Two years ago, when the large -named Chinese technology companies such as Baida and Alibaba were chasing advances in Silicon Valley in artificial intelligence with splashing announcements and new chatbots, Deepseek captivated a different approach. He reset to research.
The strategy paid off.
The Chinese start-up lasted the technical world with its claim that it has created a powerful AI model that was significantly cheaper to build than the offers of its better-funded American rivals.
In the rivalry between China and the United States over the domination of artificial intelligence, Deepseek seemed to come out of nowhere. In fact, in recent years, it has risen via the Chinese technology world on the way that was something other than conventional.
Its mission to conduct research reflects the mission of companies, such as Openai, Silicon Valley, which in the autumn of 2022 described the American signature over AI. But the similarities usually end there.
Origins Deepseek is in financing, not technology for technology. Its parent company, the Chinese Hedge Fund called High-Flyer, did not start as a laboratory dedicated to humanity from artificial intelligence such as open AI, but as an AI company to bet on the Chinese stock market.
High-Flyer prospered by earning a market dominated by Chinese retail investors who are known to jump in and out. In 2021, High-Flyer found himself under the pressure of regulatory interventions in China on speculative trading, which the authorities in Beijing felt in violation of their attempts to keep markets in peace.
Thus highly Flyer took advantage of new opportunities that they say better with the priorities of the Chinese government: Advanced AI
“We want to do things with greater value and things that go beyond the investment industry, but it was incorrectly interpreted as AI speculation,” said High-Flyer CEO, Lu Zhengzhe new team independent team that is equivalent to the second launch . ”
Deepseek was born. As with many other Chinese start -ups, Deepseek came up with a different business approach market.
The latest Deepseek model for artificial intelligence is considered almost as strong as American rivals, but much more efficient. Her success suggests that AI Silicon Valley has shrunk. The Deepseeka breakthrough, despite the efforts of Washington, to limit the Chinese approach to the advanced chips needed for the AI, raises questions about how effective these controls can be in the long term – although Deepseek’s founder acknowledged that chips are limited.
Deepseek did not rely on the production of AI -oriented products and only this month released his First chatbotWhich allows you to generate text and photos using simple commands. Instead, the company took advantage of the money that earned from trading in Bankrolsky ambitious research. This approach distinguishes it from American rivals, all of whom are eventually consumer technology companies.
This unconventional approach has also made it possible to submit strict regulations that the Chinese government has given on the use of AI by the public. Since he focused on research and sale to businesses that use their model – and until his chatbot’s release this month, not consumer application – his early work has not caused the same government restrictions.
Deepseek is operated by its CEO Liang Wenfeng, a thin, surrounded engineer who studied at Zhejiang University in the eastern city of Hangzhou. He has repeatedly stated in several interviews that he gave the Chinese media that to catch up with US innovations, Chinese companies must devote research to research before profit. Deepseek and High-Flyer did not respond to the request for comment.
What Chinese technology society is “missing in innovations is definitely not capital, but lack of trust and knowledge of how to organize a high talent density to achieve effective innovations,” VA said a widely circulated interview with Chinese technological output 36Kr.
According to interviews and public accounts, those who worked with Mr. Liang describe him as a capable manager with a deep technical background.
“It’s definitely IntP,” said Zihan Wang, a computer engineer who was working on the previous Deepseek model, referring to an introspective type of personality from the Myers-Briggs test, a popular personality test among young people in China. “IntP are really good scientists and have a willingness to explore,” Mr. Wang said. “It’s not one of those people who want to control everything.”
Mr. Liang was not very annoyed by details such as the project’s timelines, and occasionally sent research questions raising ideas to a team of scientists, Wang said. However, it seemed that Mr. Liang mostly led to the development of technology and was not focused on profits.
Unlike many Chinese companies that tend to focus on hiring programmers, Mr. Liang has gained the reputation of employing people outside the computer. Poet and humanities commanders from the best Chinese universities at Deepseek train the model to write classical Chinese poetry and ace with questions taken from a difficult entrance examination at the University of Earth.
“Most of the team graduated from the best universities in China,” said Yineng Zhang, the main software engineer in Basetene, San Francisco, who works on Sglang, a project that is not part of Deepseek that helps people build on the top Deepseek system. “They are very smart and very young.”
For years, Chinese Technology Society has promoted artificial intelligence applications used in computer vision, such as facial recognition. But the edition of Chatgpta Openii caused clearing. When no Chinese society immediately issued nothing comparable, many of them concluded that American companies had a management in advanced AI
In China, computer scientists were determined to prove that they could compete. In 2023, many companies in China released their own models of large languages, a technology that supports chatbots like Chatgpt.
However, creating advanced models would require the use of a large number of chips that would cost hundreds of millions of dollars.
High Flyer also spent. By 2021, it was one handful of Chinese companies that could supply more than 10,000 advanced NVIDIA A100 chips.
Deepseek’s research, however, gave a surprising advantage. Last year, it dramatically reduced the prices charged by developers who build applications using their model, leading the price of war with larger opponents.
Mr. Wang, an engineer who previously worked in Deepseek, said there was little discussion about commercial applications for the technology he was building. Instead, he said that the company was focused on the production of an AI system that could use a number of people for many purposes.
“During my time, we didn’t talk much about how we earn money,” Mr. Wang said. “They just focused on making a great endowment model.”
The key part of Deepseek’s popularity is that she has published the work of her developers. This kind of sharing information called Open Source was the cornerstone of the development of computer software, the Internet and now artificial intelligence.
In the United States, scientists AI and entrepreneurs AI have long since managed the progress of Deepseek technology. Last year, the company turned its heads when it released systems designed to generate their own computer programs.
A new challenge for the company can come up with a new high profile. On the same day, R1, a model behind his new chatbot, last week Mr. Liang appeared in a discussion on a round table with Li Qiang, Chinese Prime Minister.
Deepseek’s sudden popularity has thrown it into the center of the Chinese Communist Party’s efforts to incite innovation, and this could prove to be difficult to handle, said Jimmy Goodrich, head of Rand Corporation, a federally funded Think tank. “For Deepseek, it’s a big misery.” I’m sure they weren’t on the five -year government plan, he said.
“Can they keep this chaotic carefree vision when the party and the world are watching?”
Zixu Wang Research from Hong Kong contributed.
