Elephant News reporter Li Chang and contributing writer Liu Yuling
With the rapid development of global artificial intelligence (AI) technology, OpenAI, as a world-renowned AI research institution, has a nerve in the technology industry every decision. "On July 9, OpenAI will terminate the provision of API (Application Programming Interface) services to China", the revelation of this news caused an uproar in the technology community.
On July 4, the 2024 World Artificial Intelligence Conference will be held in Shanghai. China has a large market scale and rich application scenarios, which provides a broad space and conditions for the landing and application of large models, which is an opportunity for China to catch up with and surpass the United States when it is slightly inferior to the United States in terms of underlying R&D technology. OpenAI's "suspension" allows the domestic large model to switch back to the "Chinese mode", is this a "block" or "assist" for domestic manufacturers?
OpenAI "stops serving" domestic large models switch back to "Chinese mode"
OpenAI artificial intelligence is called the "fourth industrial revolution", and whoever can stay ahead of the curve will be able to occupy the top of the world's science and technology. At present, in the field of artificial intelligence, the United States and China are leading the world off a cliff and are in the first echelon.
As we all know, AI chips can be called the heart of artificial intelligence, but they are quite expensive, and many AI companies simply can't afford to burn this money, and more importantly, the United States directly bans the sale of related chips to China. This means that domestic AI companies cannot train large models like American AI companies, and can only train with low-level chips, and the training level determines that the gap will become bigger and bigger. However, at present, the ability of domestic self-developed large models, second only to the United States, is gradually showing competitiveness.
However, according to industry analysts, the gap between China and the United States still exists for at least one to two years, and the gap is likely to widen. Even so, in order to suppress the development of Chinese artificial intelligence, the United States still "moves", first let the AI chip overlord "Nvidia" ban the export of AI chips to China; Now is the time for OpenAI, the overlord of AI models, to stop providing API services to China.
The API discourse is the "Application Programming Interface", which is a software intermediary that allows two applications to talk to each other. "The gap between the domestic self-developed large model and ChatGDT-4 is already very small." Sun Yang, co-founder of Shanghai Qiansui Network Technology Co., Ltd., believes that although the "blocking order" will usher in a big blow to some companies that start up with the help of OpenAI's large model. "OpenAI is not unusable, as long as there are overseas servers, it can be used. It's just that it's more troublesome and will incur more costs. ”
Is OpenAI's "suspension" a "block" or an "assist"?
In addition to the trouble and cost issues, "OpenAI's failure to open up to China is actually two actions. It turned out that the web version was not allowed, and now even the docking method has been banned, judging from the situation in the industry, in fact, the impact is still very large. As a state-owned enterprise, Li Mugang, director of the transportation, cultural and tourism industry of the government and enterprise department of China Telecom Yunnan Company, which has been paying close attention to AI projects, believes that this move is a "hard decoupling" of the United States from China in terms of AI scientific research. Because from the perspective of the development system of AI in the two countries, the relationship between innovation and application was before, and the two are actually inseparable from each other. After the "hard decoupling", if China wants to compete in the entire AI industry chain, the cost of investment will be very huge.
"Where will the resources go? At the national level, state-owned enterprises and central enterprises must assume this social responsibility. The generative large model, which mainly consumes computing power, is mainly represented by the graphics card chip, that is, the computing power card; Computing power cards are now monopolized by 'NVIDIA', but we need to build a lot of computing power to be able to support this form of generative large model. Li Mugang said that once the computing power is put in, it will inevitably be powered on, and once it is powered on, it means that a lot of energy will be consumed to make it run, regardless of whether there is business or not, it must run empty. So far this year, China Telecom has built four 10,000-calorie-level computing centers, which consume millions of electricity per day. To catch up with the speed of the United States, Li Mugang believes that higher requirements will definitely be put forward for domestic research and development.
At present, most domestic AI manufacturers have a positive attitude towards OpenAI's "ban", believing that challenges and opportunities coexist. On June 26, the founder and chairman of 360 Group@周洪祎 posted a video on Weibo, saying that he believed: "OpenAI's suspension of services to China can only accelerate the development of China's own large model industry, which is not necessarily a bad thing", he explained, "OpenAI's API cannot be called, which forces domestic applications to choose only domestic large models, and the gap between domestic large models and GPT has gradually narrowed." ”
"I have a similar view to most people in the industry in China about OpenAI no longer opening up to China, and I think it is good for China's large model research." Sun Yang said in an interview with Elephant News that this stage will be painful, but it will be the same as Huawei's before. But the market has been caught, and it is indeed an opportunity for domestic large-scale model manufacturers.
Domestic manufacturers "grab orders" to open "moving plans"
After OpenAI's "suspension", domestic AI large model companies almost started a "collective action" - "migration plan", and in one fell swoop, they played the banner of "free moving". At present, including Alibaba, Baidu, Tencent, Baichuan Intelligence, Zhipu AI, etc., a number of domestic leading large model manufacturers have fed the "moving plan" to realize the transformation of the technology base to the user's mouth. For example, Alibaba Cloud Bailian announced that it will price the API of Qwen-plus, the main model of Tongyi Qianwen GPT4, at 1/50 of OpenAI, and provide Chinese developers with 22 million free tokens and exclusive migration services.
In addition to the leading manufacturers, several representative start-ups of domestic AI large models are also vying for this wave of traffic. SiliconFlow even makes Qwen2-7B, GLM-4-9B, Yi-1.5-9B and other open source models free forever, founder Yuan Jinhui said in the circle of friends: "Today, SiliconCloud has made a large number of models free, everyone take it lightly, don't go bankrupt." ”
Obviously, domestic large-scale model companies are trying to seize this critical "window period" to win more developer users for themselves. The "moving plan" of the large model company is not only to seize market share, but also to demonstrate its technical support and service capabilities, which helps to quickly build user trust. In addition, in terms of price, domestic large models are also more affordable. In Alibaba's free migration plan, the API pricing of the qwen-plus large model is only 1/50 of OpenAI's.
Sun Yang said that this incident is an opportunity for most domestic self-developed large models to switch the "English mode" back to the "Chinese mode". In fact, in the Chinese context, domestic self-developed large models have shown significant advantages, especially in dealing with Chinese natural language tasks, these models can better understand and generate Chinese content, so as to perform well in Chinese Q&A, dialogue, translation and other tasks.
Although the domestic large model has begun to approach GPT4 in the context of Chinese, GPT4 is also constantly developing new versions, so it is still in a catch-up situation. Sun Yang pointed out that domestic large-scale model research still needs to pay close attention to future development trends: "We have to pay attention to one problem," OpenAI's 'ban' may be just the beginning. ”。