Source: People's Daily Original Draft
In recent months, China's technology enterprise circle has been hot spots. On the front foot, Baidu, Ali, iFLYTEK and many other established Internet companies have invested in the big language model track with high profile; On the back foot, start-ups such as Lightyear Away, Shenyan Technology, and Baichuan Intelligent have also announced that they have received hundreds of millions of financial support...
This new industrial boom for "general artificial intelligence" has also triggered a new round of "grabbing people". The salary of high-end talents is rising, and the structural contradiction between the supply of talents and market demand in the field of artificial intelligence is becoming increasingly prominent. In the face of the "rapid progress" of the industry, how should the mainland make up for the shortcomings in the training of artificial intelligence talents?
Doctor's annual salary is million, high-end talents are "in demand"
"I can't finish all kinds of news and papers every day, and I almost run out of time to read." An industry insider engaged in AI research sighed. At present, many domestic Internet companies and start-ups are accelerating their layout in the field of large models, and his company is no exception.
In April this year, with the formal establishment of a new large model team in his company, he also "changed" from the original voice team to become a member of the large model team. The continuous expansion of the team is the biggest change in the company in the past few months, "the number of desks is almost out".
In this round of big model boom, the problem of talent scarcity bears the brunt. MIT Technology Review believes that among the AI ecosystems with talent, data, capital and hardware as the four factors, the importance of talent is the most prominent. "They are the main driver of innovation in algorithms and hardware, and in the long run, talent is more important than data."
As a rapidly developing emerging field, there has always been a gap in artificial intelligence talents in mainland China. The Liepin report shows that the shortage of artificial intelligence talents continues to be higher than the overall level of the Internet, especially after 2020, showing a rapid upward trend.
The start of this "big model" industry competition has pushed this contradiction between talent supply and demand to the cusp. Advertisements for heavy money are emerging in an endless stream, and it is not uncommon to poach each other, and "a team is poached at every turn" makes many companies feel helpless, and some projects even put up the slogan of "no upper limit on internship salary".
Industry insiders revealed that there are very few people in China and the United States who have really had experience in the research and development of large models, "adding up to 100 people", and these people are currently facing the pursuit of science and technology giants and start-up companies, and "the salary is very high".
Ren Fuji, academician of the European Academy of Sciences, honorary vice president of the Chinese Institute of Engineering Intelligence and chair professor of the University of Electronic Science and Technology of China, told People's Daily News: "On the whole, the talent demand of the artificial intelligence industry will not undergo subversive changes. However, there will be a talent gap in areas such as pre-trained models, conversational robots, and AIGC, which are closely related to GPT. ”
Zhang Dafang, doctoral supervisor of the School of Information Science and Engineering of Hunan University, said, "The current AI model is still essentially a kind of software, which follows the principle of software life cycle, and the demand for early development talents is large, and the demand for application talents is small. However, after the mature software is recognized by the market, the demand for application talents in the middle and late stages will increase significantly, while the demand for development talents will gradually decline. In the future, application talents who can 'use big models' will be of great use, and their salaries will also rise. ”
Top-notch Leading Talent is Scarce The talent structure still needs to be perfected
As the battle for talent intensifies, the ChatGPT team has also become a "dream team" for all parties to compete for. According to the statistics published on the official website of OpenAI, a total of 87 people have contributed to the ChatGPT project. The average age of this team of less than 100 people is only 32 years old, most of them have prestigious university degrees, as well as work experience in world-renowned companies, and the number of undergraduates, masters and PhDs is relatively balanced, accounting for almost one-third each. Its notable common features are "very young", "luxurious background", "focus on technology", "deep accumulation" and "advocating entrepreneurship".
According to a report released by the research group of the China Institute of Science and Education Strategy of Zhejiang University in 2022, in the field of AI, the United States occupies a leading position in the number of scientific research talents or industrial talents, basic talents or top-notch talents, and China's outstanding talents are densely distributed in universities and scientific research institutions, and the talent gap in the industry is large.
Ren Fuji analyzed that China has a huge population and a huge education system, but it is relatively lagging behind in the training of high-end talents. In addition, Chinese enterprises and talents pay more attention to business and application, and there are still deficiencies in the underlying technology and original innovation.
Pulse data shows that the domestic artificial intelligence industry algorithm direction talent recruitment is the most difficult and the most popular, occupying 7 seats in the top ten positions that are the most difficult to recruit. Nearly ninety percent of algorithm engineers need a master's degree and a doctorate.
From the perspective of supply, undergraduate talents are the main body, accounting for 62.7%, followed by masters, accounting for 22.1%. The largest supply of talents is the double-first-class universities, whose professional fields are mainly distributed in computer science and technology, electronic information, big data engineering, information and communication engineering and other majors.
"The emergence of ChatGPT is not a big change in the direction of artificial intelligence research, but actually the result of the integration of several major technological developments in artificial intelligence." Guo Yike, academician of the Royal Academy of Engineering, academician of the European Academy of Sciences, and chief vice president of the Hong Kong University of Science and Technology, told People's Daily News that this means that the requirements for artificial intelligence talents in the future are more comprehensive, and compound talents with profound technical attainments and profound mathematical and humanistic foundations are needed.
School-enterprise cooperation, focus on cross-cutting to create a highland for AI talent training
A number of experts and scholars pointed out that the current AI talent training in mainland universities faces two prominent problems: one is the scarcity of high-end talents, and the other is the disconnect between talent training and industrial application.
To this end, universities are constantly improving the construction of artificial intelligence disciplines. It is understood that since Beihang University established the country's first artificial intelligence major in 2017 and the University of Chinese Academy of Sciences newly established the School of Artificial Intelligence Technology, the pace of artificial intelligence professional training has gradually accelerated. At present, more than 400 colleges and universities have opened artificial intelligence undergraduate majors, and more and more universities have listed the construction of related disciplines as an important task.
New colleges, new majors, and discipline construction are still in their infancy, and the AI talent training model is also constantly being explored.
Tsinghua University's "Yao Class" all-English course, "Zhi Class" AI+X crossover project; the mathematical advantages of Peking University's "Turing" class; The "combination of departments" model of the University of Science and Technology of China... Relying on their existing advantages, colleges and universities have formed different characteristics in the cultivation of artificial intelligence talents.
In terms of curriculum, the higher school gradually condenses a teaching system that adapts to the development of disciplines. Ji Mengqi, associate professor of the Institute of Artificial Intelligence of Beihang University, introduced that Beihang has set up an undergraduate major in artificial intelligence, and the teaching system emphasizes the combination of "intelligent theory + common technology + major system platform", and pays more attention to the foundation of mathematics and the intersection of disciplines compared with other majors. He also often encourages students to participate in various competitions to develop their practical problem-solving skills.
Zeng Yi, a researcher at the Institute of Automation of the Chinese Academy of Sciences, said that the artificial intelligence science system is not only related to natural sciences, but also closely related to humanities and social sciences. In the future, the cultivation of artificial intelligence talents also needs to strengthen courses and practices in ethics, law and other aspects.
School-enterprise cooperation to improve the quality of AI talents has gradually become the consensus of universities and enterprises. In recent years, the two sides have jointly built school-enterprise joint R&D centers and school-enterprise joint laboratories, so that the teaching content can closely follow the front line of the industry, and can also cultivate talents with practical experience for enterprises in a more targeted manner. Leading enterprises have successively carried out in-depth cooperation with many universities, such as Baidu, which has jointly carried out talent training with more than 100 universities across the country.
In Guo Yike's view, the "diversity" of talents is very important. He said there was no need to impose a standard for talent, but rather to give a wide variety of talent the freedom to use their own talents. "If everyone learns one model, it's hard to develop a good team."