In January 2025, a development company of Chinese AI
In depth has published the “Deepseek-R1-Zero” and “Deepseek-R1” inference models as a source, which, according to him, has the same performances as Openai O1. Since then, Deepseek has become a burning subject in the technology industry and has been classified first in the classification of free app stores at the time of writing.
Deepseek gets the Silicon Valley to speak | Techcrunch
https://techcrunch.com/2025/01/26/deepseek-gets-silicon-valley-talking/
Deepseek is an AI development company based in Hangzhou, Province of Zhejiang, China, and announced the “Deepseek-R1-Lite-Prifewe”, “ A large-scale language model specialized for inference, in November 2024. R1-Lite-Prview is a model that makes inference through “thought chains” and has the characteristic of being able to show the various user Chains and “thoughts” in response to the user’s entry and document the process.
In addition, in December, Deepseek announced the large-scale language model “Deepseek-V3”, which At 671 billion parameters and, in some cases, GPT-4O surpasses. In January 2025, Deepseek published the “Deepseek-R1-Zero” and “Deepseek-R1” inference models, trained on Deepseek-V3, as an open source under the MIT license. Deepseek says that “Deepseek-R1” surpasses GPT-4 and Claude 3.5 Sonnet in references, and has equal or better performance than Optai-O1-1217.
“ Deepseek R1 is one of the most amazing and impressive breakthroughs I have ever seen ”. said Marc Andreessen A software developer and co-founder of the venture capital company Andreessen Horowitz.
One of the reasons why Deepseek attracts attention is its low training costs. While large AI development companies spend hundreds of millions of dollars to form models, Deepseek complaints That it costs only $ 5.6 million to form one of its latest models.
Chinese company Deepseek also drew attention to develop a high performance AI model at a time when the United States strongly restricts the export of high performance semiconductors to China. Founder and CEO Deepseek Liang Wenfeng would have said the Chinese Prime Minister Li qiang During a meeting on January 20, the export restrictions of American semiconductors remain a bottleneck.
As Deepseek has become more important in the field of AI, many consumers also try Deepseek AI. Consequently, “Deepseek – Ai” was class Number one in the category of free applications on the App Store at the time of writing the time of the editorial staff.
The rapid rise of the Chinese company Deepseek was a shock for established AI developers, with a person claiming to be a meta-work writing on the blind anonymity platform that the Division generating AI of Meta was in panic mode, analyzing the Deepseek models and trying to copy the best possible.
Neil Khosla, CEO of the AI Health Care Society, Curai Health, said: “Deepseek is a national psychological and economic war campaign by the Chinese Communist Party to make IA less profitable in the United States. They are layer approximately low costs to justify the low price fixing.
“ China Deepseek seems to have built a model of revolutionary intermediary at a cost at very low cost and without access to advanced flea, which could be the greatest threat to the American stock market. ”, ” said Holger Zaepitz economic analyst. “This questions the hundreds of billions of dollars in capital investment paid in the AI industry.”
On the other hand, some welcome the rise of Deepseek. Garry Tan, CEO of the venture capital company y combinator, said: “While training models become cheaper, faster and easier, the demand for inference (real use of AI in the real world) will develop and accelerate even faster, guaranteeing the IT offer to be used.
Yann Lecun, chief scientist of AI in Meta, supported This ascent of Deepseek should not be considered “China exceeding the United States”, but as “open source exceeding proprietary models”. “Deepseek benefits from open research and open source (such as Pytorch and Meta’s Llama). They proposed new ideas and built them on the search for other people. Their work is public and open source, so everyone can benefit from it. It is the power of open research and open source, “he said.