In the Chinese AI industry, Tsinghua University is an unavoidable name.
Yang Zhilin, founder of Moon’s Dark Side, Tang Jie, founder of Zhipu, Wang Xiaochuan, founder of Baichuan Intelligence, Liu Zhiyuan, co-founder of Mianbi Intelligence… They are either professors at Tsinghua University or have studied in school before.
In the field of embodied intelligence, another track of AI, there are similar images. Chen Jianyu, founder of Xingdong Era, Zhao Mingguo, co-founder of Accelerated Evolution, Gao Jiyang, founder of Xinghaitu, Wang He, founder of Galaxy General Motors… are all professors and graduates from Tsinghua University.
What is more intriguing is the relationship between these people.
Yang Zhilin’s mentor was Tang Jie, Liu Zhiyuan’s mentor was Professor Sun Maosong from Tsinghua University, and Sun Maosong’s computer science department at Tsinghua University was founded by Zhang Bo, the founder of China’s artificial intelligence field.
This is not an alumni record, but a family tree, four generations, a chain, the story of Chinese artificial intelligence, which is the story of this chain.
01
In most people’s impression, China’s AI has only become popular in recent years, as if nothing had happened before.
But the fact is not so.
In 1978, Professor Zhang Bo from the Department of Automatic Control at Tsinghua University transferred to the Department of Computer Science and set himself a research direction: artificial intelligence and intelligent control.
He immediately opened a new course called “Introduction to Artificial Intelligence”, and even the textbook was compiled and printed page by page by combining foreign materials.
During his overseas visits, Zhang Bo gradually realized that the development of artificial intelligence requires the use of mathematical tools to improve algorithm efficiency. So he collaborated with Professor Zhang Ling from the Mathematics Department of Anhui University, and the two communicated across the ocean, using the thinnest paper and writing the smallest characters, just to save on shipping costs.
In 1984, Zhang Bo and Zhang Ling collaborated to publish a paper in a top international journal in the field of artificial intelligence, which was the first academic paper by Chinese scholars in the field of artificial intelligence. Subsequently, he became the first Chinese scientist to publish a paper at an international joint conference on artificial intelligence.
In the following years, AI experienced more than one low point. In the 1980s, the expert system foam burst, the global AI funding fell sharply, and research fell into a cold winter; From the 1990s to the early 2000s, neural networks were questioned and AI was repeatedly criticized. Every cold winter, some people leave, some turn, and some feel that this path is not feasible.
But Zhang Bo has always stayed at Tsinghua University, leading students and conducting research, from a professor to an academician, guarding AI without moving a step.
In 2015, Zhang Bo, who was nearly eighty years old, proposed the theoretical framework of “third-generation artificial intelligence” – advocating the combination of data-driven deep learning and knowledge driven symbolic reasoning to solve the fundamental problems of AI systems being inexplicable, insecure, and unreliable.
This framework has deeply influenced many large models today and is a core principle in AI theory.
02
Zhang Bo planted the roots of Chinese AI, but for the roots to grow into a big tree, someone needs to continue watering them.
Professor Sun Maosong from the Department of Computer Science at Tsinghua University is the one who watered the water.
As a senior professor at Tsinghua University, Sun Maosong has been working in the field of computer science for 46 years. If Zhang Bo is the founder of Tsinghua AI, then Sun Maosong is the generation that carries on the legacy. The Chinese word segmentation system CSegTag, which he led the research on, is the infrastructure of natural language processing and cannot be avoided by anyone who later pursued Chinese NLP.
But Sun Maosong’s contribution, as well as the proud students he brought out.
For example, Tang Jie, who started pursuing his PhD at Tsinghua University in 2002, declined the opportunity from a large company after graduation and stayed on as a teacher to focus on research. He developed an academic search engine called AMiner, which uses AI to mine relationships in academic networks, such as who has collaborated with whom, who has influenced whom, and which direction is heating up.
This project may seem like a small tool, but it possesses the key capabilities of a large model – data mining and knowledge graph.
In 2019, Tang Jie founded Zhipu. The goal is clear: to create China’s own big model and benchmark against OpenAI.
For example, Liu Zhiyuan, who enrolled in undergraduate studies in 2002, had no decent research experience, no competition experience, and did not achieve outstanding results. However, Sun Maosong still recognized his research enthusiasm and recruited Liu Zhiyuan to pursue a master’s degree under his leadership.
Liu Zhiyuan took a different path from Tang Jie – Tang Jie made cloud based big models, combining parameters and computing power; Liu Zhiyuan’s Face Wall Intelligence is an end-to-end large model that combines efficiency and practicality.
This tree can continue to branch and leaf.
During Tang Jie’s tenure at Tsinghua University, he brought out a student named Yang Zhilin, who taught him the basics of scientific research from the perspectives of how to find important problems, how to find solutions, how to conduct experiments, how to write articles, and how to make presentations. Tang Jie taught Yang Zhilin step by step.
Later, Yang Zhilin went to Carnegie Mellon University to pursue a PhD and studied under AI experts from Apple and Google. Later on, he returned to China and founded the Dark Side of the Moon, creating Kimi.
From Zhang Bo to Sun Maosong, to Tang Jie, Liu Zhiyuan, and then to Yang Zhilin. From 1978 to 2026, four generations, 48 years. Every generation is at the forefront of AI, and they have taken a step forward in their direction.
03
Tsinghua AI can grow into towering trees, in addition to passing on the torch, there are three magic weapons.
Firstly, the professor personally stepped down.
Tang Jie’s creation of Zhipu is not about “professors leaving to start their own businesses”, but about incubating companies directly as a Tsinghua professor with academic research results. The foundation of the GLM model is his research accumulation at Tsinghua University; The team of Zhipu is the original members of the AMiner project.
Liu Zhiyuan did the same for the face wall. He is still an associate professor in the Department of Computer Science at Tsinghua University, and co founded Face Wall Intelligence with both academic and entrepreneurial identities.
And in the field of embodied intelligence, Zhao Mingguo, the founder of accelerated evolution, is a professor in the Department of Automation who began researching bipedal robots as early as 2000.
Chen Jianyu, who founded Xingdong Era, is still an assistant professor at the Cross Information Research Institute. The company is incubated in the institute and is the only embodied intelligence enterprise owned by Tsinghua University. The results of the laboratory can be directly transformed into products for the company.
This is not common in Chinese science and technology innovation.
In most fields, there is a ‘valley of death’ between academia and industry – professors conduct research and publish papers; Students graduate and go to the company; The company takes the ideas from the paper and slowly develops the product. A few years or even a decade in between.
In the field of artificial intelligence, Tsinghua University has directly filled this valley. Professors do not need to resign, research results do not need to be transferred, and doctoral students can directly enter the project. The paper will be published today, and the model will be released tomorrow – there is almost no time difference between academic frontiers and commercial implementation.
This is also one of the reasons why Tsinghua University’s artificial intelligence companies started rapidly – Zhipu was established in 2019 and now has a market value of HKD 900 billion. In just three years, its valuation has reached USD 30 billion.
Not only due to the drive of capital, but also because the distance between technology from the laboratory to the product has been compressed to the extreme.
Secondly, it is the top talent pool.
In 2005, Turing Award winner Yao Qizhi returned from Princeton University and founded the Tsinghua School Computer Science Experimental Class, also known as the “Yao Class”.
The positioning of Yao Class is very direct: to cultivate top-notch talents with the same or even higher competitiveness as undergraduate students at MIT and Princeton. In 2019, Yao Qizhi founded the “Intelligent Class” – Artificial Intelligence Class, which specializes in cultivating leading talents in the field of AI. In 2022, the Yaoban, Zhiban, and Liangxin classes under the Cross Information Research Institute will merge, with three directions: computer science, artificial intelligence, and quantum information.
Yao Class has 20 years of experience, Zhi Class has 7 years of experience, with 750+undergraduate students and 237 doctoral students from the Cross Information Institute.
This number may not sound big, but its value is extremely high. The students of Yao class were top contestants in various provincial information competitions before enrollment, and after enrollment, they received world-class research-oriented training, becoming the most elite seeds of AI in China.
Thirdly, there is a tight conversion mechanism.
In December 2020, the Intelligent Industry Research Institute of Tsinghua University was established, with Zhang Yaqin, a former Microsoft executive and foreign academician of the Chinese Academy of Engineering, serving as the dean.
Zhang Yaqin, who has experienced the forefront of the industry, came back not to teach, but to build bridges.
The positioning of the Intelligent Industry Research Institute is very clear: to empower industrial upgrading with artificial intelligence technology. It is not a traditional laboratory or incubator, but a “revolving door” in between – university professors can conduct industry level research here, companies can find cutting-edge academic technologies here, and doctoral students can work on both papers and products here.
Someone asked online, why can Tsinghua University excel in the field of artificial intelligence?
There are many people who answer, but there are a few reasons that cannot be avoided.
The first is academic. Since Zhang Bo established the direction of AI in 1978, the academic heritage of several generations has never ceased. Zhang Bo builds the theoretical framework, Sun Maosong builds the NLP infrastructure, and Tang Jie builds the knowledge graph – every generation is at the forefront and the direction has never been lost.
The second is technology. CSegTag、AMiner、GLM， The technological foundation accumulated over decades allows Tsinghua University to avoid starting from scratch when large models explode. When others are looking for direction, they are already in the right direction.
The third is industry. Professors can engage in both academic and business activities, laboratories can incubate companies, and Tsinghua University can hold shares. AIR has institutionalized this industry university research mechanism, compressing the distance between technology and products from the laboratory to the shortest.
These three reasons allow the AI bloodline of Tsinghua University to continue and flourish.
The field of AI has undergone too many reshuffs. The expert system foam burst, the neural network was called down, deep learning suddenly emerged, and the big model overturned everything – every paradigm shift, someone was eliminated, someone fell behind, someone started again.
If the industry is cut off, we need to find a new direction, accumulate again, and start from scratch. Many areas in China are like this – after ten years of pursuit, the direction changes and all previous efforts are in vain.
But such a regret did not happen with Tsinghua’s artificial intelligence, it always had someone present. When people are there, the battlefield is there; The battlefield is there, the accumulation is there.
That’s why Chinese AI can compete with OpenAI on the same stage – not only because computing power has caught up, nor just because capital has poured in, but because for more than 40 years, there have been people unwilling to step down from the table, and new players have always been on the table. The Tsinghua AI desk has not been available for 48 years.
相关文章
