If ChatGPT is a gust of wind, after blowing to China, in addition to attracting attention from all walks of life to AI large models, what else does it bring? At the sixth Digital China Construction Summit held recently, many officials and scholars are discussing this issue.

Needless to say, with the explosion of ChatGPT, artificial intelligence has once again stood on the cusp. Many domestic companies have seized the opportunity and launched their own AI models, such as Baidu's "Wenxin Yiyan", Ali's "Tongyi Qianwen", Huawei's "Pangu Big Model", iFLYTEK's "Spark Cognitive Big Model"... According to incomplete statistics, more than 20 large AI models have been launched in China so far.

Faced with so many AI models, Cao Shumin, deputy director of the Cyberspace Administration of China, is a little worried, "and can't rush up."

She mentioned the "chaos" of rushing to build data centers in some places. Wang Jian, academician of the Chinese Academy of Engineering and founder of Alibaba Cloud, once commented: Many places have built data centers into real estate-like projects, and the actual business of some data centers is "leasing", dividing the space of data centers to different customers... Don't build a data center for the sake of building it. Cao Shumin said, "All fields must not overheat."

At present, artificial intelligence is increasingly integrated into all fields of China's economic development, becoming an important driving force for promoting the leapfrog development of science and technology, industrial transformation and upgrading, and the overall leap of productivity, providing strong support for the conversion of new and old kinetic energy and high-quality economic development.

According to data released by the China Electronics Information Industry Development Research Institute, from 2017 to 2021, the scale of China's artificial intelligence industry increased by 2.6 times, accounting for 16.8% of the world. The share of patent applications in the world continued to expand, from 2012% in 13 to 2021.70% in 9.

Cao Shumin said that although a lot of progress and achievements have been made, there are still some gaps between the overall development level of artificial intelligence in China and developed countries, especially in some basic theories, original technologies and high-end core technologies.

She hopes that all walks of life can take advantage of the east wind brought by ChatGPT to truly sink their hearts into the development of artificial intelligence and seize opportunities, "while maintaining concentration and being able to develop rationally."

How AI big models will evolve

The development of artificial intelligence has gone through three waves, and the breakthrough of deep learning based on Internet big data has achieved the current grand situation of artificial intelligence. Zeng Dajun, deputy director of the Institute of Automation of the Chinese Academy of Sciences, believes that the ChatGPT system itself is not precisely a big breakthrough in science, but a big breakthrough in the integration and use effect after the technology accumulation to a certain extent, which is a milestone in engineering practice.

The consensus in the industry is that the super algorithm plus the super computing power and the coupling of super data have finally created an engineering miracle.

The Institute of Automation of the Chinese Academy of Sciences summarizes the capabilities and characteristics of ChatGPT as "four links": smooth and natural natural language generation, coverage of general knowledge system in all fields, universal and smooth human-computer interaction in multiple natural language scenarios, intent recognition and logical reasoning. Effectively, ChatGPT achieves a fairly high level of human-like comprehension and verbal dialogue capabilities.

Behind this is the support of AI large models, which is why many companies have launched AI large models after ChatGPT. Zeng said that the success of the big model behind ChatGPT has explored a different path for core AI challenges such as world cognition and modeling.

Overcoming key core technologies and basic theories such as big models, large computing power, and big data will become the key to grasping the commanding heights of artificial intelligence development. What people want to know is what the road of the big model will go in the future.

At the 6th Digital China Construction Summit Artificial Intelligence Sub-forum, Zeng Dajun introduced their views: the application and innovation ecology is undergoing drastic changes or at least the potential for drastic changes, large models promote the rapid development of decision-making intelligence, the need for miniaturization and regionalization of large models is very urgent, and more general artificial intelligence is expected to be realized.

He said that the big model is like the prototype of a human brain, by feeding various data, realizing various intelligent capabilities, the big model is redefining the interaction between people and computers, and is expected to become the main interface between computers and people in the future.

At the scene, Zeng Dajun emphasized the development of miniaturization and regionalization of large models. He said that the computing power and energy consumption challenges of existing large models will prompt a lot of work to develop in the direction of domain-specific, lightweight small models or large and small model mixes, especially in the fields of finance, education, medical care, transportation, etc., and a lot of work is trying to reduce the cost of large models.

Ji Rongrong, Nanqiang Distinguished Professor of Xiamen University and head of the Institute of Artificial Intelligence, also talked about this. He said that large models will eventually be used in vehicle-mounted, airborne, electric and other applications, in drones and other equipment, the hardware is larger, the cost will rise in multiples, a little more computing space will increase the huge hardware cost, so "each company should not only focus on the model to be bigger, but also think about how to make the big model small."

What path our country takes

According to the "National New Generation Artificial Intelligence Development Plan" issued by the State Council in 2017, by 2025, the basic theory of artificial intelligence will achieve major breakthroughs, some technologies and applications will reach the world's leading level, artificial intelligence will become the main driving force for China's industrial upgrading and economic transformation, and the construction of intelligent society will make positive progress. Five years later, in 5, China's artificial intelligence theory, technology and application will generally reach the world's leading level, becoming the world's major artificial intelligence innovation center.

Mei Jianping, deputy director of the High-tech Department and first-level inspector of the Ministry of Science and Technology, said that according to the plan, in recent years, the Ministry of Science and Technology and relevant departments have been vigorously promoting innovation in the basic theory, key technologies and demonstration applications of artificial intelligence.

For example, in terms of basic technology, especially related to infrastructure construction, the Ministry of Science and Technology has made a systematic layout of research and development from digital, data to computing and connection, and has achieved a series of results. Recently, the Ministry of Science and Technology is also actively promoting the construction of computing power network, hoping to build a computing power base for supercomputing and intelligent computing through the research and development and application of new technologies, and provide strong support for the construction of digital China.

Zeng Dajun believes that at present, China's artificial intelligence innovation level has entered the world's first echelon, and the overall development trend is good, but there are also many challenges. The main reason is that there are relatively few breakthroughs in the sky, and the gap between original and subversive core algorithms is relatively large compared with developed countries such as the United States; The ability to stand on the site is weak, the major and systematic integrated innovation ability is insufficient, and there is a problem of low-level duplicate construction.

For AI large models, he suggested that on the one hand, China needs to accelerate the application of large model technology, "China has a good reserve and application of large model technology, and the threshold of ChatGPT-like technology has been greatly reduced recently"; On the other hand, it is suggested to concentrate superior forces to break through the general base large model, "it is necessary to build a new type of organized scientific research force around the general large model, unswervingly continue to tackle key problems, and actively seek resource input from multiple channels".

He also said that GPT-4 has shown some rudiments of general artificial intelligence capabilities, but this technology path has very high power consumption, which is unaffordable for the development of the digital economy. Zeng Dajun believes that China has a good research foundation in the intersection of artificial intelligence and brain science, and it is necessary to further integrate large models and brain science to develop ethical and sustainable low-carbon AI.

No matter which path you take, the focus is on "conquering key core technologies and basic theories". Cao Shumin said that only breakthroughs in key core technologies of the entire chain can truly achieve the comprehensive development of artificial intelligence, which involves a large number of algorithm problems, model problems, intelligent computing power related industries, etc., "which need to be deeply cultivated."

Especially in the chip industry, which is directly related to the development and breakthrough of artificial intelligence computing power.

Make technology work for good

On the same occasion, Li Min, chairman of Rockchip, sighed that it is necessary to consider artificial intelligence training and reasoning methods that are in line with the development of China's realistic chip industry.

This year, Lipin participated in 7 large-scale exhibitions held in many countries, and almost all customers believe that in the next 3 years, all the IT industry will be swept by artificial intelligence and AIoT (artificial intelligence Internet of Things), "because in 3 years, everyone can not imagine what ChatGPT 7.0 will look like".

The trend has emerged, standing at the inflection point of artificial intelligence, our response is that we must keep up with the demand for computing power, and in the final analysis, we must rely on the underlying technology.

Li Min said that to think about the new industrial artificial intelligence model, can not copy ChatGPT, the domestic artificial intelligence model within 3 years should be adapted to the domestic chip development within 3 years. The development of related industries should also be "horse racing instead of Soma, and cannot rely on PPT to do the industry".

Cao Shumin also mentioned the issue of development security. Previously, OpenAI's data breach caused privacy controversy, and later, thousands of technology leaders, including Musk and Hinton, signed an open letter calling on all AI labs to immediately suspend training AI systems more powerful than GPT-4 for at least 6 months, all concerned about the various risks posed by the development of artificial intelligence.

"How to coordinate development and security is also a major issue facing us." Cao Shumin said that every new technology is a double-edged sword, bringing huge development opportunities, but also bringing many challenges, especially when the new technical capabilities are very strong, the risks brought are also very serious.

On April 4, the Cyberspace Administration of China issued the Measures for the Administration of Generative Artificial Intelligence Services (Draft for Comments), advocating the establishment of a regulatory system with a large scale of computing power and professional verticals, and the conclusion of code conventions to determine ethical norms for artificial intelligence.

Cao Shumin said that we must actively respond to the challenges brought by artificial intelligence in terms of possible public safety, privacy infringement, digital divide, ethical demonstration, etc., how to make technology good and benefit mankind, how to promote the development of artificial intelligence more fair, more sustainable and safer, so that technology can better benefit the people, and let technology better promote the healthy and sustainable development of the economy and society, "This is a major responsibility faced by all our subjects."

Zhang Junbin, reporter of China Youth Daily Source: China Youth Daily