Deepseek: All you need to know about the AI ​​Chatbot app

Dibsic virus has gone.

Deepseek stormed the Chinese artificial intelligence prevailing this week after that Its Chatbot app has risen to the top of Apple App Store ((And Google Play, as well). Deepseek models of artificial intelligence, which have been trained using providing account techniques, He led Wall Street – And technicians – To ask whether the United States can maintain its progress in the artificial intelligence race and whether the demand for artificial intelligence chips will maintain it.

But where did Depsik come from, and how did it rise to international fame so quickly?

Deepseek merchant assets

Deepseek supports high capital management, a Chinese quantitative hedge box that uses artificial intelligence to inform its commercial decisions.

An artificial intelligence enthusiastic Liang Winfing Participated in the establishment of a high table in 2015. Wenfeng, who was reported to be trading while a student at Zhejiang University launched Flyer Capital Management as a hedge box in 2019 focusing on developing and publishing AI’s algorithms.

In 2023, Deepseek began as a dedicated laboratory to search for artificial intelligence tools separate from her financial work. With high mutations as one of its investors, the laboratory set out to his private company, which is also called Deepseek.

From the first day, Deepseek built its data center collections for models training. But like other artificial intelligence companies in China, Deepseek was affected by the American export ban on the devices. To train one of its most modern models, the company has been forced to use NVIDIA H800 chips, which is a less powerful version of the chip, H100, available to American companies.

Deepseek’s technical team is said to tend to Young. Company It is said that the recruits are strong Doctorate researchers are one of the most important Chinese universities. Deepseek also rented people without any computer science background To help its technology better understand a wide range of topics, according to the New York Times.

Strong Depsic models

Deepseek revealed its first collection of Deepseke Coder, Deepseek Llm and Deepseek chat-in November 2023. But that was not until last spring, when Startup released the Deepseek-V2 family of the next generation, which began to consider the artificial intelligence industry.

The performance of Deepseek-V2, a system of text and images analysis for general purposes, was well in various artificial intelligence standards-and it was much cheaper for operating than similar models at that time. The local competition for Deepseek, including Bytedance and Alibaba, has been forced to reduce the prices of use for some of their models, and make others completely free.

Deepseek-V3It was launched in December 2024, it was added only to the reputation of Dibsic.

According to DeepSeek’s internal test, Deepseek V3 is both both both, and is openly available like Meta’s Lama And “closed” models that can only be reached through the application programming interface, such as Openai’s GPT-4O.

Equally impressive, “Thinking” model in Deepseek. It was released in January, claims Dibsic R1 performs as well as the O1 Openai model on the main criteria.

Since it is a model of thinking, the R1 is effectively dividing the facts, which helps it avoid some of the pitfalls that are usually on the models. Thinking forms takes a little longer-secondly to a longer minutes-to reach compared solutions with an unbalanced model. The upward trend is that they tend to be more reliable in fields such as physics, science and mathematics.

There is a negative side for R1, Deepseek V3 and other Deepseek. Being the Chinese Amnesty International, it is subject to Measurement By the Internet organizer in China to ensure that its responses “embody basic socialist values”. In the Deepseek Chatbot app, for example, R1 will not answer questions about Tiananmen Square or Taiwan’s independence.

Sabotage

If Deepseek has a business model, it is not clear what this model is, exactly. The company regrets its products and services much lower than the market value – and gives others free. It also does not take the money of the investorDespite a lot of VC attention.

How to tell Deepseek, enabled her that efficiency breakthroughs have enabled her to maintain the competitiveness of the intense cost. Some experts Dispute The numbers provided by the company.

Whatever the situation, the developers have taken Deepseek models, which are not open source because the phrase is common but available under ease licenses that allow commercial use. According to Clem Delangue, CEO of Huging Face, one of the platforms that host Deepseek models, The embracing developers created more than 500 “derivative” models from R1 That achieved 2.5 million downloads combined.

Dibsic’s success was more and more firmly rivals It is described as “lifting artificial intelligence” and “More than that.” The company’s success was at least responsible for The NVIDIA share price decreased by 18 % In January and Devoting a general response From Openai CEO Sam Al -Tamman.

Microsoft It announced that Deepseek is available in the AZURE AIMicrosoft platform that combines AI services for institutions under one banner. When asked about the influence of Dibsic on Meta’s Amnesty International spending during the first quarter profit call, CEO Mark Zuckerberg said Amnesty International’s infrastructure will remain a “strategic advantage” For meta. In March, Openai called Deepseek as “state and” controlled by the state “,”, “. It recommends that the US government think about banning models from Deepseek.

During the fourth quarter profit call, NVIDIA, CEO Jensen Huang confirmed the excellent “innovation” of Deepseek, “ The saying that it and other “thinking” models are great for NVIDIA because it needs more account.

At the same time, Some companies prohibit DeepseekAnd as well as all Countries and Governmentsand Including South Korea. New York State also Deepseek prevented it from using it on government agencies.

As for what the future of Dibsic might hold, it is not clear. Futured models are given. But it seems that the United States government It is increasingly cautious about what it considers to be a harmful foreign influence. In March, the Wall Street Journal reported this The United States is likely to prohibit government agencies.

This story was originally published on January 28, 2025, and will be updated regularly.