laitimes

Interview with Wu Hequan: After the mobile phone is embedded in the large model, many applications can be used without the cloud, which will bring trillions of yuan in scale of replacement

author:National Business Daily

Every reporter: Zhang Rui Every editor: Chen Xing

The 2024 Zhongguancun Forum Annual Conference will be held from April 25 to 29. With the theme of "Innovation: Building a Better World", hundreds of top experts and thousands of speakers will focus on topics such as scientific and technological innovation leading social progress, improving people's livelihood and sustainable development, discussing open innovation, seeking common development opportunities, and contributing wisdom and strength to building a better world.

The Zhongguancun Forum is a national-level platform for China's global scientific and technological innovation exchanges and cooperation, and has been successfully held for 14 sessions. Nearly 120 activities will be held, and more than 150 foreign government departments and international organizations from more than 100 countries and regions will be invited to participate in the forum.

During the Global Science and Technology Innovation High-end Think Tank Forum held on the 28th, Wu Hequan, academician of the Chinese Academy of Engineering and former vice president of the Chinese Academy of Engineering, was interviewed by a reporter from the National Business Daily (hereinafter referred to as NBD).

Interview with Wu Hequan: After the mobile phone is embedded in the large model, many applications can be used without the cloud, which will bring trillions of yuan in scale of replacement

Wu Hequan made a speech Image source: Photo by reporter Zhang Rui

The embedded large model of the mobile phone will bring a new round of replacement craze

NBD: You mentioned earlier that in the future, we will also embed large models in mobile phones, and what subversive functions will appear after mobile phones are embedded in large models?

Wu Hequan: After the mobile phone is embedded in the large model, you can use many applications without going to the cloud. Why don't you go to the cloud? Because many times mobile phones involve privacy, and you can better protect your privacy by not going to the cloud. In addition, not going to the cloud also reduces the latency, and many processes can be operated on the mobile phone, such as Wensheng pictures, Wensheng songs, and Wensheng videos, which can be done without using the cloud. In other words, users can use their mobile phones to perform corresponding operations when there is no cloud or network. Of course, you can also go to the cloud for collaboration.

In addition, the mobile phone is embedded with a large model, and when there is a network, users can achieve good communication. For example, when communicating with foreigners, especially foreigners in small languages, the so-called simultaneous interpretation on the Internet is actually not simultaneous, and you need him to speak a sentence, and you can speak it after he finishes translating. In the future, if mobile phones are embedded in large models, simultaneous interpretation can be realized.

Another example is to communicate with a deaf and mute person, he can only use sign language, which ordinary people do not understand, if the mobile phone is embedded in a large model, you use the mobile phone to point at him, his sign language can be translated into a language you can understand.

Another example is that China is gradually entering an aging society, and healthcare is a high-demand market. As long as you point your phone at your face or tongue and feel your pulse, it can provide some health advice and more reliable health prescriptions. This is a bit like the look and feel of traditional Chinese medicine, and the big model in the mobile phone is the old Chinese medicine doctor, which is equivalent to having a family doctor.

Of course, if you have children at home, you can also use your mobile phone to tutor them. Nowadays, many students can't be tutored by their parents when they reach junior high school and high school, and at this time, mobile phones can help you tutor their children.

When 4G came out, it achieved short videos on mobile phones, and now 5G is also the same short video, without a breakthrough, mobile phones can neither do 3D nor AR. In the future, after the large model is landed, it can do 3D and AR, which can make short videos richer, and users can also generate them themselves. So I think it will bring a new round of mobile phone replacement craze.

NBD: As you mentioned, will this generational boom be on a trillion-level scale?

Wu Hequan: Yes, China now calculates 1.4 billion mobile phones, generally speaking, once every three years, and 400 million to 500 million mobile phones a year. Assuming that a mobile phone is calculated at 3,000 yuan, the market size is at least 1.2 trillion yuan per year.

Interview with Wu Hequan: After the mobile phone is embedded in the large model, many applications can be used without the cloud, which will bring trillions of yuan in scale of replacement

To obtain high-quality data, it is necessary to build a knowledge base, and at the same time, it is hoped that it can be open source

NBD: We understand that many enterprises are particularly difficult to obtain high-quality data in the process of training large models, would you like to talk about what kind of high-quality data is, and why do we lack high-quality data?

Wu Hequan: There are a lot of netizens in China, and there is a lot of data on the Internet, but the data on the Internet cannot be called high-quality data. There are high-quality data on the Internet, but more of them are user-generated data, which is uneven, plausible, and even misleading. So if this data is used for training, it is definitely not good data.

So there is a lot of data to really be trained, in the West is used encyclopedias and the original old data of the library, etc., of course, there are many libraries in China, but there are not many digitized books, and there are many open source databases abroad, although China also has databases, but there are not many open sources.

Therefore, in order to solve the problem of difficulty in obtaining high-quality data, it is necessary to have a knowledge base and industry knowledge base in the future, and at the same time, it is hoped that it can be open source, so that it can be used for large model training.

Interview with Wu Hequan: After the mobile phone is embedded in the large model, many applications can be used without the cloud, which will bring trillions of yuan in scale of replacement

Wu Hequan Image source: Courtesy of the organizer

NBD: Are there any other solutions?

Wu Hequan: Of course, the government is the department that has the most high-quality data, but the government's data must be processed before it can be opened, and there can be no state secrets, corporate secrets, personal privacy, etc., and there is still a lot of work to be done in this regard. In addition, even if these efforts are done, it remains to be seen whether the government will be able to actively open up the data.

NBD: What do you think is the next stage of technology for large models?

Wu Hequan: Now the main thing is the basic large model, the next stage is to the industry large model and special large model, and may even hope to develop a small model, because now the large model consumes too much computing power, so the industry is also thinking about whether it can become a small model, hoping to use a small model to solve the problem of a large model.

NBD: Now that some local governments have issued hash power coupons, do you think it is effective in solving the computing power gap under the boom of large models?

Wu Hequan: The main purpose of computing power coupons is not to solve the computing power gap, but to facilitate the use of computing power by small and medium-sized enterprises, because no matter who uses it, the computing power is so much, and now the key is that small and medium-sized enterprises have a cost problem.

National Business Daily

Read on