What is Deepseek, Chinese Ai startup who sends shock waves via Global Tech? | Technology

Deepseek, a little known Chinese startup, sent shock waves through the global technology sector by publishing artificial intelligence models (AI) whose capabilities submit to Google and Openi creations.

Deepseek-R1 Creator says his model has been developed using less advanced and fewer computer chips than those employed in technological giants in the United States.

In a research work published last week, the development team of the model said they spent less than $ 6 million on a computer power to train Model-Djelić multi-million dollar AI budgets enjoyed by US technological giants like Openi, Apcebet and Meta.

Marc Andreessen, one of the most influential technological risk capitalists in the Silicon Valley, welcomed the publication of the model as “Ai’s moment Sputnik”.

The sudden appearance of a small Chinese startup that is capable of submitting the best players of the silicon valley has challenged the assumptions of the US dominance in AI and raised fear that market assessments of companies such as Nvidia, alphabet and target could be separated from reality.

On Monday, Nvidia, which holds almost a monopoly in the production of the semiconductor generated by AI, lost nearly $ 600 billion market capitalization after its shares fell 17 percent.

US President Donald Trump, who posted last week to launch a $ 500 billion initiative led by Openii, Texas based in Texas and Japanese Softbank, said Deepseek should serve as a “awake call” about the need for the US industry ” Laser focused on a winning competition ”.

What is Deepseek?

Deepseek, based in Hangzhou, was founded by Liang Weenfeng, a serial entrepreneur who also runs a HEDGE High Flight Fund, at the end of 2023.

Although little is known outside China, Liang has an extensive history of combining growing technologies and investments.

In 2013, he faced the Hangzhou Jacobi Investment Management, an investment company that hired AI in the implementation of trade strategies, along with Ko-Alumin University of Zhejiang, according to the Chinese media outlet of Son Finance.

Liang continued to establish two more companies focused on computer-focused investments-Hangzhou Huanfang Technology Co and Ningbo Huanfang Qualitative Investment Partnership 2015 and 2016.

In an interview with Chinese media outlets in 2023, Liang rejected the proposal that it was too late for the startups to be involved in the AI ​​or that it should be considered forbidden expensive.

“The reproduction itself is relatively cheap-theme on public works and an open code code, minimal training times, or even sufficient adjustments. The study, however, includes extensive experiments, comparisons and larger computer and talents, “Liang said, according to the translation of his comments published by The Chinatalk Supsack.

Liang said his interest in AI was conducted primarily with “curiosity”.

“From a broader perspective, we want to confirm certain hypotheses. For example, we assume that the essence of human intelligence could be language, and human thought could basically be a language process, “he said, according to a transcript.

“What you consider” thinking “could actually be your brain weaving language. It suggests that agi similar to a man can get out of great language models,” he added, referring to artificial general intelligence (AGI), a type of AI that tries to imitate cognitive abilities of the human mind.

Deepseek did not immediately respond to the comment request.

On Monday, Gregory Zuckerman, a journalist from The Wall Street Journal, said that he learned that Liang, whom he had not heard before, wrote a preface to the Chinese edition of the book about the late manager of the American Hedge Fund Jim Simons.

“Simons obviously left a deep impact,” Zuckerman wrote in the column, describing how Liang praised his book as a Torah that “reveals many previously unresolved mysteries and brings us many experiences we will learn from.”

“Even my mother didn’t get that much out of the book,” Zuckerman wrote.

Why did the Deepseek storm take the technological world?

Simply put, the success of the company raised existential questions about AI approach that is taken by both Silicon Valley and the US Government.

It was assumed that US technological companies had a critical advantage in AI, especially because of their huge size, allowing them to draw top talent from all over the world and invest huge amounts in the construction of data centers and buy large amounts of expensive high-end chips.

Deepseek’s arrival on stage challenged the assumption that billions of dollars at the forefront of AI.

“Openii was founded 10 years ago, has 4,500 employees and raised $ 6.6 billion in capital. Deepseek was founded less than two years ago, has 200 employees and has been developed by less than $ 10 million, “said Adam Kobeissi, founder of the Kobeissi market analysis on Monday.

“How are these two companies now competitors?”

In their research work, Deepseek engineers said they used about 2,000 NVIDIA H800 chips for training the model, which are less advanced than the most modern chips.

The team said he used more specialized models that work together to allow slower chips for more effectively analysis of data.

For the US Government, Deepseek’s arrival on stage asked questions about his attempt to restrain Chinese AI progress by limiting the exports of top chips.

Deepseek’s research work suggests that or the most advanced chips are not needed to create high-performance of AI models or that Chinese companies can continue to perform chips in sufficient amounts or a combination of both.

The California Nvidia H800 chips, which were designed to be in line with US export control, were freely exported to China until October 2023, when the management of the then President Joe Biden added to their list of limited cases.

In the 2023 interview with Waves, Lian said his company had supplied 10,000 NVIDIA A100 GPU before they were banned for export. The GPU -O or graphics units are electronic circles used to accelerate the graphics and processing of the image on computer devices.

Tanishq Abraham, former director of research in AI stability, said he was not surprised by the Chinese level of progress in AI, given the presentation of various models by Chinese companies such as Alibaba and Baichuan.

“Although there were limitations in China’s ability to get GPU, China still managed to innovate and squeeze out performance from what they have,” Abraham told Al Jazeera.

“I think it’s a lesson to US companies that there is still a lot of performance I can pull out.”

Tara Javidi, co-Director of the Mechanical Engineering Center, Calculation and Safety at San Diego California University, said Deepseek was excited about “rapid progress” in the development of AI around the world.

“My only hope is that the attention that will be given to this announcement will stimulate greater intellectual interest in the topic, further expand the talent pool and, last but not the least important, increase private and public investments in the US research in the US,” he said, “he said,” he said, “he said,” he said Javidi said Al Jazeera

In the meantime, the confidence of investors in the American technology scene hit – at least in the short term.

In addition to the dramatic slide of Nvidia, on Monday, Google Matiiona Alphabet and Microsoft recorded on Monday that the shares prices drop 4.03 percent, or 2.14 percent, although Apple and Amazon ended up more.

“If Deepseek’s costs are real, now almost every major organization is in any company that can build it and host it,” said Tim Miller, a professor specializing in AI at the University of Queensland, for Al Jazeera.

“So in this sense the game has completely changed because there is a new” rule “that anyone can play.”

Does this mean that China wins in the AI ​​race?

Not necessarily.

Although technological analysts are widely agreed that Deepseek-R1 acts at a similar level as a chatgpt-or even better for certain tasks-it moves quickly.

Openi Executive Director Sam Altman said earlier this month that the company would publish its latest AI model AI, O3 Mini, within a few weeks after considering the user feedback.

On Monday, Altman admitted that Deepseek-R1 was “impressive” while defending his company’s focus on greater computer power.

“Obviously we will deliver a lot better models, and it is also legally empowering that we have a new competitor! We will pull some editions, “Altman said on X.

“But we are mostly excited to continue to execute our research plan and believe that more computations are more important than ever before to succeed in our mission.”

Abraham, a former director of research in AI stability, said perceptions can be distorted by the fact that, unlike Deepseek, companies like Openii did not make their most advanced models available to the public.

“Deepseek made his best model available for free for use. Openai’s best model, on the other hand, is not free, “he said.

“So most people who use the Chatgpt for free are shocked by Deepsek and believe that there is a huge jump able to have had a similar model that has been paid for several months. This wall of the border AI model salaries leads to people do not understand the progress and abilities of AI. “

Miller, professor at the University of Queensland, said Deepseek’s progress and other recent events suggest that China is at least “up” with the USA in AI.

“At the end of last year, I made a somewhat rejected prediction that the next scientific breakthrough in AI could come from a small player like a single university researcher who doesn’t have much access to a lot of computer power – they should be smarter to compete,” he said.

“The obvious progress of Deepseek is almost an example of this: if they do not have enough computer forces to build a model as big as Chatgpt, they had to be smart. Necessity is the mother of invention. “