From catching up with the United States, to being copied by the United States

An AI (artificial intelligence) team composed of three students from Stanford University in the United States plagiarized the farce of the "small steel cannon" MiniCPM-Llama3-V 2.5, an open-source achievement of the Chinese large-scale model company "Facing Wall Intelligence", triggering a comparison of the research and development speed of AI technology in China and the United States.

After the "Stanford plagiarism incident", the "Global Times" reporter met with the face-to-wall team at the 6th Beijing Zhiyuan Conference, whether it was the founder Liu Zhiyuan or the CEO Li Dahai, they tried their best to downplay the matter. The appearance of Zeng Guoyang, CTO (Chief Technology Officer) of the wall, has attracted a lot of attention.

This 26-year-old boy, known as the "AI genius", was asked by the host with a smile when he took the stage, "Have you graduated?" Known as the "Spring Festival Gala of the AI World", the Beijing Zhiyuan Conference is a pure technology venue, gathering many domestic and foreign talents in the field of AI, which makes the Global Times reporter very curious, what is the ecology of China's AI industry? Did the "plagiarism incident" shake up China's AI community? What is the state of AI research and development in China and the United States? The Global Times reporter interviewed Zeng Guoyang on this.

Zeng Guoyang was interviewed by the Global Times. Photo by Wang Dong

抄袭事件涟漪

Three weeks after the plagiarism incident, Zeng Guoyang restored the matter to the Global Times reporter from his own perspective. "When we first learned about this, it was feedback from some users of the open source website GitHub, and we found some relevant clues and evidence, which finally proved that there was indeed a not-so-good shell phenomenon [at Stanford University]." Zeng Guoyang said that the incident finally ended with the Stanford undergraduate team apologizing and deleting the relevant code and products in the form of emails.

In this regard, Manning, director of the Stanford AI Lab, issued a condemnation of plagiarism students, and called MiniCPM "a good open source work". Zeng Guoyang said that after the incident, he felt the support of many people, and let more people know that many Chinese large models in the field of open source are worth paying attention to.

Liu Zhiyuan, chief scientist of Facewall Intelligence and tenured associate professor of Tsinghua University, also expressed the same feeling in an interview with the Global Times, he believes that China's large models, including Facewall, have not attracted attention because of plagiarism, and China's R&D progress in the field of AI has been concerned before.

Does the large domestic model, which has always been considered to be "catching up with the United States", now be the object of plagiarism, does it mean that China has taken the lead? Zeng Guoyang believes that it cannot be said that the R&D level of China and the United States is in front and behind, but it is a state of "you chase after me". In the field of open source, China has always been a very important player in the field of large models, and there will continue to be cutting-edge breakthroughs in China and the United States. Although relevant research institutions in many countries are also involved in the research and development of large models, "in terms of volume, the United States and China are still the main heavyweight players."

Liu Zhiyuan said that in terms of academic research, 10 years ago, China was able to publish many good academic papers internationally. However, he pointed out to reporters that compared with foreign countries, the current domestic repeated development of basic large models is a kind of follow-up and waste, and domestic needs to build a platform system for high-quality manufacturing large models. He called it the construction of a "large-scale lithography machine".

"We also grasp the scientific principles of lithography machines, but the fundamental reason why China's high-end lithography machines are 'stuck' is that we have not done so well in terms of systematization and platformization." Liu Zhiyuan cited the American artificial intelligence start-up OpenAI as an example, saying that when ChatGPT was launched, there were hundreds of contributors listed, but by February this year, the Wensheng video model Sora came out, and there were only 13 core members. What happened in between? Liu Zhiyuan believes that OpenAI has automated and platformized the basic work completed by a large number of programmers and scientists to form its own "lithography machine with large models". Sora is developed on this basis, and naturally does not need so many people to participate.

Liu Zhiyuan said that it is necessary to be more soberly aware of the real gap between China and the level of advanced technology, "If we are just satisfied with replicating ChatGPT, it means that we will not be able to make Sora and ChatGPT4o in the future." We want to build our own 'large model lithography machine', which is the direction we want to work towards. ”

"The Three-Body Problem has influenced my career direction"

Zeng Guoyang was Liu Zhiyuan's student at Tsinghua University, and he taught himself programming at the age of 8 and was sent to Tsinghua University as the champion of the Olympiad. At the end of 2015, when Altman, Musk and others founded OpenAI, Zeng Guoyang, who was 17 years old at the time, came into contact with the concept of AI for the first time.

Zeng Guoyang's earliest contact with large models was GPT-3 released by OpenAI in May 2020, when he went to register an account for the first time to try, and he felt that "artificial intelligence will have a major breakthrough opportunity, and there must be a lot of imagination in the future", which also contributed to Liu Zhiyuan, Zeng Guoyang and others decided to make the first large model in China at the end of 2020.

"During 2019, we felt that in the field of NLP (Natural Language Processing, which is an important direction in the field of computer science and AI - editor's note), we basically did not have much gap with foreign countries. However, after the emergence of GPT-3, it has had a great impact on the NLP field. He said.

In 2021, Liu Zhiyuan took the lead in establishing Facewall Intelligence, and Zeng Guoyang, who graduated from Tsinghua University with a bachelor's degree in computer science, is one of the three core figures of entrepreneurship. Why is it called a face wall? Zeng Guoyang recalled that they first founded an artificial intelligence open source community called OpenBMB, which reads like "facing the wall", and at the same time, there are many people in the team who are fans of Chinese science fiction writer Liu Cixin's science fiction novel "The Three-Body Problem", and the "wall-facer" plays a very important role in "The Three-Body Problem", and finally the team chose to use the name "Facing the Wall Intelligence", which is full of science and technology.

Zeng Guoyang said that he likes "The Three-Body Problem" very much, "It can be said that Mr. Liu Cixin's "Three-Body Problem" has influenced the career direction of a group of Chinese people, including me."

From 10 to 100 people

In the United States, there are also a number of young people who are starting businesses in the field of AI. The youngest member of Sora's core team was born in 2003. The founding team of Pika labs, an American AI start-up, is only 4 people, one of whom is said to be Guo Wenjing, a post-95 "Hangzhou Harvard girl".

Zeng Guoyang said that the large model is a new technological change, giving many young people with a spirit of challenge and willing to try new technologies the opportunity to change themselves.

In the emerging field of AI in China, there are both gray-haired scholars and young people like Zeng Guoyang. At the 6th Beijing Zhiyuan Conference, a reporter from the Global Times saw that a middle-aged engineer respectfully called Zeng Guoyang "teacher" when he asked a question on the spot. Zeng Guoyang said that in China's AI circle, everyone calls each other "teacher", and everyone is willing to learn, accept new ideas, and take the initiative to learn some technologies that they don't understand.

In his spare time, Zeng likes to watch the news, go to the open source community to communicate, and sometimes even miss the days of writing code. At the beginning of last year, there were less than 10 people in the wall-facing team, and now it has expanded to more than 100 people, and Zeng Guoyang needs to be responsible for the research and development of the large model as a whole. At the age of 26, when many young people have just graduated from college and started working, Zeng Guoyang is already the CTO of Facewall Intelligence.

Regarding the outside world calling him an "AI genius" and "China's version of Sheldon", Zeng Guoyang admitted that he has been learning "okay" all along, and he has never watched the American drama "Young Sheldon". "Joining the computer field and doing artificial intelligence has always been my hobby that drives me, AI is a very, very creative work, and my work is like creating a new intelligent life." Zeng Guoyang said that he prefers to call himself an "AI artist" and that "I am creating an artistic creature."

Will AI replace humans in the future? Zeng Guoyang believes that AI will be used as an assistant to improve human work efficiency in the future, and cannot completely replace humans. The 26-year-old is a little yearning for AI to completely replace human work, "maybe humans may not need to work, they will do something more creative, or do something they particularly want to do."