Hello, everyone! Welcome to the People's Reception Hall. In recent years, China's artificial intelligence has accelerated its development and formed a relatively complete industrial system. Artificial intelligence is increasingly integrated into thousands of industries, becoming an important driving force for industrial transformation and upgrading and promoting the construction of new industrialization. This year's government work report also emphasized that we should vigorously develop the digital economy, which means that the digital economy such as artificial intelligence will usher in new development prospects.

Today, our studio invited Huang Tiejun, president of Beijing Zhiyuan Artificial Intelligence Research Institute and professor of Peking University School of Computer Science, Sun Maosong, professor of Tsinghua University Department of Computer Science and Technology, executive vice president of Tsinghua University Institute of Artificial Intelligence, and Zhou Jingren, chief technology officer of Alibaba Cloud Intelligence. Welcome everyone.

Today we will discuss the development of China's AI industry driven by the big model. We know that a series of phenomenal AI products generated by natural language have appeared recently at home and abroad. We want to ask three people first, what is the root cause of such a product?

Maybe you are concerned about the product. In fact, the overall AI trend is unchanged no matter which product appears at this time, and the product is just a typical thing that you are exposed to. In fact, the development of artificial intelligence, of course, can be said to have gone through almost 60 or 70 years, and can also be said to have reached such a moment after more than 10 years of development since 2012. What kind of moment? It is today's big model. What is a big model? In fact, it is an intelligent carrier. Intelligence comes from big data, massive data. Human intelligence, like biology, comes from a complex environment like the earth, which has evolved our intelligence. Today's AI, which is generated by computer algorithms, is actually based on massive data. Data is also an expression of the environment. Since almost the new century, the generation of massive data has provided the material conditions for generating intelligence. This is the first one.

Second, today's computer is particularly powerful. It can extract an intelligent model from massive data, which is also a basic condition. This condition has not been met before, and today's computing power has also met this condition. Together with the progress of artificial intelligence algorithms, it is possible to train models with higher intelligence levels today. With such a model behind it, it is not surprising that there are various products.

Just now, Mr. Huang said that there are actually three "big", three carriages, big data, big models and big computing power. These three things do not come together, but actually produce chemical reactions. But it's definitely not possible without these three things. It should be said that the results of these three "big" opportunities are the same. What is particularly important here is the role of natural language. Natural language is human language. The importance of human language is self-evident. We all know that language is unique to human beings and can reflect very profound thoughts. Many of our unique human activities, such as reasoning, decision-making, etc., actually depend on language. But before, we didn't find a good way to calculate and put it into the model. Over the years, in fact, we have found a way to express human language through calculation without omission and deal with it once again. What was the result of this? Any concept that we human beings have, this concept is very large, including the concept of rough force and the concept of fine force. Any concept can be in the form of a hidden vector representation that we call vector representation. Vector representation is actually a number, which can be connected in a way that can be calculated, Interaction can lead to more complex behaviors, such as big model and big data fusion, which may lead to emergence, which will lead to some sudden changes in its performance. Therefore, the recent performance of big models, including AIGC, is amazing. Our language processing is a very important reason.

In addition, this set of ideas is that we have further expanded multimodality. For example, when it comes to images, we take natural language as a bridge, and then combine the advantages of the image itself, combining the respective strengths of the text and image, to produce a good image generation effect now. This is also the bridge of language, which is a big step forward. I think these points may be an internal reason why our products are doing well.

Just now, I agree with Professor Huang and Professor Sun. The main reasons are from three elements, including data, computing power, especially the birth of cloud computing, and a series of progress in its own computing model. As Professor Huang mentioned just now, the whole data now exists in a very rich form, not only in exponential growth, but also in the past, we may have more text data. Now we have pictures and various multimedia data. The whole data volume is also developing in an exponential scale. With data, you need to have the ability to calculate to process these data. Just in the past few years, our entire cloud computing has made great progress. Not only has the computing power of a single server developed rapidly, but also our cloud computing has effectively connected thousands of machines to form a supercomputer, which also provides a solid foundation for the development of today's models.

In addition, we have made great breakthroughs in the model. First of all, the model itself is becoming more and more complex, especially the architecture of the Transformer model in recent years, which enables the model to have very large parameters and more parameters. To a certain extent, the model can capture a lot of knowledge details. In addition, a series of changes have also taken place in the model training paradigm. In the past, our AI models started more from labeled data, that is, people need to label in advance to tell the machine what results to learn. But today, because of the birth of a new AI paradigm, we are more through a so-called self-monitoring learning method, We can automatically label data from our massive data. Only in this way can we develop the training of models on a large scale and create such a large model.

Back to the essence of models, why do today's models have such a powerful series of applications? We are also constantly exploring. There may be several reasons. We know that today's big models have to some extent represented agents that are similar to human beings. One of the most important aspects of human intelligence is memory, that is to say, memory is the first thing to have other deductive abilities. Today, in the big model, because it can deal with all kinds of information seen on the Internet today, or in books, it can first read all the information, and the model itself has a strong memory ability, it can effectively combine our information, our knowledge, and organize through the powerful internal network, That is the epitome of efficiency we often talk about, that is, it has efficient storage to effectively store these knowledge. With this knowledge, when we solve practical problems, we will find that we can draw inferences from one instance. We often say that we can recite even if we are familiar with three hundred Tang poems. We often encounter practical problems that can be solved through our knowledge system. Even today, we have introduced some random variables into the model, which can make today's deep applications based on large models very rich and powerful.

We think that today's big model is still in the early stage of development. The so-called big model brings a series of applications, as well as the real core capabilities of the big model. I think we may only see part of it now. In the next few years, we should be able to see more innovative applications develop.

Just now, all three talked about the birth of phenomenal AI products like the current one, which cannot be separated from the development of AI technology. Will the development of AIGC technology bring some changes to our whole society or production structure in the future? This may have some connection with the lives of ordinary people. I would like to ask two professors about this question. First, please ask Professor Sun to talk to us.

AIGC is generated, so first of all, it should have a direct and profound impact on content production and content creation. I think there may be several points. One is that it can significantly reduce the production cost. In the past, you had to do a lot of work to make a painting. Now, with a big model, he has learned all the human paintings. Later, he can quickly do a lot of things without you. From the perspective of production, if from the perspective of industrial production, he will significantly reduce the cost.

Moreover, it will significantly reduce the threshold of access. For example, if you want to do content production, it is usually a big organization or a large enterprise. It is difficult for individuals to do this. Made from. But now with AIGC's AI tools, not only 2B, but also big B can do this, and individuals can do it. 2C and 2B may be on the same plane, which is actually greatly conducive to mass entrepreneurship and innovation.

On the other hand, it will significantly improve the efficiency of production. It is fast. This obviously comes out in seconds.

Moreover, it will significantly improve the quality of products. In particular, diversity and personalized production. If you are not satisfied with a painting made by computer or artificial intelligence, I will make another one for you. It is almost unlimited until one meets your requirements. If you modify it on this basis, you can still inspire you. The possible space for the machine to generate may be larger than our imagination. It inspires you, It may not be completely reasonable. It is also important to inspire you to do some more creativity. Creativity on this basis will greatly improve its diversity and personalization.

The direct effects, including related industries, are very clear, and the space is very broad. Going further, we are not limited to this. With the big model, we can sort out information in the form of knowledge and get through to it. At this time, we can do other more complex tasks, such as macroeconomic decision-making and more complex tasks, which may be better than before. It is production directly. However, if we move forward, there will be wider application space.

Thank you Professor Sun. Just now Professor Sun said that AIGC technology has brought great savings and changes to our production and life, including costs. Will it bring some pressure to people. I would like to ask Professor Huang to tell you something.

AIGC, AI generates content. As we know, several years ago, UGC users generated content. Going forward, we will not say, but we all know that it is professionals who produce content, and it is writers and painters who can produce high-quality content after long-term training. Now that AI can do it, it will definitely have a profound impact on people in all aspects.

On the one hand, we traditionally think of creativity. For example, I am a writer and I write a novel that everyone is willing to read. I think it is very creative. I am a painter and graduated from the Academy of Fine Arts. How much is my painting worth. At that time, we thought that this was creativity. Today, we look back and see that this kind of creativity can be divided into two parts. One is true originality and the other is composition, imitation or creation based on the works of predecessors. These are two kinds of creativity. The latter one is now available for AI. It is truly original. For example, Qi Baishi School of Painting, a new kind of painting has emerged, Not before. It started with him. Einstein of physics, we didn't have the concept of relativity before, he had such a flash of inspiration in his mind, and produced a new theory. This kind of creativity, today's AI, I think is impossible, at least the current AI still lacks such ability. But like the others just mentioned, we originally thought that AI could do more than 90% creativity, because most people engaged in more than 90% creativity, rather than the top creativity, would definitely have a profound impact on this kind of career and work.

Because many of us today, in addition to the top few just mentioned, there are also some people engaged in material production. A large number of people are working on content or cultural or spiritual products, which will be directly affected. But I think the biggest impact may be that if this thing can be replaced by AI, if people don't do it, AI will be cheap, convenient, and easy to get. In fact, the production of the top achievements may also be affected, because if people don't work hard to follow the previous less creative work, they may not be able to follow the original things. So, for the development of human society, we should think about what we should do if we want to hand over the physical labor to the machine, but now we want to hand over most of the mental labor to the machine? Although I said just now that there are still some people who can make innovations, the problem is that they may not get there. So, this is a very profound problem. On the one hand, in a sense, it is welfare. AI has helped us do so much mental work, but on the other hand, we should think about the social structure and how our human beings develop.

Just after listening to Professor Huang's statement, I watched the news a few days ago. AI anchor is very popular now, and many programs have invited AI anchor. But I think AI anchor should not be able to participate in our interview programs, and may not be able to do so at present.

This is not difficult. I don't think it's a problem to replace the anchor. Dialogue and exchange.

It's all possible in the future, isn't it?

Even soon.

I still have the pressure.

There must be.

Just now, when we talk about large-scale models, we mean large-scale AI models. Now it has become the infrastructure of the core technology for the development of artificial intelligence. Compared with the small models trained for specific application needs, it can provide a key impetus for artificial intelligence to move from special intelligence to general intelligence, so as to better provide inclusive applications, What do you think of the current development of our domestic big models and AI? What is the standard?

We just mentioned that the development of large-scale models, especially the super large-scale models, requires strong computing power, as well as a series of research and development of the models themselves. Today, we have arrived at a very large-scale model, which is actually a representation of the overall technology of AI+cloud computing. In other words, all today's big models can not be separated from the following infrastructure, especially the computing power such as cloud computing, which can help us to do model training and model services. In these two aspects, our country has very early investment, and there are also many innovations in this area. For example, in the whole computing power part of the foundation, Alibaba Cloud is also the top three cloud service provider in the world in terms of cloud computing. In particular, Alibaba Cloud also has its own large-scale operating system, Apsara Cloud operating system, which can organically connect millions of machines to form a super computer. Provide relevant computing power for everyone through public service. The entire operating system can also provide a solid foundation for all aspects of our AI training, including data processing, through today's internal software resource scheduling.

Our large model training also requires some special hardware and scenarios of intelligent computing, especially a high-density computing cluster. In this regard, not only does each machine have enough computing power. These, together with our storage, our network and a series of optimization of our software scheduling, really lay a solid foundation for the development of our big model and AI. I believe that the first and foremost condition is how to make good use of cloud computing today, which can lay a solid foundation for the effective and rapid innovation and development of our AI computing power.

The second aspect is the model. Our country also has a lot of innovations in models. Today, including the colleges of Professor Sun and Professor Huang, there are many researches in this field.

Our national research on large-scale models is still very early. For example, even in Ali, the Ali Dharma Institute has been doing a series of large-scale multimodal pre training models for a long time. At first, our idea is very simple. When we see a thing, how to understand it? It may be a commodity or a service. There are language descriptions, text descriptions, and pictures about such a thing. We realize that in order to effectively understand such a thing today, we need to understand such an object from multiple channels, from the perspective of text, natural language, pictures, and vision, to understand such a commodity through our knowledge system in many ways, so as to integrate knowledge, Only in this way can we effectively recognize and understand everything. This is also the original intention of our first multi-mode pre training model. About 2019 and 2020, we will start to make a lot of investment in this area. In the past few years, from 2021, we have also released a 10 billion parameter scale Chinese pre training multi-modal large model. Then we went from 10 billion to one trillion, and to one billion also means that our country has the ability to train the entire large-scale model technology. After that, we have also made a lot of research, development and innovation of multimodal models in various aspects, including dialogue and vision. In particular, we were able to make the model exceed human evaluation standards for the first time in the VAQ contest of reading pictures in the past few years, and also exceeded human evaluation standards for the first time in Chinese reading comprehension, This also represents that all our models have made great progress in the past few years.

Last year, Ali gathered all the models together and released the "Tongyi" series of large models. The reason why we gathered all the models together can form a multimodal training paradigm, can provide a unified training framework, and more importantly, these models can be shared, open source, and open to everyone, We can let all walks of life create and continue the secondary development on it. Only in this way can we gradually establish the whole ecology of today's model.

Thank you for your efforts in this field. I think that the two professors are at the forefront of the development of large models and AI in China, which should be said to be the most direct research work. I want to consult two professors on this issue. Professor Huang first came to talk to us about the current development of AI in China.

Our AI practitioners and netizens may not pay much attention to each other. For example, after the product comes out, everyone's eyes will suddenly light up, as if AI has suddenly changed a lot. But for AI people, maybe the river is flowing, and some waves appear at this node. So, I think from this point of view, when I talk about rivers, in fact, many people around the world are doing AI research. There are many people doing AI research in China, and China's research is originally part of the world's research. China's talents, papers and other contributions are now obvious to all, and it is an important force for the whole world to do AI research. Therefore, in this sense, we believe that the time has come to this point. China has accumulated this technology and talent. Now we need to do a product or a service. These conditions are all met. However, from a real AI researcher to a service that can be used by thousands of households, there is still a lot of investment from enterprises and industries. First of all, our basic conditions are quite good. However, there is still a lot of work to be done if we want to look at this problem in such an infrastructure as the big model or in such an era as the big model.

I would like to take a long view. I don't care how many models have been released this year or next year or how many products have been launched. Of course, I have a very direct feeling to everyone. I think we can look at this from a slightly longer perspective. The big model itself is an era. What kind of era is it? It extracts from massive data and condenses such a model to provide services to thousands of households in an intelligent way. If you want to think about this kind of service, we say that the future is an intelligent era. What is the feeling of the intelligent era? Everything in the home is connected with an intelligent cloud and network, just like the appliances in the home today are connected with electricity. When they are turned on, they light up. In the future, every device or thing we use will be smart. Where does it come from? It comes from the intelligence continuously sent through the network. Therefore, in that sense, the whole society is constructed as a set of intellectual infrastructure, which is similar to the electric power infrastructure being used today, for the same reason. Therefore, there will be thousands of enterprises providing services at different links. Therefore, in that sense, on the one hand, this matter is not a special technical secret, as if who can use it, who can do it, and who cannot do it, is not the same thing. Just like power generation, there are more people in the world who can generate electricity, and there are also many large power plants that can operate. However, there are only a few power grids in the final power grid, and there are only two power grids in China, the State Grid and China Southern Power Grid. In the future, intelligence will be similar. To construct such infrastructure and intellectual services, it is necessary to promote the formation of this matter from many aspects, rather than a few simple products. Maybe different institutions, different industries, universities, research institutes, and the whole society, including investment and financing. From this perspective, everyone should find a more appropriate position to make their own contributions to the future common vision.

Institutions like our school or the Zhiyuan Research Institute where I am now working, because we are engaged in technology, our contribution is ultimately reflected in technology. Our focus now is actually to build an open source and open big model technology system, because sooner or later, no matter who runs the model, you always need to train the model technology to be as advanced as possible, so as to reduce the cost and popularize it to thousands of households. Therefore, in order to open source, many researchers are doing it. If we do not open source, these things will be closed to a few monopolies. For example, we care about domestic and foreign products. Of course, this product should face users directly. But as I said just now, in the long run, this thing is inclusive. Inclusive things need more people to bring their intelligence, such as the improvement of algorithms and training things, together in an open way to promote the faster arrival of this society. Now, the whole society may be a little worried about the products. This is not urgent. It must be to build an intellectual society. At least it should have a vision of several years, not months. Now it is urgent to make an immediate response. It is a bit extreme, which is not good.

Thank you, Professor Huang, for sharing your views. What's your opinion on this issue, Professor Sun? In fact, many people are concerned about whether talent development and training can keep up with the pace of rapid technological development?

I particularly agree with what Mr. Huang said just now that we should look at this issue from a higher perspective, and we should take a broad view of the long scenery. Wittgenstein, a famous philosopher, once said that "the limit of my language is the limit of my world". Let's deduce that the limit of human language is the limit of the human world. The big model actually integrates all the text information in our world, as well as further expanding image information, multimodal information, etc. through language. In fact, this matter can be said to be the limit of the big model, which is the limit of our future intelligent information processing. It may not be exaggerated. Just as Mr. Huang said, it is an infrastructure of intelligent information processing. Maybe we should look at this problem from this perspective. If we look at it this way, it will not only be a problem of enterprises, but also rise to the national level.

Our country should say that since 2010, when this wave of AI has reached its climax, our country has made remarkable progress in AI. Just now, Mr. Huang said that we have the ability to do this, which is not easy, very difficult. AI is actually a result of the game of various forces around the world, including countries, enterprises, and individuals. It is a peak competition. I call it a peak competition. It is very difficult, and we can keep up. Therefore, this should be an important guarantee for our country to have made a lot of achievements in large model and AI technology. Where are we weak? Our ability to follow is very strong, and our ability to lead is now a big weakness. On the whole, we can keep up with us, but there are few leaders. This is a problem we should think deeply about in the future. But the ability to follow is also important. Why? Our ability to keep up shows that we have the ability to innovate in some important AI application scenarios. Because AI application scenarios are also a major challenge, foreign countries are also exploring this aspect. It seems that on the application level, it is actually closer to the national economy and the people's livelihood. This is a big weakness, and the whole world is a big weakness. Our ability to keep up with cutting-edge technologies indicates that we have the ability to make breakthroughs in major application scenarios, which is also very important.

Of course, it would be even better if we could lead the way in the future. This problem is complicated, and we may not have time to discuss it today. The talent cultivation you just mentioned is actually one of the important factors. Our country has cultivated, on average, high-quality talents and artificial intelligence talents. These years, there is no problem. A large number of talents have been cultivated, which has supported the development of artificial intelligence in our country, including universities, research institutes and even enterprises. But we also have a big weakness in the cultivation of top talents, which matches our top achievements and leadership. Of course, this problem is quite profound. For example, I would like to say that we may further improve the students' vision, that is, their academic vision. For example, artificial intelligence really needs to have major innovation, which may require interdisciplinary crossing, especially a very deep mathematical foundation. We students are weak in learning from physics and chemistry. For example, computer learners may only focus on computer knowledge. But the development from abroad, such as the Diffusion Model, actually draws lessons from physics and chemistry. You can see that their students can read the paper Physical Review, and we may have fewer students. Therefore, the awareness of interdisciplinary cooperation may need to be cultivated. In short, this issue is very complicated, and there may not be time to start today.

Thank you Professor Sun for sharing with us. After listening to the three people's explanations just now, our domestic big models include the development of AI technology, and now we have also made many achievements. As Professor Sun said just now, it may take a long time for us to change from following to the ideal leading state. We see that this year's government work report also emphasizes the need to vigorously develop the digital economy, which is a very big environment and opportunity. How can we promote the innovation and development of AI technology to better integrate China's digital economy and the real economy?

As we mentioned just now, the essence of big models is computing power and breakthroughs in models. In order to further promote the overall innovation and continuous exploration of AI in our country, I think there are several aspects: first, we should be able to provide corresponding computing support on the entire infrastructure. Because the training basis of all models still depends on the general computing power. In this regard, we hope to have a more large-scale computing platform that can provide more public services, perhaps more organically combined with the current development of cloud computing.

Just now, a series of changes are taking place in the paradigm of model training. Often, we used to train a model for a specific problem, because with such a pre training model framework, we can begin with more knowledge accumulation, and not train the model with any specific problem. This part, in this stage, is often called the pre training stage, which requires a lot of computing power and data processing. In this regard, two professors just mentioned that today we are going to compete with top overseas institutions in this regard. Today's competition is not only about AI, but also about all aspects of AI+computing power. In this regard, we must lay a solid foundation for the rapid development of AI in the whole country in terms of infrastructure, which is the first point.

Secondly, I also agree with the two professors. Today we will talk about the rapid development of AI, which requires our model to have such an ecosystem, and more importantly, we need to let everyone have a goal of open source sharing. In fact, today we all pay more attention to the underlying model, that is, today we pay more attention to language models and so on. Generally speaking, professionals talk about basic models. The basic model requires a lot of computing power and knowledge accumulation. At the same time, we also need industry models. For example, if we really want to apply these models to an industry, we need a large number of industry algorithm engineers to constantly tune according to our current problems. Of course, with the development of the whole big model today, a series of changes have taken place in the tuning of the upper model. In the past, we used to do more fine-tuning, but now we see that fine-tuning is not necessary. Through a series of mandatory innovations, we can change a series of behaviors of today's models in the actual process and help us solve problems effectively. Therefore, in this regard, I hope that you will not only focus on our underlying models, but also explore models related to each industry line, including a series of applications of AI, which will play a crucial role in the development of AI in all aspects of our country.

Third, with the development of AI today, we still need to lower the threshold of AI use. As mentioned just now, AI training costs are relatively high, and sometimes it also requires quite professional knowledge. Today, every industry also needs to carry out specific production for this industry, and how to further reduce the cost of using AI. Even today we can let students, even primary school students, today our primary school students can program on PAD, in fact, they should also be able to quickly use these models. Last year, we set up a community called Mota Community to try to solve this problem. In just a few months, nearly hundreds of thousands of developers have come to our website to download models. In particular, we provide a simple and easy-to-use programming interface, which can use a very complex model with just a few lines of code. This is just an example. I want to say that today we need to reduce the application cost of models, so that people in all walks of life, even those who are not computer professionals, can use AI in all aspects, and can be applied to your actual scenarios. Only in this way can we truly achieve the common development of AI's benefits and AI's innovation.

Thank you, President Zhou, for your explanation. Now let's invite two professors to share with us how to make our AI industry develop rapidly.

The combination of digital economy and real economy mentioned just now may need to be viewed from a longer perspective. I even want to make an analogy like this. For example, the Industrial Revolution and the steam engine have entered an era of machinery dominated by human and animal power for hundreds of years. It should be said that in the early 1900s, the whole power industry began to enter thousands of households and enterprises. Let's think about the difference between the society with and without electricity. Over the past 100 years, information such as computers, the Internet, and mobile Internet, in fact, I think that information itself may be difficult to compare with the intellectual age that we are going to see next. Because when information is provided to people, or people are the main body to process and use the information. The so-called intelligence era is the whole information system. AI system is using this information to generate new abilities and new intelligence. So, as I said just now, the industrial revolution, the electric power revolution and the intellectual revolution were of the same magnitude. Those in front of the information are all preludes to preparing materials, data and computing power for this intellectual revolution. Only when intelligence, like electricity, continuously supplies the whole society, will we enter a new era.

How do we view this problem in this sense? When we talk about the digital economy, the digital economy must be based on new technologies such as data, big data and AI, so that all kinds of applications, people, enterprises and all elements of the whole society can simply use intelligence to drive. Because a lot of things that run in this society depend on intelligence. We build houses, our agriculture and industry, not to mention all these activities related to content processing and information production, are all intellectual. We used to rely on human intelligence. Just now we said that the industrial revolution has liberated our physical strength to a certain extent. This time, we will liberate our intelligence. It will bring particularly profound and huge changes. As I said just now, China is in the forefront of this era, and we are not backward. Unlike the previous two times, the historical reason is that we missed a period of time in a sense. This time, we are almost ready when mankind is about to enter this era.

We look forward to such an intelligent era. Thank you Professor Huang for sharing. We also asked Professor Sun to sum up how we can promote our AI technology to enable our digital economy and real economy, in fact, ultimately to bring people a happier life.

The big model and AIGC actually create infinite possibilities for the digital economy. Because it uses computers to understand all human information, text information, image information, and all kinds of information from the perspective of computing, I will do information processing on this basis, so its depth and breadth have a qualitative improvement than before, so the possibility is unlimited. However, it is still very difficult to implement the possibility into reality. You can't expect the big model to be a panacea, which can be used everywhere. You can't expect it. When it comes to major application scenarios, it still requires deep cultivation, deep cultivation, and hard intellectual labor. It is not a simple matter.

At this point, the whole world is actually in the exploratory stage. You can think of the big model AIGC as a concept, which shows a wonderful prospect. There is still a long way to go before this concept can be sold to thousands of households. In fact, it brings a good opportunity to our country because our country has a large market and various application scenarios. At the same time, it also brings a lot of opportunities to various enterprises, whether large or small, especially newly started enterprises, as well as a lot of space for personal development. I just said that AIGC itself can be very personalized, which lowers the threshold.

In general, our country has worked harder in recent years. With our strong ability to follow, we should be able to make world leading achievements in major application scenarios. Of course, there are many problems to be solved. For example, Professor Huang said just now that there is a big problem with data.

Is this a security issue?

no You must first have big data to make a big model. Big data is the advantage of our country, but it does exist. For example, you are subject to various restrictions, and this data may not be available. Our ideal state of big data, how can you break the barriers in the industry, such as sectoral barriers, regional barriers, and under the support of morality and ethics in line with the world, For example, on the premise that we maintain personal privacy and comply with national laws and regulations, how to integrate these data for better sharing, utilization and development is a big deal in itself. Our country has made many efforts over the years, but this problem has not been solved satisfactorily. This time, the country has established the National Data Bureau, which I am also looking forward to. It also reflects the country's determination in this regard to take this matter forward. This matter has taken a big step forward, and the construction of the big model has been implemented from the national level. In fact, there is still a lot of hard work for us to do, but it also provides many opportunities.

Thank you Professor Sun. The development of AI industry should have a broad future, but the hardships on the road can be seen.

But hardship heralds opportunity.

thank you. I think the development of China's AI industry may require the joint efforts of our schools, scientific research institutions and our enterprises. Our country's digital economy may usher in a new development prospect, and our people's lives will be better and better. Once again, thank you three for coming to our reception hall. Thank you three. Thank you for watching this program. See you next time.