Open China embraces the traditional wisdom about artificial intelligence

China is adopting an Amnesty International Open Source Models in Trends Market Observer and the Opinion they say it enhances the adoption of artificial intelligence and innovation in the country, with some that it is the moment of Android to the sector.

Open source transformation A leader was carried out by Ai Startup Deepseek, which was released earlier this year challenge American technical hegemony and ask questions about the huge Big Tech's huge spending on large language models and data centers.

While the R1 was created in this sector due to its performance and low cost demands, some analysts say the most important effect of Deepseek was stimulating the adoption of open source AI models.

“The success of Dibsic proves that open source strategies can lead to faster innovation and wide adoption,” said Wei Sun, the main analyst of artificial intelligence in corresponding research.

“Now, we see that R1 reshapes the scene of artificial intelligence in China, with large companies such as Bido Moving to opening their LLMS source in a strategic response, “she added.

On March 16, Bido It released the latest version of artificial intelligence Model, Ernie 4.5, as well as the new thinking model, Ernie X1, making it free for individual users. Bido also plans To make the Ernie 4.5 series an open source model from the end of June.

Experts say open source plans are a broader shift in China, away from the work strategy that focuses on ownership license.

“Baidu has always been very supportive of the royal business model and was explicit against the open source, but the disturbances like Deepseek have proven that the open models of sources can be competitive and reliable like royal models,” Previously said cnbc.

The artificial intelligence models called themselves were open source before the appearance of Dibsic, with DeadHowever, GEMMA from GEWAMA and Google are preliminary examples in the United States, Some experts It argues that these models are not really open source because their licenses restrict some uses and modifications, and that their training data groups are not general.

The R1 Deepseek is distributed under the “MIT License”, which is described by Sun Counterpoint as one of the most nail licenses and building widely for wide dependence, which facilitates unrestricted use, modification and distribution, including for commercial purposes.

The Deepseek team was even set up.Open source week“Last month, which witnessed the issuance of more technical details about developing its R1 model.

Although the Deepseek model itself is free, starting starting fees for the application programming interface, which allows integrating artificial intelligence models and capabilities in other companies' applications. However, its API fees are announced to be much cheaper compared to the latest Openai and anthropic offers.

Openai and Anthropic also generates revenues by imposing fees on individual users and institutions to reach some of their models. These models are “closed source”, as their data sets, and algorithms are not open for public access.

In addition to Baidu, other Chinese technology giants like Ali Baba Group and Tinetnet Their artificial intelligence offers have been increasingly served, making more open source models.

For example, Alibaba Cloud said last month that she was open source Artificial Intelligence Models to generate videoWhile Tinent was released Five new open source models Earlier this month with the ability to convert the text and images into 3D images.

The smaller players are also strengthening the direction. Manusai, a Chinese company, Amnesty International, recently revealed You have an agent This claims that he is superior to Openai's deep research, he said that he would turn towards the open source.

“This will not be possible without an amazing, open source society, and for this reason we are committed to returning.” Experimental video. “Manusai works as a multi -agent system, supported by several distinct models, so later this year, we will open the source of some of these models,” he added.

Zhipu Ai, one of the leading startups in the country, announced this month in WeChat that the year 2025 will be “the year of open source”.

Ray Wang, the main analyst and founder of Constellation Research, told CNBC that companies had had to take these moves after the appearance of Deepseek.

“With Deepseek Free, it is impossible for any other Chinese competitors to impose fees on the same thing. They have to move to open source business models for competition,” Wang said.

Amnesty International researcher and businessman Kay Fu Lee believes that this dynamic will affect Openai, noting recently. Social media yet It will be difficult for the company to justify its prices when the competition is “free and huge”.

“The biggest revelation of Deepseek is that the open source has won,” Lee, whose Chinese youth 01.AI built a LLM platform for institutions seeking to use Deepseek.

Openai – which began artificial intelligence when she released BOT ATCHGPT in November 2022 – has not noted that she plans to move from her own business model. The company, which started as non -profit organizations in 2015, is heading towards a For profit building.

Sun says Openai and Deepseek represent very different ends of artificial intelligence space. You add that The sector can continue to see the partition between open players who create each other and closed source companies that have been pressured to maintain high -cost sophisticated models.

Set open direction to question the huge funds collected by companies such as Openai. Microsoft has invested $ 13 billion in the company. In talks to raise up to $ 40 billion in A. Finance round This would raise his evaluation to $ 340 billion, as CNBC stressed at the end of January.

ISeptember, confirmed CNBC The company expects about $ 5 billion in losses, with revenue revenue of $ 3.7 billion. Openai Cfo Sarah Friar also said that revenues are $ 11 billion.Certainly in the world of possibility“The company this year.

On the other hand, Chinese companies have chosen the open -source road because they are competing with the most intelligent approach for American companies, Wang said constellation. “They hope to adopt faster than the closed models of the United States,” he added.

Talk to CNBC'sStreet brands Asia“On Wednesday, Tim Wang, the administrative partner of the hedge fund that focuses on technology, said that models of companies such as Deepseek were” great empowerment and complications in China, “which shows how things can be done with more limited resources.

According to Wang, open source models have led to low costs, and opened the doors to innovate products-which says that Chinese companies were very good.

The development is called “Android”, referring to when Android from Google has done Operating system source code is available freelyEnhancing Innovation and development In the ecosystem of the application other than Apple.

“We are used to believe that China was late 12 to 24 months (the United States) in artificial intelligence and now we believe this may range from three to six months,” said Wang.

However, other experts have underestimated the idea that the open source artificial intelligence should be seen through the lens of China and the American competition. In fact, many American companies have Integrated and benefited From Deepseek's R1.

“I think the so -called Deepseek's moment is not about whether China has a better Amnesty International than the United States or vice versa. It is really related to the power of open source,” the head of the Ali Baba Jo Tsey Group told CNBC in Singapore earlier this month.

Tsai added that open source models give the strength of artificial intelligence to all of the young entrepreneurs to large companies, which will lead to more development, innovation and the spread of artificial intelligence applications.