Faisal Bashir Lightrockket | Gety pictures
Dibsic in China has become the largest theme in technology this week, as many have focused on industry and Wall Street on the number one: $ 6 million.
In Deepseek paper About the latest artificial intelligence model, the company said that the total training costs amounted to 5.576 million dollars, based on the rental price Nafidia Graphics processing units. Deepseek included a clear warning, saying that the number included only “official training” of the model and excluding the costs associated with “research and attempts previously on structure, algorithms or data.”
Early in the week, Deepseek assistant took over the desirable place for the most downloaded free application in the United States appleApp Store, Chatgpt from Openai. Global technology shares were sold, with NVIDIA chips and Broadcom loss Completely 800 billion dollars in the market on Monday.
A A new report from SemianalysisThe semiconductor research and consulting company added more context to Deepseek expenses. The company estimated that the spending of Deepseek devices “is higher than $ 500 million on the history of the company,” adding that the costs of research, development and the total cost of ownership are large. Semianlysis wrote that the generation of “artificial data” of the model to be trained requires “a large amount of account.”
The report said that Claude 3.5 Sonit of the human cost is “$ 10 of millions for training”, but he pointed out that the Antarubor raised billions of dollars in dollars. Amazon and GoogleIndicating the amount of funds required to operate forms and company.
“Because they have to try, reach new structures, collect data and clean them, pay employees, and much more,” said Semienaelik.
The private Deepseek paper does not include its account costs. The company did not immediately respond to a request for comment.
“To be unique Deepsek is unique in that they have achieved this level of cost and capabilities first.” The company added that Deepse's R1 is “a very good model” and that “catching up with the edge of thinking this is objectively impressive.”
Experts and analysts described this week the quality of the Debceic model, and they noticed the extent of admiration in the United States Exporting a supplement chip To China three times in three years. This led to fears that the United States is behind its main opponent in the market It is expected that the highest $ 1 trillion In revenues during a decade.
Bernstein's analysts wrote in a note on Monday that “according to many (sometimes hysteries), we saw (during the weekend) the effects of anywhere” this is interesting “as we know it.”
Deepseek was founded in 2023 by Liang Wenfeng, co-founder of the High-Flyer, a quantitative hedge box focusing on artificial intelligence. According to what AI Startup originated from the AI Research Unit for the hedge fund in April 2023 to focus on large language models and access to artificial general intelligence, or AGI – a branch of artificial intelligence that equals or exceeds human thought in a wide range of tasks, and this is Openai and others follow .
Deepseek is still fully owned and funded by high models, according to analysts at Jefferies.
Deepseek began to capture Steam earlier this month, when Startup R1 released the thinking model that competes with Openai's O1. It is open source, which means that any Amnesty International developer can use it.
Like other Chinese Chatbots, Deepseek has restrictions on some topics: When he was asked about some of the Chinese leader Shi Jinping's policies, for example, it is said to Dibsic The user directs away From similar lines of interrogation.
Sam Germany, CEO of Openai, praised the model publicly, but the company also said it believes that there is evidence that Deepseek Incorrect Openai data to build its product.
At an event in Washington, DC, on Thursday, which Openai hosted, Altman said that Deepseek is clearly a great model. “
“This is a reminder of the level of competition and the need for a democratic victory,” he said. He also indicates “the level of interest in thinking and the level of interest in the open source.”
He watches: The CEO of Nvidia Jensen Huang and President Trump meets the policy of artificial intelligence
