China News Service, Beijing, February 25 (Reporter Xia Bin) From the beginning of last year when ChatGPT set off a craze for artificial intelligence (AI) large models, to the beginning of this year when Sora surprised the world with Vincent video technology, international cutting-edge artificial intelligence technology has iterated rapidly.

How should China take its own AI path?

  "I think China still has advantages." Zhou Hongyi, a member of the National Committee of the Chinese People's Political Consultative Conference and founder of 360 Group, said in an interview with China News Service and other media that it is entirely possible for China's large AI models in vertical fields to surpass GPT-4.

  Many Chinese companies have started the "Battle of 100 Models", the Central Economic Work Conference has made it clear to "accelerate the development of artificial intelligence", and the State-owned Assets Supervision and Administration Commission of the State Council has held a special promotion meeting on artificial intelligence for central enterprises... China is actively embracing the AI ​​development trend.

According to data from the Ministry of Industry and Information Technology, the scale of China's core artificial intelligence industry has reached 500 billion yuan, with more than 4,400 companies and nearly 10,000 digital workshops and smart factories built.

  From the perspective of corporate vitality, policy dividends, technology patents, R&D motivation, and industrial strength, China is in the "first echelon" of the global artificial intelligence field.

However, Zhou Hongyi also reminded that there is indeed a gap between China and the United States, which is the leader in this field, mainly in the direction of original technology.

  "This is an objective fact. Only when you see the gap will you know how to catch up." He pointed out that regardless of the Transformer model or Sora, they are still software in nature. "Now (China) is lagging behind for about one to two years, and it can be solved." .

  Zhou Hongyi further explained that the most difficult thing about technology is to find an original direction. "Whoever chooses the right direction and the right architecture will follow suit."

As long as China has a clear correct structure, subsequent progress will be rapid, because Chinese companies have a strong learning ability.

  After the general direction is clear, you can also "find another way."

Zhou Hongyi pointed out that for Chinese companies, there is actually no need to wait until technology has fully caught up with GPT-4 or Sora before exploring application implementation.

Chinese companies have been catching up well in the field of general large models in the past two years. 2024 should be the "year of application" of AI in China. In many vertical fields at the enterprise level, large models will have great potential.

  "The big model should produce an industrial revolution in the future. It must follow enterprises into hundreds of industries and be integrated with business processes and product functions." Zhou Hongyi made a comparison. GPT-4 "knows everything, but is not specialized in everything." "If we can accumulate unique business data in a certain business field, train a large model well in a vertical field, and combine it with enterprise business tools, such a large model will not only have a "brain", but also There is unique knowledge, even "hands" and "feet".

“It is entirely possible that China’s large models in some vertical fields will surpass GPT-4.”

  Zhou Hongyi believes that in the future, enterprises will not only have one super large model, but will have multiple small-scale large models with tens of billions of parameters. Each large model focuses on the work of a scene. Such large models will be closely related to the business of the enterprise. Platform integration.

According to this model, for many companies today, large models are "completely affordable and can be used very well."

  Talking about the current hot word "new quality productivity", Zhou Hongyi said that it should be productivity driven by technological innovation and productivity that can bring high-quality development.

At present, technological innovation still relies on industrial digitization to empower most traditional industries.

The essence of digitalization is ultimately intelligence. From this point of view, using large models to drive industries is precisely creating new productivity.

  Zhou Hongyi revealed that his proposal during this year's National Two Sessions recommended promoting the R&D and application of large models at multiple levels and turning large models into new productivity tools.

Specifically, the first is to adhere to the bottom line of safe development, and lay out policies and standards to ensure the safety of large models in various cities and industries; the second is to develop private large models, explore more application scenarios, and develop Professional, vertical, miniaturized, and low-cost enterprise-level large models will help digital transformation of hundreds of industries; third, we will create inclusive large models and provide special support to reduce the cost of private large models so that the majority of small, medium and micro enterprises can afford and use them. Use it to help complete the "chain" digital transformation.