a large number of white-collar workers lose their jobs, social stability will face severe tests. This is not the fault of the technology itself, but the lag of the social governance mechanism. Behind this is still the question that has been asked thousands of times but has made very slow progress: While promoting the development of AI, how do we need to establish a more perfect distribution mechanism to ensure that the benefits of technology can benefit all people?

The only remaining private land for humans: care and "thought experiments"?

Chomsky wrote in a New York Times article a few years ago: AI will never completely replace humans, because humans have a skill that AI cannot match - doing "thought experiments." That is, you can imagine things that do not exist out of thin air. For example, Einstein's study of black holes was imagined by himself, and then he used theoretical deduction to prove its existence. Chomsky believes that AI does not have this kind of thinking experiment ability of "imagining things that do not exist at all and then proving their existence".

So far, thought experiments may be the leading point of human creativity. But in the future, can AI also conduct thought experiments, imagine and prove things that do not exist? If it can be done, humans may have one less piece of their own land.

But there is still an ultimate land: "human care" between humans - AI can never replace it.

Mr. Chen Cunren described the scene of the old Chinese medicine clinic in "Life History in the Silver Dollar Era": bustling patients, doctors led several apprentices, and the apprentices divided the work and cooperated to receive patients, record prescriptions, and assist in massage, acupuncture and medicine. In addition to observing, smelling, asking and palpating, doctors also taught and guided their apprentices orally. The master-apprentice relationship at that time had subtle similarities with today's team of doctors, nurses and interns. However, the biggest difference is that the apprentices can eventually set up their own business and save the world, while the career development of nurses is limited to the field of nursing.

So, will jobs like nurses and secretaries be replaced by ChatGPT? It seems that clerical work such as recording doctor's orders and sorting medical records can be completely handed over to AI; jobs such as ultrasound examinations that rely on image interpretation seem to be easily replaced by AI. But what about helping you lift your clothes, helping you lie down and sit up, and applying warm liquid on your belly? ChatGPT and DeepSeek can't do these things, maybe humanoid robots can do it - but can you feel that it is "worrying" and "caring" about you?

Can it face the fear of surgery for you? Can it help you experience the joy of recovering from a serious illness? Or if you think a little more, can AI help you experience "death"? The emergence of AI will certainly change the landscape of many industries and even reshape our lifestyle. But human emotions, empathy, and the pursuit of the meaning of life are all beyond the reach of AI - and only humans can empathize with each other.

Does the future of mankind really only have "thought experiments" and "human care" as its own private land? No! This is too pessimistic! We must know that the ubiquitous AI in the future and the everlasting mountains, rivers, wind and moon are all the living environment of mankind. Humans create AI and will innovate AI with different technical paths, all for the service of mankind itself. Although AI can replace many human jobs and has more knowledge, skills and wisdom than a single person, it does not mean that it can replace humans, just as there are already many people with high IQs on this earth, but there is still room for other people with high IQs or even low IQs to survive.

People who have "Li Bai's Poems" in their bookcases are different from those who have read "Li Bai's Poems". You travel the Yangtze River and recite a poem. The poem may not be as good as Li Bai's, but it is an improvement in your life realm. Humans have created AI, but it does not mean that people can do nothing, or even necessarily do nothing. For a person or a company, it is not whether they have AI, but whether they can master AI and use it effectively.

Through accumulation: enlightenment of China's innovation industry

The last thought is to shift the focus from oneself to the country: Is DeepSeek "national destiny"?

DeepSeek has also triggered a heated discussion on China's innovation and technology industry. Public opinion is polarized, with praise and doubt intertwined, and national pride and technological anxiety coexisting. Some people regard it as a "Xiaomi plus rifle" victory under technological blockade, while others question its innovation, worry that it relies on banned graphics cards or data infringement, and even advocate that technology should not be open source.

The success of DeepSeek is an important step in the development of global AI and a manifestation of technological innovation under the open source model. Its success is also based on learning and drawing on the technology of predecessors. Innovation is a process of accumulation and inheritance, which cannot be separated from the wisdom of predecessors. And it is precisely the open source success of DeepSeek that provides valuable experience for the development of China's large model industry.

In the global AI arena, competition and cooperation coexist. The emergence of DeepSeek is just the beginning. We have reason to believe that China will play an increasingly important role on the global AI stage. Strengthening international cooperation and making full use of global technical resources and talent advantages are important bridges for development; actively participating in open source projects and communicating and cooperating with the world's top developers can not only learn advanced technologies, but also share Chinese wisdom and enhance international influence. China's large model industry, and even the broader innovative industry, technological upgrading, and the development of new quality productivity should all seek and progress with such an attitude.