Two years ago, when large Chinese technology companies such as Baidu and Alibaba were chasing Silicon Valley’s advances in artificial intelligence with announcements and new chatbots, Deepseek got a different approach. Research machine.
The strategy has attributed.
The Chinese start has pulled the technological world on its claim that it has created a powerful AI model that was significantly cheaper to build on the offers of its best funded US opponents.
In the rivalry between China and the United States about the sovereignty of artificial intelligence, Deepseek seemed to come out of nowhere. In fact, it has been so far through the Chinese technological world in recent years with a path that was only conventional.
Its mission to seek research reflects that of companies such as Openai, Silicon Valley, which marked an American signature for AI in the fall of 2022, but the similarities are mainly there.
The origin of Deepseek is in funding, not technology for technology. Its parent company, a Chinese risk capital called High-Flyer, did not start as a laboratory dedicated to the preservation of humanity by AI such as Open AI, but as a business that uses AI to bet on the Chinese stock market.
High-flyer had been developed by exploiting a market dominated by China’s retail investors, who are known for the jump and stocks. In 2021, High-Flyer was found pressured by regulatory repressions in China for speculative transactions, which authorities in Beijing felt contradicted their efforts to keep the markets calm.
Thus, High-Flyer followed a new opportunity that stated that he was better aligned 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 have been misinterpreted as AI shares,” said High-Flyer CEO Lu Zhengzhe, told 2023. A new team independent of investment, which is equivalent to a second start. ”
Deepseek was born. As with many other Chinese startups, Deepseek came to an established market with a different business approach.
The latest model of Deepseek’s artificial intelligence is believed to be almost as powerful as American opponents but much more effective. His success suggests that Silicon Valley’s AI lead has shrunk. The discovery of Deepseek, despite Washington’s efforts to limit Chinese access to the advanced brands required for AI, raises questions about how effective these controls can be long -term – though the founder of Deepseek recognized that restrictions on Chips is a restriction.
Deepseek was not based on the production of AI for revenue and only this month released its first chatbot, which allows anyone to create text and photos with simple commands. Instead, the company used the money made by the high-flyer by the trading of shares in Bankroll ambitious research. The approach has put it in addition to our opponents, who are ultimately consumer technology companies.
This unusual approach also allowed Deepseek to bypass the strict regulations that the Chinese government has used the public by the public. Because its focus was the research and sale of businesses that use its model – and, until the liberation of Chatbot this month, not for consumer applications – its early work did not cause the same government limitations.
Deepseek is run by its CEO, Liang Wenfeng, a slim, landscaped engineer who studied at Zhejiang University in the eastern town of Hangzhou. He has repeatedly said in the few interviews he gave to Chinese media that in order to cover American innovation, Chinese companies need to research before profits. Deepseek and High-Flyers did not respond to comments for comments.
What Chinese technology companies “do not have innovation is definitely not capital, but the lack of confidence and knowledge of how to organize a high density of talent to achieve effective innovation,” he said in a widely circulated interview with Chinese technology .
Those who worked with Mr. Liang describe him as a capable director with a deep technical background, according to interviews and public accounts.
“It’s definitely an Intp,” said Zihan Wang, a computer engineer who worked on a previous Deepseek model, referring to an endoscopic personality type from the Myers-Briggs test, a popular personality test among young people in China. “Intps are really good researchers and they are willing to explore,” Mr Wang said. “It’s not one of those people who want to control everything.”
Mr Liang did not disturb much with details such as project schedules and occasionally sent research questions that cause thinking to the entire team of researchers, Mr Wang said. But for the most part, Mr Liang seemed to promote technology and did not focus on profits.
Unlike many Chinese companies, which tend to focus on recruiting developers, Mr Liang has gained a reputation for employing people out of calculation. Poets and humanitarian studies from China’s leading universities in his staff Deepseek train the model to write classic Chinese poetry and ace questions received by the difficult entry into the country’s college.
“Most of the team graduated from China’s top universities,” said Yineng Zhang, a software engineer at Baseten in San Francisco, who works on Sglang, a project that is not part of Deepseek that helps people rely on top of the Deepseek system. “They are very smart and very young.”
For years, Chinese technology companies have pioneered artificial intelligence applications used in computer vision, such as facial recognition. But the liberation of Chatgpt by Openai caused an estimate. When no Chinese company immediately released anything comparable, many concluded that US companies had a leading role in Advanced AI
In China, computer scientists were determined to prove that they could compete. In 2023, many companies in China have released their own large linguistic models, technology that supports chatbots such as chatgpt.
But 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 of the few Chinese companies that managed to store over 10,000 advanced Nvidia A100 brands.
However, Deepseek’s research gave him an amazing advantage. Last year, he dramatically reduced prices that charges developers who build applications using his model, causing a price war with larger opponents.
Mr Wang, the engineer who worked in Deepseek, said there was little debate on commercial applications for technology they were building. Instead, he said, the company focused on producing an AI system that could be used by a number of people for many purposes.
“During my time there, we didn’t talk much about how we make money,” Mr Wang said. “They just focused on constructing a large foundation model.”
A crucial part of Deepseek’s popularity is that it has done the work of its developers. This type of sharing of information, called Open Source, is a cornerstone of the development of computer software, internet and now artificial intelligence.
In the United States, AI researchers and businessmen have long followed the progress of Deepseek technology. Last year, the company turned its heads when it was released systems designed to create their own computer programs.
A new challenge for the company can come with its new high profile. The same day the R1 was released, the model behind his new Chatbot last week, Mr Liang appeared in a roundtable talk with Li Qiang, the Prime Minister of China.
Deepseek’s sudden popularity has pushed it into the center of the Chinese Communist Party’s efforts to promote innovation and this could prove to be difficult to manage, said Jimmy Goodrich, a senior consultant for the analysis of technology in Rand Corporation tank. “It’s a great difficulty for Deepseek. I am sure it was not in the five -year government plan,” he said.
“Can they maintain this chaotic carefree vision when they watch both the party and the world?”
Zixu wang He contributed research by Hong Kong.
