2023 World Artificial Intelligence Conference Opening Ceremony (live transcript)

《》In the days after it was broadcast, there are content and videos about it on Douyin, Weibo, Station B and other places, especially the opening ceremony of the 2023 World Artificial Intelligence Conference (live transcript) was recentlyThe audience has discussed it the most, so today I will chat with all my friends.

Some models in the picture, including GPT 3 to be released in 2020, will not be officially released until December 2022.In this in-depth process of modern data, on the one hand, we perform some operations on the overall neural network, including the processing of some information, and gradually make some progress in the algorithm, and the algorithm becomes more and more advanced., the algorithm becomes more and more deep, including processing some things more and more powerfully.The other layer is actually its bottom layer. In fact, it is some of the developments in the chip industry that are quite popular now, especially the emergence of CPUs and the overall emergence of CPUs. It has actually pushed the overall computing power to a very high level.The high ground.Well, it allows us to truly integrate and digest a lot of human knowledge. It took him several years to do it.But it turns out that if our computing power is not enough, that is, when the underlying hardware support is not enough, this thing cannot be done.So as our understanding of algorithms, including some models, gradually iterates to the other side now, in fact, the arrival of its underlying logic makes these things truly serve us.Is this somewhat understandable to me?Can.

Understand, in fact, all technologies are not born out of the blue, yes, they must have been accumulated over a long period of time.Yes, and the level of development and iteration of this technology has led to this, well, I can’t say it has led to this, so that this technology suddenly became known to everyone. Maybe the process he is working on is invisible to us.of.Yes, but what we see is the result, the result.Well, what about you, Grandma Wu?Do you have any knowledge about these technologies?You can make some supplements with us.

I definitely can’t get to know their technical staff in depth, but why are the fields different?Yes, but everyone will pay attention.Well, because I think that for people like me who engage in science popularization or science education, or even science and technology innovation education, because I mainly do these three jobs, then I will definitely pay attention to these technical issues.But I think how to master this technical problem?How to let children understand and master it, I think is very important for all our teachers.Well, if the teacher doesn't understand him, I think there will be some regrets in the teaching, so I will still pay attention to it, especially in the deep city that I mentioned just now.Many people are carefully studying how the robot learns during the development and training of robots. So how does its neural network gradually become more robust?I.

I would like to make a small request. What kind of language, simple and vivid language, would you use to explain this concept to teenagers or young people who don’t understand?

It seems that many teenagers can understand this just by talking about it.Well, because they are like you in normal times. For example, some people in our family will have some small ones, small ones, small ones, etc., right, right, and then they also have chips, right?They knew about this chip during the learning process. How do I want this microcontroller to work?Why should I teach him? Do I want to program?It has to be poured in, so he can still understand these things.

I am also born with this curiosity.right.

Yes, yes, I think it is too easy for teenagers to talk. If you talk to some elderly people, it may be a tiring day.Yes, well, it’s easy for you to tell it, and it’s easy for them, but the key is to let them know what we use this thing for, and what practical problems I can solve.So when I guide teenagers to do projects, I don’t mean to learn 3D printing, digital cutting or something like how to build a model, but to solve your own problems.I just hope that he can find any problems around him or in his studies or life that deserve his attention.Well, he was able to detect this problem keenly.

Yes, alas, everyone hopes to solve it, and I may be able to solve it. I may solve it. In that case, I think this is a particularly good phenomenon.Therefore, if he wants to solve his problem, he must learn. Then his learning process will be particularly vivid and interesting. He himself also has a very strong desire to do this, and then he will master it during the learning process.Some techniques.

I think the most important thing for teenagers is why should they learn this technology?He must make it clear that he is not saying that I will teach programming to a class, and everyone will learn it. After learning, they will stare at the screen for a long time, and in the end it will be over without knowing where to use it. This semester's class is over.I think this is not particularly important. What is important is why he is learning technology. He only has this strong desire. Can I give you an example?OK.

There was a classmate who said that his mother walked into the hallway of his home on her way home.Well, he wanted to hear some of the footsteps as he entered.Whether the mother's footsteps are heavy or light, it means that the mother is tired today, or is at work, or there will be some unpleasant things.She is still very happy. She wants to understand her mother's mood. As a daughter, she cares about her mother. When mother comes in, should I put down the homework in my hands and go pour a glass of water for my mother, get slippers, etc.?Sometimes he is very nervous about homework and may ignore his mother's questions, so he needs to use artificial intelligence. He needs to install a microphone in the corridor, right?Accept the sound of his mother's footsteps, and he needs to do some research on what kind of sound his mother's footsteps will reflect his mood.Well, then you need to enter the sampling volume into the computer, collect a large amount of data, and then compare the data with some of his mother's performance.This is a good case, but it also uses artificial intelligence. This is a bit of a calculation, yes.

Bar?Yes, yes.

Right, forget it, I see Jianguo is smiling all the time. I wonder if we both think the same thing. She is a considerate girl. If it were us, maybe we would think the same thing.Well, is this going to be a desertion?That's right, I'll hurry up after hearing if the parents are back.

Put on the TV.

I quickly returned to this desk and started doing my homework and reading.Okay, I think what you are conveying is that our life and study are inseparable.Well, there may be scientific principles everywhere in our lives, things that we need to observe and see if they can be applied to life to answer questions in life.Not just for exams, not just for that score or a certain node, such as going to college, or not just for these, but for some further uses.

Yes, well, in fact, after the student union has studied it, he will not fail in the exam or study well. If he studies well, he will do well in the exam.Because I often tell parents that many of them question how you can do well in the exam if you don’t study the questions and do research.In fact, from two perspectives, first, during the learning process, if he works hard to study extracurricular review, it does not mean that he does not do the questions, but he also studies and reviews the questions, and has a thorough understanding of the knowledge to do this.topic.Well, it's relatively simple. Then by then, he will already have a habit of research. Then when he takes the exam, first of all, won't he have mastered all this knowledge?He must have done well in the exam.The second most important point is that because the current examination has also been reformed, there are many situational questions to be tested.

Well, that’s the question for your exam. It’s not just about drawing a model diagram for you based on what is known, such as that wire or a stick, its force, etc. It’s not that simple.It will tell you what scene it may appear in, um, even Beidou or something.Just one paragraph describes the phenomenon. Some of the data in the known conditions are not necessarily used. Unlike in the past when one data was omitted, I was very nervous. It seemed that my answer was wrong, but some of them now require you to build the model yourself..So do you think people who have been studying questions are very nervous when they see questions like this that they have never studied before?If he gets nervous, it will affect his performance.That.When people who have always been used to doing research see such a problem, their habit immediately becomes, oh, no matter how well they do it, they will study this topic again.Alas, actually.

Behavior forms a good habit, thinking.

Habit.Well, a habit of thought is what it's conditioned to become.Hmm, then it must be very good and can solve some problems.Well, so during the exam you are always the one who knows how to research, and it must be about nucleic acids.

Well, so it’s not that we don’t answer questions. What we answer are questions in daily life.

Well, the word "brush" is, don't say "brush", the word "brush" is a bit radical.good.

We don't talk about brushing.

This word can mean doing questions and researching questions.

right.Has Jianguo paid attention?Grandma?

I just started paying attention to this yesterday.

Yes, from yesterday, from today should we, I have.

I became a fan after watching Grandma’s Final Food, hahaha, you know.

I don’t know that Grandma Wu’s number of fans has exceeded 5 million in less than half a year since this video was launched, and the number of fans has increased from 500,000 to 600,000 every month.Right, so actually this is a very gratifying phenomenon.Yes, so many netizens are saying that this is the star we should really pursue in this era.Well, yes, so we are also calling on more people to pay attention to this video, so in fact, it is the first time for Wu to start a business, and she succeeded in her first start-up, hahaha, yes, you are doing thisIs there anything in the video that just feels like something?Can you share with us whether you encountered some problems at the beginning, encountered some confusion, or encountered some difficulties, or did it go smoothly from the beginning?

In fact, what does everyone think of me, but I personally feel that I still act as my true self.But the team behind me supports me and encourages me.At first, I thought that because I actually like to play, it’s not that kind of play, like playing ball, but I like to play with these gadgets, which have physical principles in them.I feel that it is very strange and very curious.Well, usually with the students or the young teachers around us, we would, oh, take a toy, and we would study whether it could be turned into a physics teaching aid, etc., just play with these things, and sometimes we would not get started.Or he could make any fool of himself.So I am usually a very happy person, but I feel restrained in front of the camera. I feel a little embarrassed when I show my cheerfulness in front of people all over the country, but.

Very cute, hahaha, no.

People encouraged me. After encouraging me for a long time, I thought that I should just go for it, because I still need to spread knowledge.In addition, when I am happy, I hope that everyone around me is happy, but if more people, thousands of people, are also happy, it will be quite good.Well, in addition, I also have an ideal, a dream or a fantasy, which is that I hope that everyone will study physics.Physics is a very important basic subject. If many people across the country feel afraid of physics, and parents feel that when their children learn physics, they become nervous first. Then, for children, especially girls, parents will feel, ugh.Hey, don't learn badly, I'll take the outside to make up for it first.So in Shanghai, don’t we enter junior high school in sixth grade?Some sixth grade students have to go outside to make up for math and physics, and then make up for it again in the first grade of junior high school, and then start learning physics in the eighth grade. He has learned it several times, and he feels that he understands it all. In fact,He knows how to answer questions.But he doesn’t necessarily understand, or he doesn’t necessarily have feelings for physics. Then when he does it, yes, he will feel bored, or once he is a little frustrated when doing questions or taking exams, his score will not be as good as other people’s.Will be anxious.So I hope that everyone will like to play physics since childhood. In fact, for newborns, the first thing they come into contact with is a physical enlightenment, such as touching their mother’s hands or touching some soft things.Or something hard, its language is not soft and hard, but it has this concept.So, for example, when a child is two, three, or three or four years old, he can build building blocks. You will say that this is a problem of gravity, a problem of stability, and why the center of gravity should be placed on the base. He does not understand this., but he will figure it out.Oh, it’s too high. The parents who raised it said they clapped their hands, “Oh, that’s great. You made it so high. The child is happy. If it’s a little higher, he’ll think about it.”

So these are all physics problems. I learned physics at a very young age, so I have been thinking about a problem recently, because I was thinking about this sudden change from Newton to Einstein. Then people at the time thought that Einstein’s theory was unreasonable., so now our people think, well, you can learn Einstein's theory to understand it, because he is about kinematics, and it can be solved with basic mathematical formulas without advanced mathematics.

So from this point of view, why can I now understand things that I couldn't figure out at the time, and junior high school students can understand them? Then why not learn physics well when they were young, so that they can understand it in life and play?Once you have mastered physics, your grandparents can learn it and learn it together with their children.So once you learn such fun things, when your children are studying physics in junior high school or high school, they really study it in depth, right?For example, calculus was so difficult when it was first invented. Now everyone even studies it in high schools and universities, and they think it’s okay. But I think it’s time for a breakthrough in physics education. In a society with artificial intelligence, it can be moreGood for making breakthroughs in physics education.

Well, yes, I would like to discuss this with you. In fact, we can see many shining points from grandma just now. Of course, in addition to her own ability and academic ability, we also sawIt’s curiosity, and how to protect and maintain this point of interest so that it can last for a long time. Maybe this is the driving force for us to learn well.

Including when we were watching the video, we saw grandma using some very imaginative props and some vivid language to make some very unfamiliar technical words very easy to understand for us.Moreover, in many of our lives, we are accustomed to phenomena, or phenomena that people do not pay attention to. Grandma can extract them and tell us the truth behind them. These are very interesting and informative, so do you think that in artificial intelligence?Will there be more times?Do you think there will be more help to popularize this science?Or do you think how should we popularize science to everyone in this era?

First of all, in this artificial intelligence society, I hope there will be an artificial intelligence grandma.Haha, then make one for me.

?Yes.

Why?In fact, I still have some things that I find difficult.Well, it's just, um, why?One is that I am older and have more physical problems.Another point is, um, a lot of explanations, I hope, because after all, explanations in the lens world are different from communicating with students in class.Well, during the student exchanges, I saw that the students spoke out on their own, and I was very happy to interact with the students.Yes, or the interaction is a physical issue, but I don’t feel that comfortable when facing the camera. For many long-form content, the editor should not use all my words, right?Do you want to repeat it again, or what?Don't let him pick out these languages ​​bit by bit, then I need to list the languages ​​​​in detail.Well, sometimes I have to use some written manuscripts, but the written manuscripts must be seen by the eyes, right?And does he have a prompt screen now?But I can’t see clearly with my eyes, and I can’t see clearly from a little further away.My eyes are slightly different now, that is, the left and right eyes do not overlap. Well, the flash is so powerful.So this line of text only has two lines of text in the middle, which are overlapping and cannot be seen clearly at all. So I feel that there is still a lot to say, and there are still a lot of short videos and small experiments to be done.

Well, it’s too late, because netizens also asked a lot of questions, and I think netizens are helping me study.Well, that’s the question he sometimes asks.Alas, how can we solve it better?Or is that the question.Well, I don’t necessarily understand heaven and earth.Then I will study, so I am especially grateful to netizens.But it’s best to have a person help me with this, artificial intelligence to help me, then we can plan it and ask him to help me talk more.

thing.Well, actually you should start more live broadcasts.

Why?I?If you don't ask me to speak about what I have written, let a grandma speak on my behalf.Hahaha, please help.

Bar.actually.

It is possible, Jianguo needs to work hard.

That's a must, hahaha.

Are you in this medical recovery now?Yes, yes, I just mentioned whether there is some logistics.

this?What we do is not logistics, it is equivalent to talking about assisted travel.Well, help, because in addition to medical rehabilitation, we also make some products to help the elderly and disabled people do some travel.

When did you become interested in this technology?Are you at Harbin Institute of Technology?

Yes, I study computer science. When I listened to my grandma’s talk just now, it really reminded me of some things from my childhood. Because my grandma and I have my father who was also a physics teacher, hahahaha, so I actuallyEver since childhood.Yes, well, these are the things that this world is edifying.Oh, yes, yes, I feel that many things in this world can be created since I was a child. I have always had this idea since I was a child, and I feel that many things in this world can be created.It doesn't mean that the world is like this, you have to accept it, you have to adapt to its problems, um, but it means that there is a creative thinking.

Okay, let's find another time to talk about your experience, because our guest who is about to return to the scene will share it. Next, we will listen to the guest who is on the scene. Let's talk about it again.After waiting for a while, we entered the scene to watch the sharing of our guest speaker.To summarize briefly, what the two of you just said is that we need to protect everyone’s curiosity from childhood. Well, how to protect their interest points and allow them to continue to be interested in the same thing.Keep this spirit of exploration alive.Then these may be the motivation that can enable our new generation of young people to embark on this path of scientific and technological innovation, or to continuously improve themselves on their own career path.

Then we will also hear some guests share their experience in this, especially some well-known technology experts and technology geeks who are familiar to everyone. They will also come to share their experience with us later.From some stories and thoughts during the growth process, we can also extract a lot of essence from them that can be conveyed to our teenagers.So let's go and take a look now.

Good leader, distinguished guests, ladies and gentlemen, good morning, everyone. I am Huang Wei, the host of Shanghai Radio and Television Station’s China Business Network.The Intelligent Connected World Generates the Future As we saw in the opening show, the advanced AI and artists performed together. Currently, a new round of accelerated evolution of artificial intelligence, represented by deep city artificial intelligence and large models, has led this round of technology sweeping the world.Revolution and industrial revolution, mankind has never been closer to a new era of general artificial intelligence, which has also aroused common attention and heated discussions around the world.

The next person we want to invite is an old friend of our conference. He has always been at the forefront of global technological exploration and promotes technological progress and industrial development with his advanced prediction and execution capabilities.Next, we will invite Mr. Elon Musk, CEO of Tesla, to bring us his AI creation in the new era of general artificial intelligence.Please look at the big screen.

Hello, everyone in.

Shanghai.Hello everyone, friends from Shanghai, especially Secretary Chen, hello, I think artificial intelligence will play a very profound role and influence in the eyes of human beings in the future, including playing a very profound role and impact on civilization.We have also seen the explosive growth of digital computing power. The most critical indicator to consider is a ratio, which is the ratio of machine computing power and biological computing power for digital computing.What's the meaning?It means how powerful calculations can humans do, and how powerful can computers and machines do calculations?So what is the ratio of computing power between computers and humans?This rate is getting higher and higher every year, so what does that mean?That is, the gap between the computing power of machines and living things is further widening, which means that after a period of time, artificial intelligence will account for a lower and lower proportion of all intelligence. Compared with machine intelligenceIn terms of impact, this will be a fundamental and profound change. It is difficult to understand it now, but it may be said that this is the most profound period in human history.Well, Tesla's Optimus humanoid robot is still in a stage of development, still in a relatively early stage, but we will have a lot of robots in the future.So what needs to be considered immediately is another issue of proportion, that is, what is the ratio between robots and humans?It seems now that one stage will exceed 1:1, which means that the number of robots on the earth will exceed the number of humans, and their computing power will be much stronger, so this seems to be one of the development trends.This will have both positive and negative impacts.The positive impact is that we will enter a post-shortage era, and there will be no more shortages. In this case, this era will be an era of abundance, and you can get what you want immediately.Because if there are a lot of robots in these future worlds, their production efficiency will be much higher than that of human-led production, so this is a very big and profound change, so we have toGreat care is taken to ensure that its end result is beneficial to humanity.But one of our current development trends is, for example, in the field of Tesla's humanoid robots. We see that there will be more and more such robots in the future.Optimus, his human fund, can help people do some work. Even he is not very intelligent, but he has enough intelligence to do some boring, repetitive, and dangerous work that humans are unwilling to do.This is our goal, so one of the goals of the Optimus humanoid robot is to do these things that humans don't want to do, so this may still be quite useful.I don’t want to be overconfident or optimistic about Tesla now. Optimus’ role will definitely be very important, but it has autonomous driving, and Tesla is also very interested in sharing its autonomous driving technology with others.Car manufacturers make a sharing and licensing.As long as the technology permits, we think this is a very useful technology. I also think that this boring process for everyone to drive will be completely eliminated. It is beyond the times, and we will see that the use rate of this car will greatly increase. Generally speaking,Typically, a family car is used for about 10 to 20 hours a week, and most of the time is spent lying in the parking lot. However, for a fully autonomous car, the time it may be used will be much longer.It's 50-60 hours a week, a total of 168 hours a week.Therefore, its usage rate for fully autonomous vehicles will definitely increase five times faster than that of non-fully autonomous vehicles.At Tesla, we also want to provide this kind of technology, which is why we are willing to license these fully autonomous driving technologies to other car manufacturers for use.So what is the current status of our autonomous driving?Tesla, um, we think we are very close to, um, a state of fully autonomous driving without human intervention. We have tested it on the roads in the United States, and now we rarely need human intervention, so I willWhen driving a Tesla and using the latest FSD full-autonomous driving beta version of this technology system, you basically don’t need to touch any controller from point a to point b.So this is still a prediction and a guess.But I think that fully autonomous driving, or this level 4-5 fully autonomous driving stage, will probably be realized later this year.I've made predictions before and I've been wrong, but I feel like this prediction I'm making now is closer to reality than ever before.So there is another thing we need to worry about, that is, we need to worry about this kind of deep and comprehensive artificial intelligence, and we need to be a little worried, especially fully autonomous vehicles.For example, in this example, it is difficult for us to implement a fully self-driving car with limited and limited AI, but we think we can solve it soon.I have predicted that fully autonomous driving will be achieved later this year. There is no 100% guarantee, but the trend is that fully autonomous driving will be achieved later this year.But this kind of limited artificial intelligence is completely different from a comprehensive artificial intelligence.It is difficult to define comprehensive artificial intelligence. It is a type of artificial intelligence that is much smarter than humans in any aspect and is smarter in any field.Well, Tesla has not done research in this area. Other companies are doing comprehensive artificial intelligence research, but I think this is something we need to consider. It is very important now. There needs to be some supervision to supervise them to ensure that thisA very deep artificial intelligence.The kind of artificial intelligence I'm talking about is a bit close to tens of thousands of high-performance computers, or hundreds of thousands of high-performance computers, and sometimes millions of high-performance computers, the most advanced computers, working together to create a dataCollaboration takes place in the center to produce a super intelligence that is a combination of super intelligence. Such super intelligence has powerful capabilities, more powerful than humans. This is a risk and a worry.Then he may have a very positive future, but there is also a probability that he will have some negative futures.We have to do what we can to make sure that these negative, negative futures don't happen, and then the positive futures do happen.Okay, I have a lot of smart people in China. I have always admired the wisdom and energy of the Chinese people. Once China makes up its mind to do something, it will definitely be able to do it.The industry is doing very well, and all industries in various economic fields are like this, including the artificial intelligence industry, of course, so I believe that China will have strong AI capabilities. This is my prediction.Thank you very much for inviting me to attend today's event and express some opinions online. We hope you find these opinions interesting.Thanks again to all my friends in Shanghai, and to Secretary Chen. I look forward to participating offline next time.

Well, thank you for the applause. Thank you Mr. Musk for sharing the content he brought to the conference today. Looking forward to tomorrow, innovation ignites the future, original innovation leads core technology breakthroughs, and Shanghai has always been at the forefront of the world. Next, we will applaud the Chinese University of Hong Kong.Professor Tang Xiaoou came to introduce the latest progress of him and his team in the field of original innovation.Applause is invited.

I will talk about this later in the round. This is very stressful. My son Samuel really wants to buy a Tesla, so I think I will buy him one after I talk about it.Dear Secretary Chen Jining, Mayor Justice, distinguished leaders, distinguished guests and friends, good morning.Hello, Shanghai.Although this title is from my heart, it is not my original creation. The originality of this title is actually very expensive, 5.4 billion, its cost.Hello, Li Huanying, this is when I accompanied my parents during the Spring Festival of 2021.

Well, I watched the box office of a movie originally created by Jia Ling from the Northeast. The box office in China has been calculated since 1994. The box office champion in the first year was FUGPT with 25 million. So I think today’s movie is thisThis kind of family drama in China can create a box office of 5.4 billion. One of the main reasons for this miracle is that compared with 30 years ago, we all buy tickets to watch movies, and there is no piracy.That is, when the original author can make money and have enough to eat, he will have the strength to make the next movie.

Chinese literary and artistic writers do not have high demands. They just want to be given food and sunshine. Then I feel that China’s technological creators are beginning to see the light. As they are riding their bikes on the road to the canteen, they hopeThere is also food to eat.Well, I’m obviously not here to talk about movies today. I think what everyone wants to hear most is artificial intelligence and this big model. Then you can’t talk about big models without this Transformer. You may ask, this photo of TransformersWhat does it have to do with Transformer?It doesn't matter at all, it's just because their English names are the same. I put it here mainly to attract everyone's attention, because in English it means "get your attention".Because the title of the Transformer article published by Google in 2017 is called attention is all you need.

Transformer is a new type of network architecture. Through this calculation method based on attention, that is, the attention mechanism, it captures very long-distance relationships in a single calculation layer. It is currently used by almost all large models in the AI ​​field.adopted and became the cornerstone of this large model.I want to briefly review today, that is, in the big model, what original contributions have Chinese scholars made?Due to time constraints, I just want to take a few minutes to briefly review the original contributions of my three students to deep learning in their respective positions over the past ten years.Let me first review some historical events.In 2010, three Turing Award winners Hinton jointly opened the door to deep learning that they had spent many years researching.The first doorbell that opened this door was the epoch-making breakthrough in deep learning speech recognition made by Xin Tian and Microsoft's Deng Li in 2011.Alex was designed today in 2012 with his students Alex Krzyvsky and Elia Suskever.net achieved another epoch-making breakthrough in this image recognition image net competition.This SARS is now the co-founder of OpenAI and his chief scientist. So what did we do from 2011 to 2013 when deep learning just started?This is the first student I want to talk about, Wang Xiaogang.

Xiaogang was the first in the junior class and the 00 class of the University of Science and Technology of China in his undergraduate degree. Then he studied for a doctorate in my laboratory during his master's degree. He worked privately at MIT for Professor Eric Grimson, the former president of MIT.After graduating from MIT in 2009, I returned to my laboratory at a Chinese university as a professor.From 2011 to 2013, CVPR and ICC were the two most important vision conferences. There were 29 articles on deep learning around the world, 14 of which were from our laboratory and 15 from other laboratories around the world.We have 18 pieces of work that apply deep learning to visual problems for the first time in the world, including face recognition, face detection, face reconstruction, object detection, human posture, image super-resolution, three-dimensional shape recognition and other computer visionThe core question.Android Ren once commented that you have subverted computer vision.On the doorstep of deep learning, we rang the doorbell 18 times.

The second job, Xiaogang's second job is that he is working on face recognition. For the first time before Google and before Facebook, the face recognition done by robots surpasses the DDPID series done by human eyes.The third task in 2015 was that Xiaogang led the team to win the first image net world championship for Chinese scholars.It's the x net project that Hinton did earlier and participated in that competition.In terms of image nice, Xiaogang’s opponent was Google.

The second student I want to talk about is He Kaiming. His undergraduate degree was from Tsinghua University. He was the top scorer in the Guangdong College Entrance Examination in 2003. His undergraduate degree was a Ph.D. in physics, which he obtained in my laboratory at the Chinese University of China.Kaiming published his first article when he was studying for his master's degree in my laboratory, and won the best paper in CW PR.This is the first best paper to come out of Asia in the 25-year history of Westway PR2, and this was in 2009.So I kept joking with Kai Ming, saying that if you reach the peak with this move, your academic career will only go downhill from now on.As a result, he went to work at Microsoft and Facebook and rose all the way up.The first thing I want to mention about its work is that Resnet was done at MSR. Before 2015, deep learning could only be trained to more than 20 games at most.Resnet introduces a direct connection channel at each layer of the network, thus solving the gradient transfer problem of deep networks. It won the best paper in 2016 and is the most cited round in the history of computer vision, with more than 170,000Second-rate.After that, you can effectively train more than 100% of deep neural networks, and you can make the network very, very deep.In the era of large models, large models with Transformer as the core, including the GPT series, also generally adopt the Resnet structure to support the stacking of hundreds of layers of Transformers.I think to sum it up, Kaiming made the network deeper, and Google made the entrance to the network wider. Only then did it become the big model it is today.

The second work of NPS, Musk Asin, was done at Facebook. He proposed a truly high-performance object detection algorithm framework for the first time and won the ICTV best paper in 2017.Kaiming should be the only person in the world who has won the CVPRCV best paper as the first author three times in less than ten years after graduation.Mask's auto encoder was developed at Facebook. For the first time, the mask-based auto-encoding idea was used in unsupervised learning in the visual field. It opened the door to self-supervised learning in the field of computer vision and was extended to the field of three-dimensional audio and even AFO science.This morning I received an email from Hai Ming. It will have a more shocking action, which you will see in the future.The third student I want to talk about is Lin Dahua. He has an undergraduate degree from the University of Science and Technology of China, a master's degree from the Chinese University of Hong Kong, and a PhD from the Chinese University of Hong Kong. He is also a graduate student at MIT.error grim set, he won the NIPS Best Student Paper in 2010 when he was a doctoral student at MIT, which is the highest award at the top machine learning conference.In 2014, he returned to my laboratory as a professor. His first work, open mm lab, was in 2018. Starting from a small team, without investment in promotion, it became the most popular project in the world through word of mouth.Influential visual algorithm open source system.It has accumulated more than 80,000 stars on Git Hub. Currently, its users are in more than 140 countries and regions around the world, and 60% of its users are from overseas.Its new standard actually surpasses Pytorch.

The second domestic project, Shusheng Portuguese is the first multi-lingual large language model with hundreds of billions of parameters and a context length of 8K to be officially released in China. The specific evaluation details will be officially released at the Science Frontier Forum this afternoon, which I will not discuss here.Spoilers ahead.The third project, Scholar Tianji Landmark, was first proposed by Dahua’s team in December 2021.In December 2021, Dahua's team was the first to propose CT NERF technology.Earlier than Google, it expanded NERF technology from object level to city level. This Landmark is the world's first city-level nerve real-life scene. It has three large models with 200 billion parameters and covers 100 square kilometers. It will be officially released this afternoon.I’ll give you some spoilers here, please watch the video of this screen.

This is Dahua’s third-line work, so this afternoon, the large-scale scholar model system led by Dahua and Qiao Yu, including language, vision, multi-modality and three-dimensional, will be officially released at the Science Frontier Forum.I want to look back at this, why don't I move?Looking back, Xiaogang sowed many original seeds in the first few years of the rise of deep learning. Kaiming laid the foundation of deep learning very firmly and deeply.Dahua has made it flourish through open source and large models. What makes me very happy is that this big tree has begun to bear fruit.Just two weeks ago, our large model of autonomous driving stood out from 9155 articles and won the CVPR 2023 Best Paper Award.According to Google Scholar statistics, this is the first article in the top three international computer and vision conferences to be completed by Chinese scholars in more than 40 years of reform and opening up.

He is a doctor brought out by Xiaogang, and open mm lab was created by Chen Kai, a doctor brought out by Dahua.Lu Chaochao, the author of another facial recognition article in our laboratory that surpassed the human face and surpassed the human eye, also graduated from Cambridge University with a PhD and returned to Shanghai. He was working with Mr. Yao, the only Turing Award winner in China.Cooperating with the Knight Research Institute in Shanghai to engage in basic theoretical research on AI, a new generation of students has successfully started in Shanghai.

Thank you. Finally, I would like to say something to Kai Xiaogang, Kai Mingdahua, Chen Kai, Yang Yang Chaochao and all the students and teachers I have worked with, as well as my friends here in Shanghai today. I met you in the best time., but because I met you, I had this best time.This sentence is very beautiful, poetic, and comes from my heart. Unfortunately, it is not original to me.This sentence is the original creation of Mr. Yu Qian. It is his classic line in the movie Teacher.Just before going to bed every night, I was listening to the cross talk of this teacher Yu Qian, and I was thinking, how could a machine surpass such an interesting soul? I didn’t believe it.Thank you everyone, let’s work hard together, come on Shanghai.Thanks.

Thank you Professor Tang Xiaoou, thank you, thank you for your recognition of Shanghai, let’s work hard together, thank you again, we especially look forward to Professor Tang and his team being able to continue to deliver good results in the fields of originality and leadership.Dear friends, the new round of innovation in the field of artificial intelligence is driven by a holistic system model innovation, computing power, mutual promotion and mutual enablement.Next, we will applaud Huawei’s rotating chairman Mr. Hu Houkun for his keynote speech on the new era of win-win artificial intelligence.Applause is invited.

Dear Jilin Secretary, Mayor Zhengyi, distinguished leaders, expert guests.Good morning everyone, I am very honored to be invited to the World Artificial Intelligence Conference on behalf of Huawei again.I think the biggest difference between this year's conference and previous ones is that we are in a new context. Everyone knows that the emergence of ChatGPT at the end of last year pushed artificial intelligence to a new level.On the topic, it can be said that artificial intelligence, especially general artificial intelligence, has become one of the hottest topics in our human society during this period.

One of the hottest topics.So yesterday in the afternoon, as usual, I went to the exhibition hall and took a look around to see what everyone was busy with.I also saw the exhibition at this conference. In the exhibition hall, everyone focused on the research of large models on the one hand, and the application of large models in different industries on the other. Everyone displayed a lot of results.Well, I think all this shows that the development of artificial intelligence is bringing us this new expectation.

The direction of the future development of artificial intelligence is actually very clear. We all firmly believe that in a not too long time, artificial intelligence, especially general artificial intelligence, will help us rewrite everything around us. Then I think when the direction is clearFrom now on, the most important thing is the design of the path. How do we walk there?So today I also want to take the opportunity of this conference to report to you.Huawei is currently standing at this time. We hope to receive your criticism and correction on some of our thoughts on the future development of artificial intelligence.

To put it simply, Huawei's core consideration now is that in the next stage we must make every effort to promote artificial intelligence in depth and reality.So-called going deeper and more practical, we have two key measures, or two starting points. On the one hand, we need to take root in computing power and create a strong computing power base to support the development of China's artificial intelligence industry.Then the other aspect is to combine research and innovation with large models, from general large models to industry large models, to truly enable artificial intelligence to serve thousands of industries and scientific research.So we propose AI for industry and AI for science.Then I will report to you in detail below.

First of all, regarding computing power, everyone knows that in the development of artificial intelligence, computing power is definitely the basic foundation. In the current situation in China, we are in the availability of computing power.There are many challenges in terms of availability and cost.Over the years, Huawei has been deeply involved in computing power. We have focused on Kunpeng and Shengteng. We have achieved breakthroughs in their basic numbers.So what we are doing now is to support the construction of China's computing power base through architectural innovation, ecological development and flexible co-construction methods. We hope that through joint efforts with everyone, computing power will no longer become a factor in the development of artificial intelligence.bottleneck.

First, we improve computing efficiency through architectural innovation. For example, at the computing node level, we launched a revolutionary peer-to-peer purchasing architecture. We use this architecture to break through the traditional CPU-centered heterogeneous computing. It hasThe possible performance bottlenecks increase the bandwidth of the entire calculation and reduce the delay, so that the performance of the node can be improved by 30“.

On the other hand, at the data center level, we launched the Shengteng AI computing cluster in 2019. Through the cluster approach, we combine the comprehensive advantages of computing, storage, network, energy management, etc.When it is gathered together, it is equivalent to designing and managing the AI ​​data center as a supercomputer, so that its performance can be greatly improved.

Currently, the largest AI computing cluster we are building in the country is the second phase of Pengcheng Cloud Brain in Shenzhen. The current computing power is 1,000 p. We are currently planning to have ours by 2024.By the third phase, its scale will reach a level of 16,000 P.At the same time, we also deployed several thousand cards from Huawei's own computing center in Ulanqab. Our actual measurement found that through this cluster method, with the same computing power, we can get 10Efficiency increased by more than ”.So next I would like to report to you about the ecology, because the ecology of the development of the computing power industry is a key means, and it is also often a bottleneck that is difficult to overcome.

Four years ago, Huawei focused on the development of the entire computing industry and proposed a strategy of open hardware and open source software to enable partners to develop talents.So over the past four years, we have achieved certain results through cooperation with all partners.For example, we insist on further opening up in terms of hardware. This year we have launched more diversified modules and partners. Currently, we have launched hundreds of artificial intelligence hardware based on Ascend AI, which can meet the needs of this scenario in different industries.need.At the same time, in terms of software, we also insist on strengthening basic software through open source, especially when it comes to innovation of current large models.Professor Xiao Ou just talked about some of our achievements in large-scale model innovation. We also provide a full-process enabling platform to better support scientific research institutions and corporate customers with resources.At present, our ecosystem has incubated more than 20 basic large models, and has also adapted to more than ten large models that are mainstream in the industry.We have made statistics and found that half of the current large models in China are supported by AI Shengteng’s computing power, so here I would also like to express my gratitude to all of these companies and institutions that use this AI Shengteng’s computing power.Everyone's trust in us makes us more confident in further developing our computing power ecosystem in the future.Of course, we also know that although we have achieved a lot in the past four years, there is no shortcut for ecological development. We must move forward step by step. We also hope that more companies will join us in the future.Come, let’s make this ecology better together.So when developing computing power, we have another consideration, that is, our body must be flexible, and the methods and models must be diversified. This is the result of our thinking based on the current situation in China.So based on China's actual situation, we use a variety of models to build computing power.For example, in terms of computing infrastructure construction in this city, we currently support local governments in building 25 city-level artificial intelligence computing centers, including Shanghai.

On the other hand, we have seen that there are quite a few large companies and leading companies that all have demands to build their own artificial intelligence computing centers, so we will actively cooperate and help these companies build their own artificial intelligence computing centers..For example, China Mobile, iFlytek, China Southern Power Grid, etc. are currently undergoing construction, and we are actively participating in this process.At the same time, we also see that this artificial intelligence is in extremely strong demand for small and medium-sized enterprises.

So for these small and medium-sized enterprises in the development of artificial intelligence, when they have computing power needs, we will use cloud services to provide computing power services on Huawei Cloud. In this way, we can combine multiple methods toTogether, we hope that in this way we can truly realize what I just reported to you. We hope that eventually computing power will not become a bottleneck in the development of our artificial intelligence. So while Huawei is deeply cultivating computing power, we still need to do a good jobIt is to truly allow artificial intelligence to enter all online industries and serve scientific research.

In this regard, we believe that on the one hand we need to continue to improve the capabilities of general large models, but on the other hand we need to build industry models on this basis.So here I want to show you an example. You can look at the big screen, and we can see how to combine this general large model with the industry large model.For example, I have a question. I live in Futian District, and there is a 78-year-old man at home. Can the government provide me with subsidies?Then let's see if we use a general large model, what kind of answer will it give us?

Okay, let's take a look at the same time. If we use a large model of this industry, what kind of answer can it give us.This industry large model is a large government model we built in Shenzhen to help Shenzhen Futian District. So obviously we have this industry large model, which is actually based on the general large model and can provide more accurate,More valuable, this is what we should strive for.Well, Huawei has launched a new three-layer large model structure. The bottom layer is the benchmark universal large model. We call it the basic large model. This layer we vividly call it reading.There are thousands of books, which means you need to learn a lot of basic knowledge.So on top of this layer, we also created this industry model and scenario model, which we call Traveling Ten Thousand Miles. There are still many challenges that need to be overcome from reading case files to traveling thousands of miles, so this is very critical.The key point is to fully match and integrate the knowledge from all walks of life with large models. In this regard, Huawei is working hard with our partners in various industries.

Huawei's current Pangu model has supported the implementation of AI applications in more than 400 business scenarios around more than ten industries including finance, manufacturing, government affairs, electric power, coal mining, medical care, and railways.So in the future, we also hope to join hands with more people in the industry to further expand this aspect and go deeper, that is, to really get distracted and do things.

So in addition to making this artificial intelligence serve thousands of industries, we believe that we also have a very important task at the moment, which is to make artificial intelligence serve scientific research. We proposed AI for science, and we found that AI can learn massive amounts of historical dataScientific knowledge, while encoding mathematical equations into our large model, can later promote its integration with basic disciplines such as molecular dynamics, fluid mechanics, heat transfer, biology, etc., to help us discover morescientific laws.

Well, Huawei’s current Pangu model, we have released the Pangu scientific computing model. Under this scientific computing model, we include a large model of drug molecules, a large model of Pangu weather, and a large model of ocean waves. WeWorking with scientists has achieved good results.For example, as shown on this big screen, in the field of meteorological research, our current Pangea model can complete the prediction of the future global meteorological conditions for more than an hour, an hour to seven days, in a few seconds.This forecast is fast and accurate, and there are several keys to fully studying it. It requires fully studying more than 40 years of meteorological data, and at the same time, it must be combined with the knowledge of the meteorological industry to continuously conduct training and correction.

We are constantly training and correcting the same work in multiple fields of scientific research. We hope that through our efforts, we can truly focus on AI for science to bring more new ideas and methods to scientists and scientific workers.and new tools, and also input new momentum into the development of our entire human society.Finally, I also want to make this, a small advertisement, because tomorrow we at Huawei will hold our global developer conference. Tomorrow at the conference, we will explain Huawei's Pangu Model 3.0 to everyone in detail., how we serve thousands of industries and scientific research, I also hope to get everyone's attention.Dear friends, we are very fortunate to have witnessed several waves of technological revolution, from the Internet to mobility, to cloud computing to artificial intelligence. Each round of change has had a profound impact on society.There is no doubt that today, general artificial intelligence is giving us endless space for imagination and is leading us into the next golden decade.We hope to join hands with everyone to innovate together so that artificial intelligence can better serve thousands of industries and scientific research.AI for industry, AI for science, let us work together to win the new era of artificial intelligence.Thank you all.

Okay, thank you for the applause, thank you Mr. Hu Houkun for sharing your views from Huawei, and thank you all again.The explosive development of generative artificial intelligence has brought unprecedented opportunities, but at the same time, it has also made AI governance an urgent global issue.Next, Turing Award winner Mr. Yang Likun will bring his thoughts on AI and AI governance.Please share his wonderful video conversation with Mr. Yu Kai, founder and CEO of Horizon.Please look at the big screen.So in.

Your view.

What's the most exciting.

In your opinion, what is the most exciting development in artificial intelligence in the past ten years?Overall, autonomous surveillance operations have really created a revolution.Natural language processing, of course, I think language models, speech recognition, things like translation, anything that has to do with symbols or essentially sequences of symbols.All we have to do now is drive the same revolution in images and video, and it's already starting to take effect.Of course, there has also been great progress in the Transformer architecture. Now we are equivalent to arranged Transformers.Not to translation and not to alignment, right?I would say panning etc.In other words, if you pan, the input and output will also be panned.And then for the Transformer, if you reorder the input and re-token the output, the token is either allowed or not, and vice versa, which is another kind of etc.If you combine the two, you can do a lot, anything, almost everything.Yes, I have noticed recently in the public media that you have also joined in this debate about ChatGPT or university language models, rather than the end of AAGI.

i know you are.

I know you said no, so the question I have to ask is, is this technology the right path to AGI?What is the basis for your actual judgment?What are your criteria for judging?One criterion you can use is what is the function of the system?They missed something.If they're missing something very basic, that means the system you're looking at is inadequate.The system may be useful, you may want to do something with it, and it may be loaded with applications, but it will not be on the path to human-level intelligence.

I don't like the term AGI, because every intelligence is specialized.Human intelligence is also very specialized.So the first one they just trained with text, right?You also can’t train them using images or videos because they are user-generated architecture.They try to predict missing words in the text.In fact, autoregressive models like ChatGPT only predict the last word in a long article, right?This is what they are trying to do.But the problem with this is that you can't accurately predict the missing word in the text.What?You have to predict the distribution of all words in your dictionary.In the middle of a typical large language model.The maximum number of tokens for LM is around tens of thousands, so no problem.If you want to apply the same idea, the real architecture predicts what's going to happen next in the middle of the video, right?This way you no longer have tokens, but instead have video frames, and you train them to predict the next frame of the video.

So first of all, this question is very simple, because the next frame of the video will be similar to the previous frame, right?The difference will be very small, and you won’t learn much by studying these things systematically. This is the first point.The second point is that we don’t know how to represent the distribution over all possible sets of video frames. We can’t do it. We don’t have a sub-set of blurred video frames, right?So it's impossible to predict all the details that may appear in the middle of the video.

A lot of things are completely unpredictable, so what you have to do is use different non, what are you going to do?With different non-generative architectures, instead of trying to reconstruct everything, they try to predict a presentation that eliminates a lot of detail.This is the idea behind AJPA, which combines and builds a prediction architecture, instead of making predictions in the input space between pixels, making predictions in the representation space.Then the question is how do you train it?That's another question, how do you train a system to learn predictions in that space without collapsing the underlying representation, which is the first thing an LMM can do, right?First of all, LAM cannot be trained with images, unless you cheat, right?Otherwise, the more important thing about them is that they are not satisfied with one goal.You can't set a goal for them, you can only design prompts and then pray that the statistics of the data you train are of high quality.

The system can generate the correct answer based on this, but you basically have no control over it. You can't specify a goal and have the system check whether it has been achieved, right?So what is related to subsystems?Current LM or returning LM?They generate tokens one after another and cannot plan their answers in advance, so they don't understand the full picture because they are trained only on text, not on video.If they don’t understand the physical world, they can’t plan, they can’t reason, and they can’t achieve their goals.This means that to realize intelligent systems, we are still missing very important components.We may have LMs who can pass the bar exam for the final exam.It is not a very complicated question. It tests the ability of information retrieval.But we don’t have a smart enough system. Even if we want a robot that can handle these things, I don’t have a system that cleans the table and puts the dishes in the dishwasher, so you need a perspective model, in the middle of the artificial intelligence system,This model is mainly trained through observation, and a small amount of interaction can handle uncertainty. The world model is just like the way humans understand the world, understanding everything about the world through physics, intuition, etc.

What I just mentioned here, I think will be a huge challenge for artificial intelligence research in the next 10 years.If AI is allowed to form consciousness and intelligence like humans, how can we meet the requirements of ethics and governance?After all, humans have spent a long time establishing their current moral norms and incentives.In this case, what is the mechanism of human intervention in AI?I think this is also an important issue right now. In your opinion, what should we do?I want to make headphones.I think if you think that the best way to achieve human-level artificial intelligence is to make larger autoregressive l l LLM models and then use multi-model data for testing, you may think that these artificial intelligence systems are unsafe, butI actually don't think these systems can be very intelligent.The way to make them smart is also the way to make them controllable.That's the goal.The idea that drives artificial intelligence is essentially to give them goals that they must meet.Yes, some of these goals are goals defined by tasks, such as did you answer the question?Did you drive your car?Have you cleared the table?Other goals are safety guardrails, such as not hurting others.

Well, it's kind of like human fear or pain, if you violate one of the goals, it goes against your nature and you can't do it, right?This is also similar, these systems will not penetrate bit by bit to deceive or dominate human behavior.We can set goals that force the AI ​​to be honest, for example, to make it subservient to humans.Be wary of targets that make them curious or accept resources they shouldn't have.So for things like that, I think the system will be completely controllable and maneuverable.It is not easy to design a system to be safe, but the goal of designing is to make the system safe. This is a very difficult engineering challenge. We may not necessarily get it right from the beginning. We can start with a system as smart as a mouse.Bar?We target it, turn it into a good mouse, then make it as smart as a cat, a dog, or a chimpanzee, and then gradually improve.As we work, we will also address some of the target issues to make these systems work properly.We put it into a sandbox and simulation environment for testing. We want to ensure that they are safe. This is a very important issue.

You know there are people who, because they're afraid of the consequences of AI, basically want AI to be heavily regulated.Because they say that if anyone can master artificial intelligence, because he can do whatever he wants with artificial intelligence, the situation will be dangerous, so they must be strictly regulated.I totally disagree with this statement.On the contrary, I believe that the only way to make an AI platform safe, good, and useful in the long term is to make it open source.

Imagine that in the future, each of us will interact with the digital world through an artificial intelligence assistant. For example, in 10 or 15 years, we only need to exchange with our artificial intelligence assistant system, and all our information will pass through this artificial intelligence assistant system..It's not a good thing if technology is only controlled by a few controlling companies.Future artificial intelligence systems should become the protector of all human knowledge, and the way they are trained must be based on many sources. In this way, we hope to see more open source LLM and more open source AI systems.

Okay, applause, thank you again for the wonderful perspectives you two brought to us during this conversation, thank you.It shows us how to find new ways to solve innovative problems and use innovative methods to solve them.Hello leaders, distinguished guests, ladies and gentlemen, next we will present the Outstanding Artificial Intelligence Leader SEO Award on site. This is the highest honor of the World Artificial Intelligence Conference. The previous winners of this award have outstanding innovation achievements.Leading the development trend of global science and technology and industry.Now let us witness the birth of this year’s Outstanding Artificial Intelligence Leader Cell Award.Please look at the big screen.

The 2023 Artificial Intelligence Leader Award was awarded to the Commercial Aircraft Corporation of China Co., Ltd. Shanghai Aircraft Design and Research Institute's three-dimensional supercritical wing fluid simulation heavy, Dongfang Yifeng Shangfeiyuan aerodynamic design team's needs for complex flow simulation scenarios of aircraft three-dimensional wings, based on Shengteng AI and Shengmindspore AI frameworks, the industrial-grade fluid 3D simulation model Dongfang Yifeng was created, which can improve simulation efficiency while ensuring simulation accuracy, greatly shortening the research and development cycle, inserting smart wings into large aircraft, and using scientific intelligenceEmpowering a leapfrog upgrade in the pneumatic field.

Huawei Cloud Computing Technology Co., Ltd. Huawei Cloud Tianchi AI solver solves various types of role optimization problems and can meet various decision-making optimization needs on the cloud.Solver-based decision-making optimization tool methods can bring new operating methods to enterprises, reduce dependence on people in decision-making mechanisms, and especially bring qualitative improvements in supply chain and manufacturing management, thereby significantly increasing the revenue growth of enterprises.speed, increasing the space for corporate growth.

The artificial intelligence engine of Qualcomm Wireless Communications Technologies China Co., Ltd.’s second-generation Snapdragon 8 mobile platform.The latest generation Qualcomm AI engine equipped on Qualcomm's second-generation Snapdragon 8 mobile platform provides excellent hardware acceleration and software solution capabilities. Combined with Qualcomm AI software station and AI studio, it can provide full-stack AI optimization.It is the first in the industry to support generative AI use cases on the terminal side, and collaborates with the cloud to create a hybrid AI architecture suitable for the era of large models.

Jingtai Pharmaceutical Technology Shanghai Co., Ltd. Jingtai Technology's intelligent automated drug research and development platform uses cutting-edge technologies and capabilities such as quantum physics, artificial intelligence, cloud computing and large-scale experimental robot clusters to make drug research and development smarter and make biological innovation within reach..It has served many well-known domestic and foreign pharmaceutical companies in more than 180 innovative drug pipelines and served more than 200 customers.

Zhang Yunwei, University of Cambridge, combines machine learning with anti-corrosion technology to predict the aging of lithium batteries.This paper innovatively combines machine learning algorithms with electrochemical impedance spectroscopy technology to develop a set of intelligent battery prediction methods that can accurately predict the health status and remaining service life of lithium batteries. This is an artificial intelligence application in batteries.important applications in diagnostics.This battery prediction method has practical application value in battery management systems for large electrical equipment such as electric vehicles and portable medical equipment.

Everyone, let us congratulate the above five award-winning projects with warm applause. Please invite the award-winning representatives to the stage.Applause welcomes everyone to the stage. Please, welcome everyone. Next, we would like to invite Mr. Chen Jie, Vice Minister of Education, and Mr. Shu Wei, full-time vice chairman and secretary of the Secretariat of the China Association for Science and Technology, to come to the stage to present awards to the winning projects.Hello, please, let us give another warm round of applause to congratulate the five projects we have on stage today, leading the development of global technology and industry with outstanding innovative results, and invite our award-winning guests to join them.Take a group photo as a souvenir.We also specially propose that everyone hold high the honor and trophy that belongs to you and congratulate you.Well, congratulations to everyone again, thank you, please take a seat, thank you, and look forward to continued efforts in the future.Distinguished guests, ladies and gentlemen, in recent years, Shanghai has made every effort to develop the cutting-edge technology innovation field of artificial intelligence. Basic research and original innovation have been continuously strengthened, some key core technologies have achieved breakthroughs, and many research projects have achieved significant results.China has accelerated its entry into the ranks of innovative countries.Then we will release the first show of a number of major innovations on site. They are the Scholar Universal Large Model System, Fudan Digital Twin Brain, Zhangjiang Super City, and Fourier Universal Humanoid Robot Gr One. Let us give you a round of applause.Please, Mr. Wu Qing, Executive Vice Mayor of Shanghai, came to the stage to launch the first release ceremony of the results.Mayor Wu, please applaud, okay, let us witness this glorious moment together.Mayor Wu, please be prepared for the 321 release. Thank you, Mayor Wu. Please take a seat. Thank you.

The Shanghai Artificial Intelligence Laboratory released the scholar's general large-scale model system, which includes three excellent-performance base models of multi-modal language and three-dimensional space, achieving seamless cross-modal integration and running through the five major links of data training, alignment, deployment, and evaluation, forThe entire chain of industry, academia and research is open source and open, providing inspiration and support for unlimited innovation, and leading the future with original technology.

what you do?The world's first digital twin brain with a size of 86 billion neurons and a structure and function similar to the human brain has mathematically solved the problem of exascale parameter inference for large models, and has broken through the million-level b per second in communication.Massive pulse Great Wall transmission congestion bottleneck.Successfully cognitively simulates affective assessment of human visual and auditory tasks.This work is of great significance for future applications in the fields of computing, neuroscience, brain-like intelligence and brain health.The shareholder Zhangjiang Science City uses the world's most advanced large-scale digital base technology to create a unique 4.1 square kilometer Zhangjiang future city.The digital twin spatio-temporal computing generative AI fully controls traffic flow, fully replicates urban street scenes, fully perceives industrial space, accelerates data algorithms, and integrates breakthroughs in computing power to carry comprehensive connections between upstream, midstream, and downstream innovative technologies.Zhangjiang Future City is a place of innovation and dreams.Physics Industry Intelligence makes its debut with the new universal humanoid robot Grone, setting off a new wave of AI embodied intelligence.With years of technological innovation and accumulation, GR one has impressive athletic abilities. It can walk quickly, avoid obstacles quickly, walk steadily with straight legs, cope with impact interference on uphill and downhill slopes, and complete actions in coordination with others. It has demonstrated movements comparable to humans.Performance.In the era of universal robots, Fourier Intelligence will continue to contribute Chinese wisdom and solutions to the world.

Okay, let us once again congratulate the above major innovative achievements with applause for their debut at the opening ceremony of the conference. Congratulations to them. Thank you, leaders, distinguished guests, ladies and gentlemen. Next, we will invite the former Executive Vice President of Microsoft.President and foreign academician of the National Academy of Engineering, Mr. Shen Xiangyang, and IEEE Chairman and CEO, Mr. Saif Ramen, came to the stage to talk about scientific and technological cooperation in the AI ​​era.Applause, please.

Professor Ramen, welcome to Shanghai, which is also a beautiful hometown of my father, so we are all here in Shanghai, and you are also the chairman and CEO of IEE.Well, according to my understanding, the IEE organization can actually be said to be the largest one, a professional organization in our engineering field.Can you tell us about the IEE organization?Okay, thank you very much for having me here today.IEE actually has more than 430,000 members around the world, and China has the third largest number of members. So we not only have our members, but we also develop relevant standards, we also publish papers, and we alsoWe will organize some conferences, which are very important, because IEE actually holds more than 2,000 conferences every year around the world, and 200 of these more than 2,000 conferences are held in China.Therefore, we have also made such a commitment to China and Chinese scientists. It is great to increase the relevance of our organization.

We are now in China. Can you briefly talk about what IEE has done for us?At the same time, as a leader, what new measures have you taken in our i e?In fact, I think that if we go to see our organization mainly, just like us, just like the AGI that Elon Musk mentioned earlier, we also have many other open source and artificial intelligenceAll of this work is done by us, and we are also open source. We hope to ensure that all the development of general artificial intelligence, including the development of artificial intelligence as a whole, can be open. How to achieve this?How to guarantee?We have to make sure that all these people, if they are conducting research in these fields, they have a place to discuss their openness, their inventions if they can be discussed openly, if they feel it is tooIf it is dangerous, then there needs to be a certain degree of defense, which is openness.At the same time, we also need to use open source. We want its foundation to be open source.

In addition, another point is that artificial intelligence also requires manpower and computing power, right?We have also heard it before, so most of us cannot do it in a secret state. We cannot suddenly say that I suddenly came up with something, and then you have to supervise it, so we provide a platform and a foundation., allowing scientists, engineers and developers to come together to discuss the work they are doing, hoping that this will ensure that we can control it.

Okay, thank you very much Professor Raman. I agree with what you said before. We are very interested to see that the two previous guests, Elo Max and many others have also mentioned that this is very scary and needs to be discussed.supervision.But some people ask, is there anything that needs supervision?I think this is a different point of view, a different angle. From a scientist's point of view, we need to do open research. I am from the industry, and we develop AI products.

I very much agree with what you said earlier. We still need to have certain supervision and certain guarantees, because artificial intelligence is becoming more and more powerful. In order to achieve this, we also recognize the power of society. We mustWork together.For example, Microsoft, Google, and other companies have also established our partners. They are organizations like Bing and AI, and their influence is also very large.

In addition, I think organizations like IEE and the organization you represent also have great power because you are in a neutral position and you can influence relevant companies and countries, right??Yes, I think you are very right.In fact, as the chairman of IEEE, we focus on openness and globalization, so I have emphasized many times that good science is beneficial to science everywhere, no matter where it occurs, and I encourage this.Whether it is the Middle East, South Africa, China, North Africa, or North America, no matter where we are, we all need to discuss it together.I also encourage open science, and I also encourage everyone to publish your results in our conferences and journals to increase visibility.

Our meeting today is a good example. I also hope to see all these results, not only in Shanghai, but also visible around the world. Our knowledge and our ideas can become moreIn fact, this is indeed the case.In addition, I have always respected the IEE very much, because this organization has actually helped many professionals. Electrical engineers, electronic engineers, and computer scientists are called scientists. So don’t you guys go from basic members to advanced members to researchers?Personnel, when it comes to scientific development, I think this IEE researcher is actually very respected everywhere, not just in China. I am also very proud. I have 7 students who are IEE researchers.personnel.

Then I know that you are also very concerned about further promoting the career development of these professionals, especially expanding the number of members of China IEE, and at the same time increasing the number of Chinese professionals who can join an organization like IEE.Share with us some of the work your organization is currently doing in this area?Before I answer this question, I also want to point out a number like this, which is IEEE. In fact, the papers we publish every year have about 300,000 authors from China, so there are actually many of these Chineseof authors have published in our journals.

We have established a community. We established such a community in China a few years ago to encourage more engineers to become our members and senior members.In fact, I became a researcher many years ago. I hope to see more senior members from China become our researchers, including those from Hong Kong and Taiwan. This is also my encouragement to everyone.To increase visibility, I will find recommenders for everyone to become our researchers, so that they can be recognized globally.So we are committed to openness and communication and collaboration.

Another point, in fact, I also respect the IEE very much. As you just said, you organize many conferences and publish and sponsor many of these journals. I was also very surprised when I heard that you sawThe numbers mentioned are more than several thousand, so this is the first time I published my paper in IEE, which is about our machine intelligence.

I feel very excited. The world is also changing. We are also turning to a digital world and digital libraries. Now we are also participating in the World Artificial Intelligence Conference. We are talking about AI. In the era of AI, in this era of ChatGPT technology, what will you do?For example, will you make any changes to your journals and conferences?Because I think it’s very good to be able to find some new forms to help people understand more about these AI-related knowledge.

In fact, we have learned a lot from the epidemic. We cannot travel a lot during the epidemic.Now we have some meetings in a hybrid format, many of which are online, and we will also have a holographic meeting, which feels very realistic.We use this holographic image to feel that you are actually participating in the meeting, but it can also be done remotely to ensure that you can attend the meeting in London and New York even though you are here. This is also our mission, to make IEE aIt is a bigger beacon of innovation and inclusiveness. Anyone, for example, a child in Nanjing can communicate with me in Washington without any barriers.

Great, even if you can't come to Shanghai, if you don't come to Shanghai, you won't be able to taste Shanghai's local food.Let me ask you one last question. I know you have been to China many times. The last time you visited China, you should have had the title of a professor in VI universities in China.So share with us your experience in China.Yes, I usually come to China and I have been to many universities.What I find very interesting is that young people are very curious and ask a lot of questions.Why?Because one day they hope to become well-known scientists themselves, and they want to make their own contribution.Therefore, I think it is most important to set high goals when asking questions out of curiosity, and to work hard to ensure that your goals can be achieved.

Okay, with your encouragement and your leadership, I believe that more and more Chinese professionals will join IEE. At the same time, we can double the number of IEE members to more than1 million.Then I also mentioned here that it is also a good opportunity for our Chinese universities and Chinese industry to cooperate more closely with IEE.Thank you again for coming to our World Artificial Intelligence Conference, thank you Professor Raman, thank you.

Okay, okay, thank you, thank you very much MR safe Ramen, thank you.Thank you, thank you Mr. Shen Xiangyang, thank you Saif, Mr. Rahman, thank you both. Here we would like to especially thank Mr. Shen Xiangyang. At the opening ceremony of the conference every year, he will bring a friend to the stage to be with you.Get up and share their conversation.Thanks again to AIGC for not only reshaping the industrial landscape in the field of artificial intelligence, but also for creating a new model and path for AI empowerment.Next, we would like to invite Mr. Hou Yang, Senior Vice President of Microsoft and Chairman and CEO of Microsoft Greater China, to bring us his wonderful sharing and applause.Please.

Dear Secretary Chen, Mayor Gong, leaders, distinguished guests, good morning, everyone. I am Hou Yang. I am very happy to represent Microsoft at the World Artificial Intelligence Conference again.The theme of this year's World Artificial Intelligence Conference is "Intelligent World Generates the Future." I would also like to take this opportunity to share with you some of Microsoft's experiences and thoughts gained from expanding the ecological innovation of generative artificial intelligence and accelerating the advancement of industrial intelligence.Artificial intelligence as a scientific research project first appeared in 1956 and has been developed for nearly 70 years.With the overnight popularity of ChatGPT at the end of last year, large models and AIGC generative artificial intelligence seemed to explode in an instant. Even many practitioners in the technology industry were surprised by the sudden emergence of AIGC.In fact, from Microsoft's perspective, the so-called emergence is no accident. Countless outstanding scientific researchers, decades of Ruyi's research foundation, and the investment of massive computing resources have created such innovative results.Perhaps you all know that the breakthrough of OpenAI chat GPT relies on the infrastructure and computing power support provided by Microsoft Intelligent Cloud.Since 2019, Microsoft has launched in-depth cooperation with OpenAI, using massive cloud computing resources to support the research of OpenAI's large language model. The basis of the strategic cooperation between the two parties is that both our companies sincerely hope to create solutions that benefit every enterprise.and artificial intelligence technology for every consumer.

At the World Economic Forum in Davos in January this year, Microsoft CEO Mr. Nadler mentioned that the golden age of artificial intelligence has arrived, and Microsoft Intelligent Cloud has thus started a new model of accelerating development.We comprehensively integrate the latest AI intelligent technology with enterprise-level cloud services, thereby empowering enterprises to achieve true digital intelligence integration in the process of industrial digital transformation.Our h OpenAI Enterprise Edition service has launched five major models, including GPT 4 and enterprise-level ChatGPT, to support customers in creating customized intelligent services required by different industries. It also focuses on Microsoft 365 office Dynamics 365 business applications and Git Hub open source community development, digital information security protection, employee experience improvement, and the Windows operating system that each of us is familiar with, Microsoft has also launched a series of AI-driven Copilot intelligent co-pilot services. Users can use natural language to make demands, and Copilot can help users efficientlyComplete a series of tasks.For example, you can quickly write code, develop an application, or design a complete set of PPT presentations with pictures and texts based on the outline of the speech.

So the scenarios I just gave examples are no longer fictions out of thin air today. They are created by Microsoft, including Microsoft, and are no longer fictions out of thin air. This is a reality that many companies, including Microsoft, have already practiced.Work scenarios, such as the much-anticipated HOPTI enterprise-level service, which has been approved by more than 4,500 companies around the world in the past few months since it was launched, and is already being implemented in the production environment of manufacturing, retail, finance, services and other industries.of innovation.

Another example is Git Hub Copilot, which is used to assist programming. Since its launch one and a half years ago, more than 1 million developers around the world have used it. Nearly half of their codes have been completed with the help of Copilot, and the programming speed has increased by 50%.above.In these active attempts by global customers to use Microsoft's artificial intelligence services, we not only see the enthusiasm of enterprises to accelerate digital transformation and innovation, but also feel the market's urgent demand for a new generation of generative artificial intelligence.Therefore, Microsoft continues to increase investment and fully promote the development and popularization of generative artificial intelligence.At the Microsoft Global Developers Conference held in May this year, Microsoft continuously released more than 50 new technologies and services related to generative artificial intelligence development.The most important one is that we have opened the Copilot intelligent co-pilot and plugins plug-in expansion system to developers around the world, so that developers, partners and enterprise users around the world can seize the innovative opportunities brought by generative artificial intelligence and createDevelop an unprecedented new generation of smart applications.Among them, the plugins intelligent plugins plug-in expansion system adopts the same technical standards as OpenAI, which can build safe and reliable connections between third-party application customers, business scenarios and generative artificial intelligence.

By accessing real-time updated information streams and a variety of applications and services through plugins, you can add more professional computing capabilities to the AI ​​system, creating a richer variety, more convenient to use, and more accurate information.Conversation service.It is expected that by the time Microsoft 365 Copilot is officially released, we will provide more than 1,000 plugins for developers to choose from.

What is even more worth looking forward to is that developers and partners around the world will be able to independently use these plug-in interfaces and intelligent services to develop more rich and unique innovative applications for intelligent co-pilots.Well, there is no doubt that these are based on self-powered natural language conversations.The new generation of intelligent interactive applications will open up a more exciting digital world for us.Microsoft firmly believes that every company in the future will need to have the ability to control digital technology. We also see that as generative artificial intelligence continues to show great potential, every application in every company will be driven by artificial intelligence in the future.

Technological breakthroughs in artificial intelligence are also bringing once-in-a-lifetime innovation opportunities and challenges to all walks of life, which also prompts us to think about how to use it to enhance the creativity and competitiveness of enterprises.While achieving technological breakthroughs, Microsoft is thinking about how to transform research results into productivity.We hope to combine generative artificial intelligence with industry needs as soon as possible to accelerate industry upgrading and innovation.What I present here are six key innovative scenarios for the application of artificial intelligence in industries summarized based on the recent industrial intelligent solutions of global customers, including optimizing supply chain resilience in the manufacturing and energy industries and popularizing predictivemaintenance, and improve the intelligent driving experience.For another example, we can innovate the intelligent customer service of retail e-commerce, realize interactive search, and guide consumption trends.We can also build lifelike NPC characters in the game and generate unlimited plots and realistic dialogues.In the financial industry, we can obtain analysis reports on financial market conditions at any time, discover earlier, and manage potential financial transaction risks faster.At the same time, in the field of life science research, we can also improve the analysis capabilities of clinical experimental data, accelerate the research of drugs and vaccines, and achieve more accurate medical image recognition and diagnosis.At the same time, in the field of education, we can also bring students a more inspiring, interactive, customized and exploratory learning method that is not restricted by geography. We can help cultivate creative and creative people who are more suitable for future market needs.Talents with lifelong learning ability.As generative artificial intelligence continues to accelerate industrial integration, I believe that in the next few months, more colorful and imaginative application scenarios will continue to emerge.As artificial intelligence accelerates its development, people will inevitably have concerns about potential security risks.Microsoft has always advocated and strictly adhered to these six principles for building responsible artificial intelligence. We have also always strictly adhered to various requirements related to data privacy and security compliance.At the same time, we also actively advocate global technology companies to form an industrial consensus through exchanges and cooperation to ensure that the artificial intelligence technology we develop can responsibly benefit all mankind.Facing the huge innovation opportunities brought by the new round of technological change, Microsoft in China is willing to start from our own technology and advantageous resources, deeply cultivate China's local ecosystem, and continue to work with enterprise organizations and partners from all walks of life across the country.We will expand technical exchanges and business cooperation with the whole country to continuously explore the application potential of digital intelligence in various industries, truly promote intelligent innovation and digital transformation in various industries, and contribute our most active strength.The golden age of artificial intelligence has arrived, and Microsoft will continue to work hard in this golden age to fully tap the potential of technology, benefit everyone and every organization around the world, and achieve extraordinary results.Thank you all, thank you.

Thank you, thank you to Mr. Hou Yang for his wonderful sharing. We have also seen more of Mr. Hou Yang’s enthusiasm for creating the future. Thank you again for your side-by-side cooperation to inspire continuous breakthroughs in artificial intelligence and to jointly create a better future civilization.Next, we will invite several experts, scholars and entrepreneurs to conduct on-site discussions and dialogues from different fields and perspectives around the theme of creating ideas and discussing the possibility of qualitative change.Next, we will invite Yao Qizhi, winner of the Turing Award and director of the Shanghai Seven Systems Research Institute, Yuan Yang, assistant professor of the School of Cross-Information at Tsinghua University, Yang Zhiling, assistant professor of the Institute of Cross-Information at Tsinghua University, founder of Mengshot AI,Jack is the first author and Assistant Professor of School of Computer Science and Engineering at Nanyang Technological University Pan Xingang.We would like to invite Mr. Xu Li, Chairman and CEO of SenseTime Technology, to be the host of this session.Next, we hand over the stage and time to them. Come on, everyone, we welcome them with applause.

Well, dear guests, I am honored to host this event today, because this event is hosted by Academician Yao, the title of our computer industry, and our three very young academic stars. It can be said that these three represent our artificial intelligence.some new development directions.So without further ado, I will start with a brief introduction and look at the forum that will launch this segment of ours.

Well, first of all, Academician Yao, we know that Academician Yao is the winner of the Turing Award, and he founded the Quantum School of Cross-Information at Tsinghua University.And in fact, the development speed of large models is very fast now. I would like to ask Academician Yao whether there are any breakthroughs in the basic theories you have seen in the development of large models today, as well as breakthroughs in basic theories.What are the next development directions?These.

Very good question.Today we heard from Professor Tang Xiao'e. He has already talked about how our Chinese scientists have made many breakthrough contributions in the development of modern AI.Well, I would like to mention one here. We have a young teacher Gao Yang. He made a very important contribution to algorithm breakthroughs more than a year ago, and received a lot of international attention. So it is basicallyIt can speed up reinforcement learning, which is very mainstream now, by hundreds of times.

So let me explain, after this ChatGPT, the next very important goal in the future is to make intelligent robots, which have such multiple sensory capabilities as vision, hearing, etc., that can operate in a variety of different situations.In the environment, they can learn various new skills independently, but not now.General reinforcement learning methods are too slow, because they often take several months to learn these new technologies.Then Mr. Gao Yang’s breakthrough enabled reinforcement learning to be done in a few hours.So you can see that in the future development of these intelligent robots, this work must be included.

At the same time, this is not only a practical issue, but also has a theoretical contribution. In the past six or seven years, these thinkers at the highest level of artificial intelligence have had a line dispute, which means that we now putIs this route that relies on reinforcement learning correct?So there are a lot of debates. I think Professor Gao Yang made his breakthrough more than a year ago. I think that on the other side of the scale, we should stick to our current path. For general artificial intelligence, itPerfection still has a long way to go.It’s no wonder that OpenAI’s co-founder, Schumer, was in an interview not long ago. In fact, he regarded the work of Teacher Gao Yang as one of the most important highlights in reinforcement learning in recent years.I am just making an advertisement now. Tomorrow our Automotive Research Institute will host a sub-forum on embodied general intelligence, where you can meet Teacher Gao Yang. At the same time, there will be many other Chinese and foreign contributions to basic theories and frameworks..

Well, okay, thank you Yao Yuanshi, we also look forward to greater achievements and breakthroughs from Quanhua Intelligence in this field of embodied intelligence.Then our next three professors are very young, two are professors from the College of Education, and one is a professor from NTU.Then I remembered the proposal in 1956 when artificial intelligence was coming up. In fact, the average age of the four scholars was 33 years old. I took a look and the average age of our three professors was less than 33 years old.Next, we would like to ask Professor Yuanyang. In fact, Yuanyang also does large-scale models and has done a lot of research on intelligent medical care.So I want to say that in the evolution and development process of large models, do these interdisciplinary subjects actually have any further help in the development of models?

Thank you host.I think that big models are now especially emphasized in some cross-industry industries and can be put into practice.Everyone will talk about multi-modality, which is definitely very important, but I think everyone’s understanding of multi-modality may be rough. People often think of multi-modality as being able to see pictures, read text, and have a sense of touch., There is a feeling of temperature.But I feel that in order to truly achieve a specific industry and be able to solve the problems in the industry, it needs to be done in more detail.

For example, let me give you an example, well, like the Dragon that Teacher Pan may talk about later, if we only consider the generation of text to pictures, then you say a sentence to draw a dog, and then it generates a picture of the dog., but in this case you are likely to find that the dog picture you generated is not the posture or mode you want. Then you can drag it with the mouse and modify the way the mouse drags it.

In my opinion, this is a new mode. The user uses a better way to tell the big model what he wants to express so that he can understand it.Although it may just be a mouse drag, I think this kind of multi-modal input is very important in specific applications.When it comes to more specific industries, such as medical care, law, and education, I think we should not just give text or images to the model, let it have some professional data, and hope that it can solve professional problems.

We should dig deeper into this industry to find out what the core problems are, and then find out what kind of modal data we need in this problem, and what kind of information can accurately express the problem we want to solve.?I call this a modal completion, and we also need it after modal completion.Well, collect enough data based on the completed modalities and align the modalities.After modal completion and modal alignment are completed, I believe it can give large models more powerful capabilities to solve more core cross-domain problems.

Okay, thank you, Teacher Yuan.After modal completion, it can be equivalent to know how in this aspect, which will be of great help to subsequent development.Let’s ask this Yang Chiling. This Yang Chiling is also known as a genius boy. And I have seen a lot of his work, including his participation in some very early work on large language models.Well, I think large language models are now widely used, but they will also encounter some problems in practice, such as the hallucinations we often mention and a series of challenges.So I would like to ask what difficulties and challenges we have in the actual use of large language models?Or what specific aspects should be paid attention to.

point?Very good question. There are indeed many unresolved issues regarding this large model. For example, how can security be very controllable?And that includes preventing it from producing this kind of illusion, and not making up some non-existent things, including that it actually has no way now, such as creating new knowledge like scientists, or being like the top leaders in many industries.Sales or top-level product development jobs may not be possible right now.

Well, I think a very important point here is that when we think about these issues, it is not more about thinking about each issue in a single point.For example, if I want to solve the problem of hallucinations today, then he may not be trying to solve the headache, but may be more systematic in abstracting what are the common issues between these problems down to the bottom level, and returning to the more essentialTo solve it at such a level, because it is a universal model after all, we actually hope that it can be able to draw inferences from one instance in all these aspects.

So more fundamentally, I think we still need to do more regular, large-scale, and efficient compression.For example, use a better framework that is more suitable for distributed training, such as Moe, or support longer contexts, including how to better allocate computing power and achieve higher quality.Allocate more computing power to the data. I think problems like this can actually be solved more fundamentally.There are some limitations of current AI that we may have just talked about, thank you.

OK, thanks.It may not be easy to solve it from a theoretical level.And the background of our round table today is all generated by our algorithm. In fact, it is all based on the discussion model, but this new gang has single-handedly brought Gan back into everyone's field of vision. He is the Dragon.An article on the Internet talks about popular products, and then says that as long as you have the skills, everyone can create content.So I would also like to hear from Xin Gang, for example, the debate between Gan and diffusion models in the algorithm. Is there any good or bad in the route selection of the algorithm itself?In other words, when you look at the generated content later, we will know which ones can be more scalable?right.

Thank you host for your question.Gan and diffusion models are now the two main generative models for image generation. In particular, the diffusion model has clearly overtaken Qian in the near future.Then I think they have three main differences due to the different frameworks and optimization goals of the generated model.The first one is the trade off between performance and efficiency.Obviously, during the generation process of the diffusion model, it requires greater computing power, and the inference time and training time required for its iterative calculation are significantly higher than dry ones.At the same time, greater computational overhead also brings higher image generation performance.The image it generates will not be limited by Gan's mode CLAP problem. Its authenticity and diversity are significantly better than Gan, so I believe that the upper limit of the diffusion model must be higher than that of dry.If performance permits, its advantages in quality and diversity are very obvious, and its application value and application prospects are broader.But if in some specific situations, such as mobile devices, where performance or computing overhead is limited, then doing it is still a compromise option.

Then the second point is the difference between their Latin spaces.Then we know that Gan maps a Compact latent Vector to an image, but the diffusion model gradually denoises a noise map with the same resolution as the image and maps it into an image.In practice, the noise map of the diffusion model often appears to be relatively random and unstructured in its impact on image content.But Gan's Compact latent code is more of a set of low-dimensional chickens that embody this high-dimensional data. It can be very effective in editing high-level attributes in images, such as human expressions orAnimal gestures and the like.This is why we chose to use stem as the first generation model for the drag editing method to study.But I believe that how to expand to the diffusion model in the future is also a very important issue worth exploring.

The third point is the continuity of the space in which they generate images.Well, since the lip shes constrain of the two models are different when they are designed, the image space of the diffusion model is relatively discontinuous and deep, while the image space of Gan is very continuous and natural, so we use the diffusion model for Latin space editing or videoWhen editing, we often observe jumps. Then we do this kind of Dragon editing in the middle. It appears smoother and looks like animation, so this is also an advantage of doing it.So how to complement the respective advantages of these two models in the future will be a very interesting research question.

OK, thanks.That is, get has advantages in front-end or continuous video, and diffusion will actually have better performance.That hopes for a better combination.Then let me ask the last question to the four of you. It is based on our research direction. In which vertical field do you think large language models may be most promising? Which vertical direction are you most optimistic about?

I think the easiest thing is that we can think of a lot of paperwork now, so with this large model language, more work can be done by these machines, so I think this is a direct one.

It's a productivity tool.Paperwork.

Yuanyang, yes, I think it is medical care, because of course not only because I am doing intelligent medical care, but also because I think large models are now based on the paradigm of pre-training, and the essence of pre-training is that it actually learnsThe relationship between data, there are a lot of relationships in medical care, such as the relationship between the patient's symptoms, the relationship with the symptoms, the relationship with the medicine, and what will happen after taking the medicine.In fact, humans may not be able to learn and describe these relationships well. I think large models may be better than machines in this regard, so I am more optimistic about this direction.OK

Thanks.Zhiling.

I think a more important scenario is that AI should have a common memory like humans in the future.For example, if we use an AI today, we still need to re-instill some things into everyone every day and provide it with a lot of context.Well, I think that in the future, for example, if there are products like Rey recently, it can actually be used, for example, through screen recording, right?Then, all the things that humans can see can actually be seen by AI. In this way, I actually think that there is a lot of room for imagination in personal use.right.

OK, thank you.surname.

Gao, I do visual generation and visual content generation. So I think image generation is already very good after large models. Then video and three-dimensional content generation also have great prospects in the future. It can help designers and artists., helping animators and film and television special effects artists create higher content and higher quality content better and more efficiently.right.

Thank you, okay, thank you.Due to time constraints, our roundtable forum ends here. For more work, you can actually pay attention to the homepages of the teachers. We and the others actually update very quickly.OK, thank you all.

Okay, thank you, thank you again for your wonderful sharing, and thank you Xu Li and Ctrip for hosting this conversation for us. We also look forward to the future artificial intelligence will bring us in our daily social life.More changes will bring us more help.Thank you again for jointly building the AI ​​ecosystem, which is the foundation for the development of the artificial intelligence industry. When large models embrace thousands of industries, an era of data creation, data, and knowledge creating knowledge is about to begin.To this end, we will also initiate the establishment of a large model corpus data alliance on site today, so that the trickle of corpus data can become a surging driving force for the development of large models.Next, we would like to invite guests from the sponsoring units of the Large Model Corpus Data Alliance to take the stage.

Invited are Jiang Wenbo, member of the editorial board of the China Central Radio and Television Station, Wang Yanfeng, assistant director of the Shanghai Artificial Intelligence Laboratory, Zhao Zhiyun, secretary of the party committee and director of the China Institute of Scientific and Technological Information, Dai Kan, member of the standing committee of the party committee and deputy director of the National Meteorological Center, People's DailyLi Jun, full-time deputy director of the National Key Laboratory of Communication Content Cognition, Wu Jianxiong, chairman of Shanghai Data Group, Ding Bo, vice president of Shanghai Digital Industry Group, and Zhang Qi, president of Shanghai Digital Business Association, invite everyone.All guests, please move to the big screen. Okay, please put your left or right hand on our startup device. Let us count down to three and release 21 together on site. Thank you, thank you.Officially established, thank you all, please take your seats, thank you.

Develop general artificial intelligence.

Smart, thank you for creating the industry.

Innovate the ecology and draw on high-quality, multi-modal, and wide-area corpus data resources.To this end, eight units including the Shanghai Artificial Intelligence Laboratory and China Central Radio and Television Station jointly initiated the establishment of a large model corpus data alliance to connect tens of billions of data.The alliance will ensure scientific research public relations, adhere to sharing, promote safe development, cultivate industrial ecology, and allow data to fully serve general artificial intelligence.

Dear guests, at present, large models have become a universal enabling tool, triggering disruptive changes in the intelligent era, and becoming an important strategic supporting technology that promotes global economic growth, reshapes the industrial landscape, and consolidates national competitive advantages.Taking model capability evaluation and verification as the starting point, we build rigorous, scientific, and leading evaluation and verification indicators and platform services, which are conducive to promoting large models and scientific research innovation, objectively assessing the level and gaps of large model technology products, and thus promoting large-scale models.High-quality development of model technology, accelerating application innovation and industry implementation.So next, we will invite Wang Zhiqin, Vice President of China Academy of Information and Communications Technology, and Wang Ping, Deputy Secretary of the Party Committee of Shanghai Artificial Intelligence Laboratory, to take the stage to launch the establishment of a large model testing verification and collaborative innovation center.Two of you, please.

Okay, let us witness this special moment of the establishment of the Large Model Testing, Verification and Collaborative Innovation Center. Come on, we invite two guests to hold the putter in their hands and be ready to count down to 321 start with everyone. Applause.Congratulations on the establishment of the innovation center. Thank you both again. Please take a seat. Thank you, big model, leaders, ladies and gentlemen, the digital economy is a strategic choice for the country to seize the new opportunities of the new round of international revolution and industrial transformation. MobileThe information industry chain is a new paradigm of collaborative innovation in the digital economy era.

China Mobile has a great mind for the country and has the courage to shoulder the mission of being the leader of the mobile communications industry chain. It has innovatively planned a 26 billion industry chain development fund to implement the central enterprises' ability requirements to improve basic fixed chain technology to supplement the chain, integrate strong chains and optimize fast chains, and give full play toInvest and test the effect, create a verification pattern, invest and cultivate deep small and medium-sized enterprises, enhance the resilience and competitiveness of the industrial chain and supply chain, and play a supporting and leading role in the construction of the country's modern industrial system.

As an important carrier for the implementation of the strategic cooperation agreement between China Mobile and Shanghai Municipality, the Shanghai China Mobile Digital Transformation Industry Fund was jointly initiated and established by China Mobile, Chengtong Group and Pudong Leading District Fund of Funds under the guidance of the Shanghai Municipal Party Committee and Municipal Government.Reaching 10 billion yuan.At present, all preparation work has been basically completed.Next, we will invite Gao Tongqing, deputy general manager of China Mobile Communications Group, Hang Yingwei, deputy secretary of the Pudong New Area Party Committee and district chief, and Tong Laiming, deputy general manager of China Chengtong Holding Group, to take the stage.There are three welcome guests for the official launch of the startup fund.

Okay, next we will come to witness the on-site release of industrial funds.Okay, three guests, please press the start button with your right hand. Together with our friends present, we will count down to the release of 321. Congratulations again, thank you three for taking your seats. Thank you.

The Shanghai China Mobile Digital Transformation Industry Fund aims to give full play to China Mobile's long-term mission, cultivate momentum for Shanghai to build a digital economic development highland with global influence, and contribute to the national science and technology power strategy and the high-quality economic and social development of the Yangtze River Delta region.

Dear guests, China Mobile Shanghai Fund, as an important fund for Shanghai's urban digital transformation, will play a capital linkage role and assist the all-round digital transformation of Shanghai's economy, life, and governance.Win-win cooperation is the cornerstone of world common prosperity.In our previous conferences, we have witnessed wonderful moments of unity, cooperation and focus on development.So next, we sincerely invite everyone to witness the launch ceremony of the UNIDO Global Industrial and Manufacturing Artificial Intelligence Alliance and the UNIDO International Center of Excellence for the Development of Artificial Intelligence in Industry and Manufacturing.First of all, please share the video sent by Mr. Gerd Muller, Director General of the United Nations Industrial Development Organization. Please watch it on the big screen.

Okay, we are back to the studio where our China Business News broadcasts 3*24 hours 3*24 hours. Here is the 2023 World Artificial Intelligence Conference that we are live broadcasting for everyone.Back to our studio, this is it. During the entire process of live broadcasting and rebroadcasting for everyone, in fact, many friends online and offline are participating and watching, but there are also many friends who may not be able to respond in time at this time.After seeing some of the content of the entire opening ceremony of the conference so far, let me briefly introduce some highlights of today’s opening ceremony.

Okay, let me introduce to you the overall structure of this conference one by one. Remember, there is an opening ceremony and a closing ceremony. Of course, the opening ceremony is currently in progress, and then there is also a technological innovation and industrial development.There are two plenary meetings, one on technological innovation and industrial development, then two plenary meetings, another ten theme forums and n ecological forums.This is the overall architecture of it all.

Well, the highlight of this time you haven’t seen, but you have heard about today’s conference, which is the gathering of big names. We see that the current number is about 1,400 such guests, and there are four of them.With a Turing Award winner and more than 80 domestic and foreign academicians, this lineup is still very strong, so you should be there.Okay, so this year’s event, that is to say, there will be events at this conference every year. This year’s event is also very much anticipated and paid attention to, that is, the World Artificial Intelligence Innovation Competition, there is also a hackathon, etc.There are four major events in total, so there are more than 3,000 teams participating, so you can see that the competition is so fierce.Okay, this is a point, so what is the situation of this year’s exhibition?The exhibition area is 50,000 square meters, which is much larger than in previous years. As you can imagine, this area includes the large models that everyone is focusing on this time, especially focusing on. This is one, and thenIt's chips, robots, smart driving, etc. So what will be the number of exhibitors this year?Listen, 400, 400, there are more than 30 new products that will be released for the first time at the conference. In fact, if you pay attention to the news, you will see that some of them are already reporting on these aspects in the public news.Something.

Okay, what I'm saying is to give you a brief summary of the general situation this morning, so now we have this brief introduction, let's take a look at the situation on site, and let's hand it over to the site..

Okay, in this case, we have just briefly introduced these data to you. Let's go back to the studio. In this case, I will introduce to you who the big names at home and abroad are today, because we are onlineI also saw some people asking, among the Turing Award winners here, among the Turing Award winners are David Patterson, and Joseph Sfarsky. These may be unfamiliar to some of you, because you don’t know much about them.If you pay attention to this industry, there are also Manuel Bloom, Yao Qizhi and so on.Well, Nobel Prize winner Michael Levitt, as well as our domestic and foreign academicians, etc., come back to our scene.

Next, let us give a round of applause to Liu Duo, Vice Mayor of Shanghai, Zou Ciyong, Deputy Director General and Executive Director of the United Nations Industrial Development Organization, Yang Zhenbin, Secretary of the Party Committee of Shanghai Jiao Tong University, and Glo, Deputy Minister of Information and Communications Technology Development of the Ministry of Science and Technology of the Venezuelan People's Power.Leah Carvalho, China Telecom Group General Manager Shao Guanglu, Huawei Vice President of Corporate Communications Zhang Yuan, Tencent Cloud Vice President Gu Wei, Alibaba Cloud Intelligent Technology Research Center Director An Xiaopeng, German INC Innovation Center Artificial Intelligence Technology Director BennyDrescher took the stage together to preside over this glorious moment.Please, please extend your palm and please it on the screen, please.Okay, let us countdown to the release of 321 together with our friends at the scene. Thank you, thank you all.

thank you very much, thank you please be seated.

Thank you, thank you again for being here for us to officially launch.Well, leaders, ladies and gentlemen, our agenda for the opening ceremony of this morning’s conference is coming to an end. In addition to asking everyone to continue to pay attention to the agenda of our three-day conference, we will continue to pass 3* this yearThe 24-hour online live broadcast shows the live information of the conference at close range, extends the exciting content of each forum in multiple dimensions, and exclusively unlocks high-quality resources of the conference.Thank you again to all the leaders present, all guests, ladies and gentlemen, and viewers who are watching the live broadcast on the cloud. Thank you all. See you in the future. Thank you.

Space allows matter to exist, time allows everything to evolve, and wisdom allows everything to be connected.Shanghai, China, has gathered global innovative wisdom to build a highland for artificial intelligence. When the industrial genes are reconstructed, the core driver of high-quality economic development will be iterated again, helping to refresh the blueprint of life.Civilization has ushered in a new starting point and will evolve beyond imagination.When the neural network expands and the world is connected as one, future trends will be profoundly changed, and countless desires of human civilization are being generated here.Infinite attribution, infinite world in life.At the Artificial Intelligence Conference, there is only one Shennong who protects life. Together, we, the snail girl, only favor one person and protect life.We are a magpie bridge for everyone, connecting once a year and helping each other across the distance, anytime and anywhere.Ma Liang only has one magic pen to create all things.We activate countless pairs of arms to explore the laws of destiny and grasp the trend.We use data to gain insights.If human beings are stronger than we were thousands of years ago, it is because we use wisdom to connect each other closely and face challenges together.

The above is the complete content of the opening ceremony of the 2023 World Artificial Intelligence Conference (live transcript).If you want to learn more about the background, development, and character relationships of the plot and characters, you are more than welcome to pay attention to the line class.In the future, we will provide detailed and rich plot introduction, analysis and character analysis to allow the audience to more fully understand and experience the charm of the series.