Zhongguancun Forum explores the development of China's large model: open source is encouraged in the early stage of development, and basic research still needs to be innovated and strengthened

21st Century Business Herald reporter Ji Yifei reported from Beijing

At present, the development of artificial intelligence is ushering in a new inflection point. The explosion of ChatGPT triggered the "Thousand Model War", and "how Chinese companies do not fall behind in the competition of large model research and development" have also become hot topics at the recent 2023 Zhongguancun Forum.

At present, China's large model technology is developing rapidly along various technical routes, and a number of pre-trained large models with industry influence, such as Pangu, Wudao, Wen Xin Yiyan, Tongyi Qianwen, and Xinghuo Cognition, have emerged in the fields of natural language processing, machine vision and multi-modality, and there will be more large models in the future to bring new efficiency revolutions and experience upgrades to all walks of life.

A number of industry experts said during the Zhongguancun Forum that it is currently in the historical opportunity period of developing AI large models, and the application of large models has gradually moved towards the real economy, in order to continuously seek breakthroughs in general artificial intelligence technology, make up for the shortcomings of the underlying algorithm, key software and hardware and ecosystem construction, the mainland artificial intelligence industry still needs to further strengthen industry-university-research cooperation, encourage open source and open up of large models, and strengthen independent innovation and research and development.

The large model is about to bring new changes in industrial applications

"We are now at a new starting point, a new era of artificial intelligence with big models as the core, big models have changed artificial intelligence, and they are about to change the world." At the plenary session of the 2023 Zhongguancun Forum, Baidu CEO Robin Li said that more and more people recognize that the fourth industrial revolution is coming, and the symbol of the revolution is artificial intelligence, because the big model successfully compresses human cognition of the whole world and allows people to see the path to achieve general artificial intelligence.

Kai-Fu Lee, chairman and CEO of Innovation Factory, also said, "AI big model is a historical opportunity that must not be missed, and it is the largest platform ever." By developing AI big models, China can promote the development of innovative industries, drive economic growth, and realize the dual dividends of technology and economy.

Kai-Fu Lee said that the big model has slowly moved towards the real economy, and the ensuing AI 2.0 era will promote the development of AI-first applications, starting from the field of productivity and entertainment, subverting many industries such as search engines, e-commerce, finance, education, film and television entertainment, medical care, games, etc., reconstructing human work, and many repetitive labor will be replaced.

The 21st Century Business Herald reporter learned at the meeting that the industrial application of large models is gradually adapting to the actual needs of all walks of life, basically developing along two paths.

The first is that the general class of large models continues to expand the application field, Wen Xin Yiyan, Tongyi Qianwen, Zidong Taichu, Xinghuo Cognition and other Chinese general large models are developing rapidly, creating a cross-industry generalized artificial intelligence capability platform, and its application industry is accelerating the penetration from office, life, entertainment to medical, industrial, education, etc.

A number of professional large models in vertical fields such as biomedicine and remote sensing meteorology give full play to their in-depth advantages to provide high-quality professional solutions for specific business scenarios.

Huang Tiejun, Secretary-General of the New Generation Artificial Intelligence Industry Technology Innovation Strategic Alliance and President of Beijing KLCII Artificial Intelligence Research Institute, pointed out in an interview with the 21st Century Business Herald that in the next three years, models will definitely spread from language to vision, hearing, embodiment, action, robotics, automatic driving and other directions. "The previously expected pure driverless and including robots entering the manufacturing industry, the home may occur in decades, and the integration of humans and AI to reduce the production cost of previous products and the development of new products and services will be a good day of harmonious symbiosis."

14 provinces have carried out large-scale model research and development

At this forum, the New Generation Artificial Intelligence Development Research Center of the Ministry of Science and Technology released the "Research Report on the Map of Chinese Artificial Intelligence Big Model" (hereinafter referred to as the "Report"). According to incomplete statistics, so far, 79 large models with more than 1 billion parameters in China have been released, and 14 provinces/cities/regions are carrying out large model research and development, mainly concentrated in Beijing and Guangdong, including 38 in Beijing and 20 in Guangdong.

"Beijing, Guangdong, Zhejiang, Shanghai and other places have the largest number of large models, and these four places are also the regions with the highest number of AI server purchases in the past three years. Various regions are also supplementing the rapidly growing demand for artificial intelligence computing power by providing public computing power to provide more computing power support for large model research and development. Zhao Zhiyun, director of the China Institute of Science and Technology Information and director of the New Generation Artificial Intelligence Development Research Center of the Ministry of Science and Technology, suggested that the coordination of computing resources such as intelligent computing centers, supercomputing centers, and cloud computing centers should be strengthened in the future, and public data sharing catalogs and sharing rules should be formulated to promote the orderly opening of data classification and classification.

It is worth noting that Beijing has launched the "Beijing General Artificial Intelligence Industry Innovation Partnership Program" in May to promote the development and application of artificial intelligence models and empower the digital and intelligent transformation of thousands of industries, and also issued the "Several Measures to Promote the Innovation and Development of General Artificial Intelligence in Beijing (2023-2025) (Draft for Comments)", which puts forward 21 measures for the development of Beijing's large model from the five directions of computing power, data, algorithms, applications and supervision. This may become the first special measure proposed by a local government in China to closely follow the industrialization of AI large models.

"The advantage of Beijing's development of AI lies in the fact that it has the most complete elements and can effectively integrate resources, but this is not a matter of one city in Beijing, but a huge industrial ecology of the whole country and even the world." Huang Tiejun said that all localities must pay attention to industrial collaboration in the process of supporting the artificial intelligence industry, and enterprises cannot use the original product thinking to plan large model research and development, innovation needs differentiation, and it is not appropriate to rush up.

Kai-Fu Lee said that the AI 2.0 market is expected to be 10 times the scale of the mobile Internet, which can accommodate the horse racing of giants, small and medium-sized enterprises and startups at three levels, and looks forward to forming an "innovation complex" where giants and small, medium and micro innovative enterprises develop together.

The "Report" shows that in the distribution of R&D subjects, different innovation subjects such as universities, scientific research institutions, and enterprises are participating in the research and development of large models, and joint research and development between academia and industry is still insufficient. In terms of model field distribution, natural language processing is still the most active key area of large model research and development, followed by multimodal field, and there are still few large models in computer vision and intelligent speech.

Zhao Zhiyun said that large model technology is still in the early stage of development, potential basic theory and technological innovation space is huge, through distillation, quantification and other miniaturization technologies to achieve model "slimming", for large model miniaturization, green development to provide technical support, further strengthen industry-university-research cooperation, encourage large model open source, will also accelerate the progress of large model technology.

Open source and independent innovation go hand in hand

Looking at the world, to build an artificial intelligence ecosystem, there are two routes: open source and closed source. Wang Huaimin, academician of the Chinese Academy of Sciences and director of the Open Source Development Committee of the Chinese Computer Society, introduced that the big model has formed a basic pattern of simultaneous promotion of open source mode and pre-research mode due to its uncertainty.

For the construction of artificial intelligence ecology, Wu Zhaohui, vice minister of science and technology, said at the 2023 Zhongguancun Forum that China adheres to open source collaboration, strengthens continuous innovation in large model technology, collaboratively solves common problems such as transparency and stability, further promotes the open sharing of computing power resources and digital resources, and accelerates the formation of an industrial ecology of large models.

Huang Tiejun said that open source is the future of big models, and collaborative innovation needs to be continuously strengthened. After the recent big model boom swept the world, the vitality of open source has been further stimulated, and open source is becoming the leading force shaping the future development of artificial intelligence. After more than 30 years of development, China's open source must have entered a new stage of development.

From a policy perspective, in 2017, China issued the New Generation Artificial Intelligence Development Plan, which wrote open source and openness as basic principles into the plan. Science and Technology Innovation 2030 - "New Generation Artificial Intelligence" major project also proposes to guide and encourage national projects to support the open sharing of R&D results, rely on artificial intelligence leading enterprises to build a new generation of national artificial intelligence open innovation platform, build an open ecology, and promote the development of the industry.

The "Report" shows that from the perspective of open source innovation ecology, more than half of the large models have been open source, and the number of open sources and the influence of open source in Beijing, Guangdong and Shanghai rank among the top three in China. Universities/research institutions are the main force of open source. Tsinghua University's ChatGLM-6B, Fudan University's MOSS and Baidu's Wenxin series of large models have the highest open source influence.

At present, Chinese technology companies such as Baidu, Huawei, Alibaba, and iFLYTEK are actively participating in the work of opening up the framework and model of artificial intelligence. Wang Haifeng, chief technology officer of Baidu, said during the forum that thanks to Baidu's comprehensive layout and technology accumulation in the field of artificial intelligence technology, especially the joint optimization of the deep learning platform of Flypaddle and the Wenxin large model, Wenxin Yiyan has improved the reasoning performance by nearly 10 times in more than a month, with higher efficiency and better results.

However, ChatGPT, which was previously popular in the world, after the release of ChatGPT2, OpenAI implemented closed-source development for it. At the meeting, many people in the industry also emphasized that open source and open up should still adhere to independent innovation.

"Support open source, but cannot rely too much on open source model", Lee Kai-fu pointed out in an interview with the media, the open source model can not achieve the performance of the self-developed model of foreign large manufacturers, and the ability of the open source model will become a "ceiling". There is also a risk of closure of the open source technology of overseas large manufacturers, and due to different cultures, user habits and laws and regulations at home and abroad, it is risky to bring foreign-trained models to China for fine-tuning.

"China's large model companies should no longer rely on the open source models of others, and need to establish their own IP and technical advantages as soon as possible to form a moat." Lee Kai-fu said.

Dai Qionghai, academician of the Chinese Academy of Engineering and chairman of the Chinese Engineering Intelligence Society, pointed out that China has become the world's largest artificial intelligence application market, and the overall level of artificial intelligence application technology is comparable to that of the United States, but it is also necessary to note that there is still a gap between artificial intelligence head enterprises and the United States, most of the talents are concentrated in the application layer, and the basic technology is obviously weak. In the future, we should deepen the talent training and basic research of artificial intelligence from the policy, mechanism and investment, strengthen original innovation, and avoid falling into the dilemma of "source water".

For more information, please download 21 Finance APP