laitimes

The report shows that 79 large models of Chinese intelligence have been released

author:New Hunan

At the sub-forum of artificial intelligence large model development of the Zhongguancun Forum held on May 28, the "Research Report on Chinese Intelligence Large Model Map" compiled by the China Institute of Science and Technology Information and the New Generation Artificial Intelligence Development Research Center of the Ministry of Science and Technology and relevant research institutions was officially released.

The report shows that 79 large models of Chinese intelligence have been released

According to Zhao Zhiyun, Secretary of the Party Committee and 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, the artificial intelligence big model represented by ChatGPT has led a new round of global artificial intelligence technology development, and new research and new products related to large models have emerged. China's early deployment in the field of artificial intelligence has laid a solid foundation for the development of large models, and under the joint promotion of government, industry, academia and research, systematic research and development capabilities covering theoretical methods and software and hardware technologies have been established, and large model research and development has shown a vigorous development trend. With the help of the visualization of large model maps, the report analyzes the phased development characteristics of China's large models, depicts the latest overall picture of the development of China's large models, and reveals the problems and shortcomings in the development of large models, hoping to promote the future development of large models in China.

The report shows that 79 large models of Chinese intelligence have been released

The large model technology group is growing rapidly

The report sorts out the development context of global large model technology and finds that since Google released the Transformer network structure in 2017, the world has rapidly grown a huge large model technology group in just over five years, deriving a large model family covering various technical architectures, various modalities, and various scenarios.

The report shows that 79 large models of Chinese intelligence have been released

China and the United States lead the development of global big models

The analysis of the report found that Google, OpenAI and other institutions in the United States continue to lead the frontier of large model technology. More and more R&D teams in Europe, Russia, Israel, South Korea and other places are also investing in the development of large models. From the perspective of the distribution of large models released in the world, China and the United States are significantly ahead, accounting for more than 80% of the global total, and the United States has always ranked the highest in the world in terms of the number of large models.

The report shows that 79 large models of Chinese intelligence have been released

China's large models are showing a vigorous development trend

China has entered a period of rapid development of large models since 2020 and is currently maintaining a synchronous growth trend with the United States. In the natural language processing, machine vision and multimodal and other technical branches are synchronous follow-up and rapid development, a number of pre-trained large models with industry influence, such as Pangu, Wudao, Wen Xin Yiyan, Tongyi Qianwen, and Xinghuo Cognition, have emerged, forming a large model technology group that closely follows the world's frontier.

The report shows that 79 large models of Chinese intelligence have been released

Map of the distribution of large models in China

The report analyzes 79 large models released in China based on public information, and the analysis results show that at present, there are teams in 14 provinces, cities/regions in China to carry out large model research and development, with Beijing and Guangdong being the most, and the geographical concentration is relatively high. From the perspective of field distribution, natural language processing is still the most active key area for large model development, followed by multimodal field, and there are still few large models in the fields of computer vision and intelligent speech. Different innovative subjects such as domestic universities, scientific research institutions, and enterprises are actively participating in the research and development of large models, and there are relatively few joint development between academia and industry.

The report shows that 79 large models of Chinese intelligence have been released

Map of Chinese large model hashrate elements

By investigating the distribution of computing power infrastructure nationwide, the report found that 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 artificial intelligence server purchases in the past three years, showing a very obvious strong correlation and providing important support for the development and application of large models. At the same time, various places are also providing public computing power to supplement the rapidly growing demand for artificial intelligence computing power, providing more computing power support for the research and development of large models.

The report shows that 79 large models of Chinese intelligence have been released

China large model talent element map

Large models have a high threshold and require high-quality AI talents. The results of the report show that in terms of quantity, there are still not enough talents in large models in various places. The report counts the authors of domestic artificial intelligence-related papers, and from the perspective of regional distribution, Beijing is far ahead of other regions in terms of the number of artificial intelligence scholars and the number of large model scholars, reflecting obvious talent reserve advantages, Jiangsu, Guangdong, Shanghai are also relatively large model talents.

The report shows that 79 large models of Chinese intelligence have been released

Map of China's Big Model Academic Influence

The report believes that China's big model has formed a certain academic influence through the publication of academic papers. Among them, Beijing, Guangdong and Shanghai rank the highest in China in terms of both the number of publications and citations.

The report shows that 79 large models of Chinese intelligence have been released

Top 10 academic influences of China's big models

In terms of model influence, the CogView model jointly developed by Tsinghua University with Alibaba and Baidu has the highest number of citations, and Huawei's FILIP, Baidu's ERNIE3.0 and Alibaba's M6-OFA citations also rank in the forefront in China, forming a good academic influence in the field of large models, but the gap is still large compared with the academic influence of foreign leading large models.

The report shows that 79 large models of Chinese intelligence have been released

China Big Model Open Source Impact Map

Open source is an important model of AI R&D collaboration and an important concept for the development of Chinese intelligence. The analysis of the report found that China's large model R&D team actively promotes the development of open source of large models, and more than half of the large models have been open source. Beijing, Guangdong and Shanghai all rank among the top three in China in terms of the number of open source and the influence of open source.

The report shows that 79 large models of Chinese intelligence have been released

China's top ten big model open source influence

The report shows that at present, the open source of large models is more promoted by universities and institutions, and ChatGLM-6B of Tsinghua University, MOSS of Fudan University and Baidu's Wenxin series of large models have formed the highest model open source influence.

The report shows that 79 large models of Chinese intelligence have been released

The generic model evolves in parallel with the dedicated model

According to the analysis of the report, at present, there are roughly two parallel development paths for the industrialization of China's large models, and a number of generalized large models such as Wen Xin Yiyan, Tongyi Qianwen, and Zidong Taichu are developing rapidly, creating a cross-industry generalized artificial intelligence capability platform, and its application industry is accelerating its penetration from office, life, entertainment to medical, industrial, education, etc.

At the same time, a number of professional large models for vertical fields such as biopharmaceuticals, remote sensing, and meteorology are giving full play to their in-depth advantages and continuously deepening the implementation to provide high-quality professional solutions for specific business scenarios.

The report shows that 79 large models of Chinese intelligence have been released

Zhao Zhiyun pointed out that large model technology has promoted artificial intelligence to achieve a historic leap, and there is still broad space for continuous innovation in the future. The high-quality development of China's economy and society provides a rich scenario and data foundation for big model innovation, and artificial intelligence has great potential for development in China. It is recommended to strengthen the coordination of resources and R&D forces to promote the orderly development of large models; Accelerate basic research and technological innovation, and enhance the impact of academia and open source; Strengthen the scene traction role in the development of large models and create benchmark projects for large models; Strengthen international cooperation and actively participate in global AI governance. (Economic Daily reporter She Huimin)

Read on