GuShiio.com Wen| Blue Media Collection, author| Ye Er, Editor| Mr. Wei
Overnight, DeepSeek stole the limelight of almost all domestic models.
In the past year, whether it is Kimi, who has made the rounds on the C-side, or Doubao, who has taken the lead, whether it is Wenxinyan, whose daily user activity exceeded 200 million, or Tongyi Qianwen, who has topped the world’s top open source list, compared with the shock DeepSeek has brought to the global technology circle, it is much inferior.
This is not because many domestic models are not effective, but because DeepSeek is too excellent.
In the past, major domestic manufacturers have been discussing how many years there is a gap between OpenAI, but here at DeepSeeK, it is a different story. What is hotly debated in the market is whether DeepSeeK has destroyed OpenAI. The open source line it represents is already forcing OpenAI CEO Sam Altman to reflect: I personally think that we are on the wrong side of history on this issue. Now you need to come up with a different open source strategy.& rdquo;
The emergence of DeepSeek has an impact not only in the industry, but also in the C-end market.
Data shows that in just 20 days after its launch, DeepSeek’s daily activity exceeded the 20 million mark, becoming the fastest-growing AI application in the world. In comparison, it took 244 days for ChatGPT to break the 15 million mark, while DeepSeek only took 18 days. Twenty days after its launch, DeepSeek’s daily activities have reached 22.15 million, which is 41.6% of ChatGPT daily activities users and far exceeding the 16.95 million Doubao daily activities users.
This is an extremely exaggerated AI storm, and what is completely different from the past is that this is an AI storm truly led by China startups.
The question is, why DeepSeek?
You know, in the past two years, major domestic Internet manufacturers have invested heavily on large model tracks, and have also produced many products. The market also generally has expectations, hoping that one of them can catch up with OpenAI in the morning and compete with Silicon Valley AI.
But in the end, it was DeepSeek that broke the game. What the big factory failed to do, it achieved it.
Deep cultivation for a long time
In essence, DeepSeek’s current explosion is a built-up explosion.
Although DeepSeek is a blockbuster this time, its team has been in the AI field for many years. The timeline is even earlier than the big factory, and the layout width and depth are not much worse than the big factory.
Public data shows that DeepSeek was born by the well-known private equity giant Magic Square, and its founder is Liang Wenfeng.
In fact, as early as in college, even at that time, artificial intelligence was still an empty theory without substance, but Liang Wenfeng was extremely convinced that artificial intelligence would definitely change the world.
This has also become his ultimate vision since he started his business.
In 2015, Liang Wenfeng founded Magic Square, which is the earliest company in China to use artificial intelligence for quantitative trading. In 2016, the first trading position generated by deep learning was executed online. In 2017, deep learning technology was fully used for trading.
In 2018, Magic Square’s official website will identify AI as the company’s main development direction and write it into the company’s major events. Another year, Magic Square simply changed its organizational structure and established Magic Square AI. When introducing itself to the outside world, it always said that it is an artificial intelligence company with large-scale in-depth learning basic research and application as the core.
From 2019 to 2021, Magic Square has independently developed Firefly-1 and Firefly-2 AI clusters. Among them, Firefly-2 invested 1 billion yuan, greatly improving computing power support. At the same time, magic squares have also actively recruited a group of algorithm scientists. Founder Liang Wenfeng himself writes and runs code every day.
In terms of technology, we have been steadily reserving reserves, and in terms of infrastructure, we have not fallen behind.
Few people could have expected that when ChatGPT emerged in 2023, the market suddenly discovered that in China, the one with the most high-performance GPU chips was not an artificial intelligence company, but a magic square quantization owned by Liang Wenfeng.
At that time, according to Guosheng Securities Research Report, on the cloud computing side, except for several Internet companies (Shangtang, Baidu, Tencent, Byte, and Ali), only Magic Square had more than 10,000 A100 chips in reserve.
It shows that Magic Square’s investment in AI is not inferior to that of large factories.
anti-routine
There is also the spirit of the DeepSeek entrepreneurial team represented by Liang Wenfeng.
The AI strategies of major Internet companies often rely on existing business systems. Tencent’s AI needs to serve the social and game ecosystem, and Ali’s AI needs to be embedded in e-commerce and cloud computing scenarios. This kind of business collaboration logic can be quickly commercialized, but it also defines the path of technological evolution. The more resources are invested, the more inclined it is to optimize existing models rather than explore alternative approaches.
DeepSeek, which is backed by magic squares, not only has strong financial support, but also has the courage to start from scratch as an entrepreneur and not be afraid of trial and error. This allows DeepSeek to just wade through it along with its belief in innovation.
Regarding innovation, Liang Wenfeng’s attitude is very firm. In the past many years, China companies have become accustomed to others doing technological innovation, and we use it as application monetization, but this is not a matter of course. In this wave, our starting point is not to take advantage of the opportunity to make a fortune, but to go to the forefront of technology to promote the development of the entire ecology. rdquo;
“What we see is that China AI cannot always be in a following position. We often say that there is a year or two gap between China’s AI and the United States, but the real gap is the difference between originality and imitation. If this does not change, China will always be a follower, so some explorations cannot escape.& rdquo;
How to achieve innovation is the counter-routine of abandoning inertia.
The most direct manifestation is in the team composition.
When domestic manufacturers enter large model tracks, they usually tend to go overseas to recruit people, introduce technical giants, quickly build a team, and then work hard. The DeepSeek team is mostly composed of recent graduates from some of the Top local universities. Regardless of experience and qualifications, the criteria for selecting people have always been love and curiosity.
At the same time, in terms of working mechanism, we generally do not pre-divide labor, but naturally divide labor. Everyone has their own unique growth experience, and they all have their own ideas, so there is no need to push them. During the exploration process, if he encountered problems, he would invite others to discuss them. But when an idea shows potential, we will also deploy resources from the top down.& rdquo;
“If there is an idea, everyone can call the training cluster’s card at any time without approval. At the same time, because there is no hierarchy or cross-department, you can also flexibly call everyone as long as the other party is also interested. rdquo;
In other words, the organizational structure of a large factory is essentially a precision operating efficiency machine. But the birth of disruptive innovation requires exactly the loss of control of anti-efficiency.
And DeepSeek is doing just that.
AI Blue Media also asked DeepSeek why the big factory did not make DeepSeeK. The latter said that it was essentially the result of the combined effect of organizational inertia, commercialization pressure and technical path, and said:
This technological revolution triggered by the open source model is forcing big manufacturers to rethink their innovation logic. If it cannot break out of the existing framework, its right to speak in technology may be further weakened.
