China Startup of China Deepseek revealed its most recent form, R1, which reports us changed to Ai Technologies. The success has caused a discussion on resources that work in its development, especially using 50,000 NVIDIA GPU. The consequences of this claim is deeply, may change the landscape scenery of research and development of AI.
[Read More: DeepSeek’s 10x AI Efficiency: What’s the Real Story?]
R1 of Deepseek: Efficiency and practice
Deepseek, China’s Ai start based on Hangzhou, has developed AI models named R1 name. The company claims that R1 was developed in just two months at low than US $ 6 million. This figure is lower than competitors spending such as OpenAi, which reports US $ 5 billion in AI development.
In terms of practice, Detossek-R1 show strong results throughout the standards. According to the official release of Deepseek-R1 reached the score of 72.8%, in the MLLU-R1.
However, it is important to note that US $ 6 million worth $ 6 million on the final training series of R1. This figure does not form any other expenses such as a computer basic computer, first training or development and development. As a result, all investments in the development of R1 is higher than the number specified.
[Read More: DeepSeek’s R1 Model Redefines AI Efficiency, Challenging OpenAI GPT-4o Amid US Export Controls]
Expected on using GPU
Deepseek’s computer description for their R1 training is the topic of the debate among the industrial analyst. While Deepseekek reports that use GPU about 2,04 NVIDIA H800 during the two months for the V3 Model, did not have a GPIS release for the R1 version. Some analysts also have doubts about these claims. For example, reports from semianalyseek indicates that Deepseek may have a very extensive infrastructure, with a possibility of 50,000 billion in the development of R1.
[Read More: DeepSeek AI Faces Security and Privacy Backlash Amid OpenAI Data Theft Allegations]
Evolution of the GPU group in AI
The use of a large GPU group of AI became a common practice in the corporate company technology. Some note examples include:
Meta’s AI infrastructure:
-
In March 2024, meta has taken action of two new groups, each with 24,576 NVIDIA H100 H100 H100 H100Designed to support an advanced AI research, including an apprentices such as Llama 3.
-
By October 2024, meta revealed that it is using More than 100,000 NVIDIA H100 H100 Training Llama 4Refers to the major growth of creation of its AI infrastructure.
Supercomputer of Tesla:
-
In August 2023, Tesla Powered On Dojo For use of production and open new AI group training groups 10,000 NVIDIA H100 GPUS. This infrastructure is designed to increase the Tesla Full self driving (FSD) Technology by improving AI training efficiency.
[Read More: Italy Bans DeepSeek AI: First Nation to Block China’s AI Over Privacy Issues]
Deepsek versus PERS Industry
Deepseee reported methods opposite the standard of industry, where the GPU broad GPU is a standard for large AI. While the Company, Meta and Tesla is finely detailed using their GPU, the lack of transparency of their Deepiskk leads to doubt. Some specialist argues that such skills make it difficult to assess the true ability to assess the R1 model of Deepisek.
-
Elong musk: The CEO of Tesla and the spacex has expressed doubts about Deepseek’s claims. He questioned that the possibility of achieving the progress with resources claimed by dealsek.
-
Palmer Luckey: The founder of Oculus VR also expresses doubts about Deepseek’s confirmation, especially on their costs and size of their AI model development. He criticized US media criticism for Deepannek’s claim report, suggesting that the lecture may be influenced by strategic benefits.
-
Stacy Rasgon: Analyst at Rastein, Rasgon has significant about Deepseek security received the least mistake “and we do not think of interest.
-
Atif Malik: The Citi Analyst, Malik has been reported to Deepseek’s reported gpu, suggesting the company’s success or great modeling.
The tech community is also divided in Deepseek’s claims. Some experts praise the potential for more effective AI development, while others express an concerns about lack of transparency. Deepseek’s hardware loss of Deepseek’s hardware infrastructure has made more increased calls to confirm its claim.
[Read More: DeepSeek vs. ChatGPT: AI Knowledge Distillation Sparks Efficiency Breakthrough & Ethical Debate]
Transparent debate: Silence of Deepseek on hardware use
Deepseek’s unwillingness to disclose the details of its hardware infrastructure to question its strategy. The possibility is the company intended to keep the competitive advantage by keeping the confidential resources and methods. In the development industry that AI has been sensitive to calculating, their infrastructure can disclose the potential of the potential.
Another other view indicates that lack of transparency may indicate how highly or incomplete claims. The odd opposite between the specified cost of Deepseek and other Ai giants such as OpenAi and Meta have doubts about doubt. If the form of the company is absolutely developed with a part of a normal investment, it will represent the most important partition in AI, but without predictable.
Alternatively options of Deepseekek can be part of deliberate marketing strategy. By allowing to predict the rotation, the company gives yourself in the center of the industry discussion, may be the interest and investment. The uncertainty around AI model can also attract attention from the program, media, global researchers, the Deeptseek’s depth of the AI space.
Regardless of underlying reasons, lack of transparency transparency to build a reliable challenge. If there is no clear revelation, technology community is also divided – some people separated Deepseek, while others hold its claims with their doubts. Until the company provides uniform evidence on the process of creating infrastructure and training, questions about its true ability to be next.
[Read More: DeepSeek AI Chatbot Exposed: 1M Sensitive Records Leaked, Misinformation Raises Concerns]
This article license
Source: Financial time, Tom’s hardware hardware, Time, Restore, Engineering FB, Register, Newsweek, the verge