The day after Christmas, a small Chinese company called DeepSeek has announced a new AI system that matches the functions of cutting -edge chatbots of companies such as Openai and Google.
That would have been a milestone. However, the team behind a system called Deepseek-V3 explained a greater step. In a research papers explaining how they built technology, DeepSeek engineers have a highly specialized computer chip that major AI companies rely on systems by major AI companies. He said that he only used a part.
These chips are at the center of tense technical competition between the United States and China. The US government works to maintain the national lead in the world’s AI racing, so it is trying to limit the number of powerful chips that can be sold to China and other rivals, as it was created by Silicon Valley Company NVIDIA. I am.
However, the performance of the DeepSeek model raises questions regarding unintended results in the US government’s trade restrictions. Controls were forced to enhance creativity with a wide range of tools that Chinese researchers can freely use on the Internet.
DeepSeek’s chatbot answered questions, solved logic problems, and created a unique computer program, as already on the market, according to benchmark tests used by American AI companies. 。
And it was inexpensive, and the high -tech industry’s largest company (everyone is based in the United States) has challenged a general idea that he can afford to create a state -of -the -art AI system. Chinese engineers have only needed about $ 6 million in raw computing power to build a new system. This is about 10 times less than the high -tech meta spent on building the latest AI technology.
“The number of companies with $ 6 million is much larger than the number of companies that have $ 100 million or $ 1 billion,” said the investor in the venture capital company Page One Ventures. CHRIS V. Nicholson states. AI technology.
Since Openai released the AI Boom in 2022 with Chatgpt’s release, many experts and investors have concluded that no company can compete with market leaders without spending hundreds of dollars on specialized chips. Ta.
One of the world’s leading AI companies train chatbots using a supercomputer that uses 16,000 chips. Meanwhile, DeepSeek engineers have only needed about 2,000 specialized computer chips of NVIDIA.
With the restrictions on chips in China, Deep Shik engineers have forced them that “it may be more efficient and competitive.”
Earlier this month, the Biden administration has issued a new rule to prevent China from getting advanced AI chips through other countries. This rule has been built based on a plurality of previous restrictions so that Chinese companies cannot purchase or create state -of -the -art computer chips. President Trump has not yet shown whether he will maintain the rules or withdraw them.
The US government tried to protect advanced chips from Chinese companies about concerns that could be used for military purposes. In response, some companies in China have stocked thousands of chips, but other companies have procured them from the prosperous underground market.
DeepSeek is operated by a quantitative stock trading company called High Flyer. By 2021, the profit was led to get thousands of NVIDIA chips and used to train previous models. The company, who did not respond to the request of comments, became known in China because of the high -paid promise and the ability to follow the most intriguing research questions.
The computer engineer Zihan Wang, who worked on a previous DeepSeek model, employed people without a computer science background, technology understood poetry and Ace questions on university entrance examinations in China, and asked ACE questions. He also said that he could support the creation.
Deepseek does not manufacture consumer products, leaving engineers to focus completely on their research. This means that the technology is not sandwiched by the most strict aspects of China’s regulations on AI. This requires consumer technology to comply with the government’s management.
Major American companies have continued to move forward in the AI in December, and Openai announced a new “reasoning” system called O3, which exceeds existing technology performance, but is not widely available outside the company. 。 However, DeepSeek keeps indicating that it is not so late. This month, it has released a unique impressive model.
(New York Times complains of Openai and his partner, Microsoft, accused them of copyright infringement of news content related to AI Systems. Openai and Microsoft denied these claims.)
An important part of this rapidly changing global market is the old idea, open source software. Like many other companies, DeepSeek has raised the latest AI system. In other words, we share the basic code with other businesses and researchers. This allows others to build and distribute their own products using the same technology.
Employees of major Chinese technology companies are limited to colleagues, but they work in the open source SGLANG, saying, “If you work in open source, you are working around the world.” Yineng Zhang, the lead software engineer in San Francisco’s base ten, said. project. He supports other people and companies to build products using the DeepSeek system.
AI’s open source ecosystem collected steam when Meta freely shared the AI system called LLAMA in 2023. Many people assumed that this community would prosper only if a company like a meta (a high -tech company with a large data center filled with specialized chips) continues to open technology. 。 However, DeepSeeks and others indicate that they can also expand the power of open source technology. “
Many executives and experts argue that large US companies should not open their technology because they can be used to spread fake information or to cause other serious harm. 。 Some US members of the United States are exploring the possibility of preventing or narrowing the practice.
However, some argued that China would gain significant advancement if regulatory authorities suppressed US open source technologies. When the best open source technology comes from China, US developers argue that they will build a system on those technology. For a long time, it may put China at the center of AI R & D.
“The center of gravity of the open source community is moving to China,” said Ion Stoica, a professor at the California University Berkeley School. “This can be a big danger for the United States.” This allows China to accelerate the development of new technology.
A few hours after taking office, President Trump has withdrawn the order of the Biden administration, which threatened to suppress open source technologies.
Dr. STOICA and his student recently built an AI system called Sky-T1 called Openai O1, the latest system of Openai, in a specific benchmark test. Computing power was only $ 450.
They did this by building on two open source technology released by Chinese technical giant Alibaba.
Their $ 450 systems are not as powerful as Openai technology or DeepSeek’s new systems. And they are unlikely to generate a system that exceeds the main technology performance. However, the project has indicated that even extreme resources can build a competitive system.
REUVEN COHEN, a Toronto technology consultant, has been using Deepseek-V3 since late December. He says it is much cheaper to use it, comparable to the latest system of humanity of emerging companies of Openai, Google, and San Francisco.
“DeepSeek is a way to save money,” he said. “This is a technology that people like me want to use.”
