2025 year, the artificial intelligence technology represented by DeepSeek has caused a sensation, causing a huge change in the pattern of domestic large models. How will artificial intelligence further develop in the future? What is the path for China's development of artificial intelligence?

At the "Academician and Expert Visit" event held during the eighth Digital China Construction Summit, Academician Wu Zhiqiang, Chen Jun, Liu Yunjie of the Chinese Academy of Engineering, and Academician Tong Qingxi of the Chinese Academy of Sciences expressed their opinions on the above topics.

Ecosystem thriving

There are still 18 teams similar to DeepSeek

Host: What insights does DeepSeek bring to innovation and development?

Wu Zhiqiang: We have been paying attention to the global development of artificial intelligence. Whether it is DeepSeek or ChatGPT overseas, they did not grow overnight. In fact, there are at least 18 teams in China that are as good as DeepSeek, it's just that DeepSeek stood out. Now, artificial intelligence has formed a flourishing ecosystem, and we must understand the entire ecosystem to truly lead or control the situation.

When we use artificial intelligence technology for urban planning, we are "advancing on two fronts". On one hand, we use general artificial intelligence large models, but they have two disadvantages: the raw materials "eaten" during model training are rough, and the computational power consumption is huge.

On the other hand, we adopt fine data to train the model, eating only "fine grains". We found that under this training mode, only 5%-10% of the computational power is needed to complete more accurate content output. This means that both the "precise" and "general" paths must be taken.

Chen Jun: When DeepSeek came out this year, people were excited about such a breakthrough in Chinese technology.

DeepSeek is indeed good, but if you ask it to write a poem, it can do it; however, if you ask it to do something professional, it falls short. Therefore, I believe:

Firstly, each field and profession has a lot of knowledge. First, gather and refine this knowledge to build a knowledge base. We need to use artificial intelligence technology to sort out, summarize, and apply the knowledge of our industry well.

Secondly, current large models are language models, focusing on contextual and inferential relationships. However, if we want to involve geographic information, urban planning, and other fields, which involve a lot of spatial information like front and back, up and down, indoor and outdoor, the required relationship structure becomes three-dimensional spatiotemporal relationships. Therefore, we need to develop "spatiotemporal large models", and large models need to evolve from language models into "spatiotemporal large models".

Finally, I believe that we should actively embrace artificial intelligence now, but we should not rely solely on it. We need to organically combine our professional intelligence with machine intelligence.

Liu Yunjie: I would like to share two insights: First, the "six dragons" are not only DeepSeek. It is unlikely that local governments will find talents or teams like the "six dragons". The most important thing is a good innovative environment with tolerance for innovation. The appearance of the "six dragons" in Hangzhou was not accidental, but inevitable.

Second, the success of DeepSeek shows that any innovative team must bravely challenge the most difficult problems in their field.

I judge the success of DeepSeek based on problem-oriented approach. Is it necessary to have high computing power to produce high intelligence? Or is it possible to achieve higher intelligence with less computing power? I think it solved this problem. This gave me inspiration: even in some fields where we are currently at a disadvantage, there are opportunities and possibilities for innovation and surpassing others.

New professions like AI nutritionists will emerge

Let young people stand out

Host: How should the construction of Digital China in the future proceed? What should we focus on, deploy, and promote?

Tong Qingxi: Every day in China, nearly four or five hundred remote sensing satellites are orbiting, meaning the amount of data transmitted down could reach exabytes (EB). For these data, if artificial intelligence large models cannot combine with multi-dimensional spatial information of remote sensing, and only focus on international or domestic general models, it may not solve practical problems. Therefore, we should develop our large models according to spatiotemporal multi-dimensional information.

Wu Zhiqiang: I think we should pay attention to two points. One is new professional trainers. In the process of training artificial intelligence, we have increasingly felt that how to train large models specifically needed by various industries, and what kind of "material" to give during training is very important. Just like in competitions, besides athletes, the coaching team is equally important, as it concerns the final outcome. This means that specialized AI trainers and AI nutritionists will become new professions in the future.

The second point is the emergence of youth. The development of artificial intelligence is not only about national strategy and overall layout, but sometimes it is bottom-up. For example, as mentioned earlier, there are many teams similar to DeepSeek, and many are just temporarily not standing out yet.

We should allow young people to stand out. Three years ago, one of my students gave me an "AI Wu Zhiqiang" as a gift — they "played" with it themselves — trained using all the books I have written and published. It is very excellent. This product was completely unplanned; it was just a group of students who thought it was fun to make.

In the future development of China's artificial intelligence field, this "emergence of youth" is a phenomenon we need to pay special attention to. We need to see them, encourage them, and give them nourishment. They might "play" out world-class topics.

Chen Jun: On one hand, we still need to create public products for artificial intelligence. For example, in daily life, we don't know whether the electricity we use is nuclear, thermal, or hydroelectric, as long as it's connected, it works. In the future, artificial intelligence should also be like this — it should not form monopolies or let everyone do it; instead, some general tools should be provided so that everyone can improve application scenarios on this platform.

On the other hand, we need to build a good ecosystem. If we develop artificial intelligence and end up eliminating our own professions or industries, it would be a sin, and we should not take that path.

My understanding is that after artificial intelligence comes out, it will assign many labor-intensive and dangerous tasks to machines, allowing humans to work as knowledge engineers, summarizing and refining knowledge.

Back when cars were invented, some people abroad were very worried, thinking that carriage drivers would be unemployed. Later, gas stations, 4S stores, repair shops, taxi industries, and other new professions emerged. I think we should strive to build a good ecosystem to better develop society. Turn people in every industry into knowledge engineers in that field, let them create and process knowledge, then feed the knowledge to machines, allowing machines to help us do the work that people don't want to do. This is a good ecosystem.

Original title: "Teams like DeepSeek, there are at least 18 in China"

Column editor: Qin Hong, text editor: Song Hui, picture source: Shangguan cover picture, picture editor: Cao Liuyuan

Source: Author: Shanghai Securities News

Original article: https://www.toutiao.com/article/7498870555366441512/

Disclaimer: The article represents the author's views. Please express your attitude by clicking the "like/dislike" buttons below.