China stands on large language models (LLMS) in the artificial intelligence space.
Blackdovfx | ITSOK | Getty pictures
China covers open source modes in a Trend market and increases insiders, aI adoption and innovation in the country, and some suggest that some are a ‘Android moment’ for some.
Open source turn The R1 model, the R1 model in the beginning of this year, has formed American technological advantages and raised the questions about the mass spending of large technology in large language models and data centers.
In the R1 sector, creating a spray in the sector due to the requirements of the Sector and low cost requirements, some analysts say that the most important effects of Deepseek are in the adoption of open source models in the adoption of EI models.
“Deepseek’s success proves that open source strategies can cause faster innovation and wide adoption,” Wei Sun, the main analyst of the artificial intelligence performing a large number of companies.
“Now, we see that R1 is actively changing the EI view of China, and the fact that it is actively changed by large companies to open its LLMS in a strategic response to large companies such as Baidu,” he said.
On March 16, Baidu, the latest version of the AI ​​model, Ernie 4.5, as well as a new justification model, ENIE X1, made them free for individual users. Plans in Baidu Ernie 4.5 Model series to make an open source from the last June.
Experts say Baid’s open source plans represent a wider range in China, which is far from a business strategy aimed at the ownership license.
“Baidu always supported his owner’s work model and was vocal against the open source, but DeepSeek’s disorders proved that open source models are proven to compete and be reliable as property.” Lian jye water
Open source and property models
Open source refers to a program where the source is freely available on the Internet for the possible modification and re-distribution of the source code.
AI models that call themselves open sources were existed before Deepseek’s appearance MetaLlama and Google’s Gemma with head examples in the United States some experts As the licenses limit certain use and changes, claim that these models are not really open sources and the training information is not open to the public.
Deepseek’s R1 is described as one of the most perisitive and widely-accepted open source licenses, such as the suns of the opposite-peropoint sun, such as counter-peropoint sun, including commercial purposes.
DeepSeek team is also a “Open source week“Last month, he saw more technical information about the development of the R1 model.
Deepseek’s model itself is free, starting charges for the application programming interface that allows the integration of AI models and their capabilities to other companies. However, the API charges are advertised as it is cheaper than the latest proposals of Openai and Anthrop.
To access some of Openai and anthropical separate models, individual users and enterprises are rechargeable by receiving income. These models are considered a ‘closed source’, their database and algorithms are not open to public entrance.
China opens
Baidu, in addition to other Chinese technological giants Alibaba Group and Tencent It is increasingly providing AI offers for free and is a more model open source.
For example, Alibaba cloud said that last month was open source AI models for video generationTencentin was reported released Five new open source models With the ability to convert text and images to 3D visual in the beginning of this month.
Small players also increase the trend. Manusai, who recently told the AIRTIP agent, said that the AI ​​agent will pass the AI ​​agent to the open source.
“This is not possible without an incredibly open source society, so we are determined to give back,” he said in a product in a product Demo video. “Manusai acts as many agent systems equipped by several different models, so we will open some of these models at the end of this year,” he said.
Zhipu AI, one of the leading AI beginnings in the country this month, said this month would be “Open Source Year” in 2025.
CNBC’s main analysis and founder Ray Wang told CNBC that companies are forced to do these actions after DeepSEEK emerges.
“DeepSEeak is free, it is impossible for any other Chinese rivals to pay for the same thing. They must move to open source business models to compete,” he said.
AI scientist and entrepreneur Kai-Fu Lee also believes that this dynamic would recently influence the statement Social media post When the competition is “free and giant”, it will be difficult to justify the company’s price.
“DeepSeek has won the biggest revelation,” said Lee, China Startup 01.AI, the LLM platform for enterprises that want to use DeepSEEK.
US-Chinese competition
Openai – AI has not stated that the Chatgpt bot in November 2022, the AI ​​has begun to be angry and planned to pass through the owner of the business model. The company that started as non-commercial in 2015 is moving towards a structure for profit.
The sun says Openai and Deepseleek represents very different ends of both AI. It adds that The sector can continue to share between open source players who innovate among each other and indoor source companies that are under pressure to protect high-precious advanced models.
Open source trends, the mass foundations of companies such as Openai questioned. Microsoft invested $ 13 billion. CNBC was approved in the talks to increase the estimates to $ 40 billion in a financing round to raise up to $ 340 billion.
In September, CNBC confirmed that the company had about $ 3.7 billion in profit. Openai CFO Sarah Friar, $ 11 billion in revenue for this year’s company is “absolutely in the realm”.
On the other hand, Chinese companies have chosen an open source route for competing with more ownership approaches to US companies, and Wang of the zodiac research. “They hope to adopt the United States faster than closed models.”
CNBC’s Street Signs Asia, Tim Wang, Tim Wang, Deepseek, said that companies from companies such as Deepseek said.
Wang, open source models lowed costs, the doors opening the door for the product update – something that Chinese companies say are historically very good.
Google calls for “Android moment” development by referring to Android The operating system source code is freely availableDevelops innovation and development in an Apple’s application ecosystem.
“We thought that China’s AI lasted 12-24 months in 12-24 months, and now this is probably three to six months,” said Wang.
However, other experts have lowered the idea that Open Sources should be seen by the AI’s Chinese and the US race. In fact, several US companies have integrated and benefited from the R1 of DeepSeek.
“I think that the so-called Deep Seech is not due to the fact that China’s USA or vice versa is a better AI. This was reported by the President of the group Joe Tsai CNBC for the Converge Conference of CNBC in Singapore.
TSAI added that open source models allow everyone who will be more developed, innovative and the spread of the power to everyone from small entrepreneurs to the great corporations.
