251202 Dr. Kai-Fu Lee on How the Rise of AI in China Can Benefit the World

Thank you Dr. Dasher for inviting me to this very interesting session. My own background was as I entered Columbia College in 1979, I just got so excited by the prospects of AI and devoted my life to it. AI had its ups and downs and I came in and went out a few times but always interested, always believing that the moment would come. I ran AI at Apple and then for Microsoft Research Asia. And then we had some AI projects when I was at Google and I founded a VC firm, Sinovation Ventures, which currently primarily invests in AI, but we were the first to invest in AI. 2011 was our first AI investment. So it's been something kind of my commitment and the love of my life. I've been spending basically all of my professional time excited to see the progress. So I think two big things happened in the recent years. One was the advent of deep learning around 2012 or so and then the arrival of transformers which was in the last 7 or 8 years and of course seeing that enabling ChatGPT and the whole generative AI revolution in the past three years. These are the sort of three big things that made something I could only dream about become a reality. So that's my technology participation. In terms of US China, I've been an executive in Silicon Valley at Apple, SGI, Microsoft and Google. My jobs at Microsoft was for two years in China, five years in the US. My Google job was all in China. So it's been a cross between US and China running US multinational companies. And eventually at 2009 I decided to start my VC firm. When I saw that at 2005 I was able to hire so many super smart people in China. I think I built kind of a dream team in Google China and most of them left by 2007. So I was looking what happened? How did I lose all these people? Did I screw up? Is Google going downhill? And I didn't think so. But then what I saw was they were not leaving Google. They were entering startups. And that was the beginning of China's startup era. And then I saw well if that's the trend, if all these super smart people are hired, two-thirds have gone to startups either to start their company or to join a startup, it's probably time for me to follow their footsteps. So 2009 I left Google China and started a VC because I thought I was too old to do a startup. But then I changed my mind three years ago when I started 01.ai because this generative AI revolution is just so huge. I couldn't just stand on the sidelines investing. I needed to be part of the action. So that's kind of my history.

The top people we had. Yes. I think when I started at Microsoft and started the Microsoft research in 1998, the Chinese university research was rather backward, not even comparable to number 100 university in the US in terms of teaching quality and professor quality at that time, which is understandable because China only started modernizing a decade before that. However, the students were brilliant. So at Microsoft research we took super smart, incredibly hardworking students and retrained them and that I think paid dividends as Microsoft Research China became a top research lab in the world. But at Google it was a different story. Just between the years 1998 to 2005, many universities started hiring young faculty who were at US good school quality and many were returnees from the US PhD and so the faculty was getting better. Of course that's a process. The young faculty need to build their reputation, get their tenure. So that took time but around that time the research was getting much much better and the Chinese research tends to be very practical which was about building AI. So the students who come out are excellent researchers/engineers. These are not, even today the top breakthrough research is largely from the US. That's not what Chinese universities are known for. But they are known for working super hard on lots of experiments and very diligent coming up with interesting good ideas, maybe not breakthroughs but excellent papers. And also both in research and engineering, I think the Chinese PhD graduates were getting really quite good. So at Microsoft, frankly, I had to lower the hiring bar in order to let super smart, high potential, but not properly educated kids come in and let us retrain them. At Google, we just used the same bar worldwide and then they passed basically the Silicon Valley bar and were hired in without any difference or exception and they made exceptional contributions to the company.

Well, it's true, but it depends on what type of data. I think if you talk about large language model training, the amount of data is huge in the US and perhaps even larger than China because there needs to be a certain quality bar. But if you talk about closed loop data such as a mobile app collecting user action and learning about user behavior and using that to train an AI, China has a huge number of users and users love to try new products. They're not as concerned about privacy. They are concerned but not as concerned. So therefore they're more willing to do things online knowing their data may be captured and used to train some model but that doesn't create a concern. So I think that played to China's advantage both in the mobile era and the AI era. Even in the mobile era the equivalent of Uber in China's Didi and other companies like Pinduoduo and Meituan and especially ByteDance, these are super unicorns that emerged. They took advantage of the fact that users were curious, willing, and not so worried about their usage patterns getting captured. And then because that leads to an improvement in the application, that became a virtuous cycle. So in AI Superpowers when I talked about where China had so much data, I was referring to the data people's willingness to try new apps caused the mobile apps to really have a huge growth. ByteDance is certainly China's most powerful technology company today and it's also China's most profitable company and it benefited exactly from that combination of data, users, and using AI to cleverly learn about these users. And the reason you find TikTok exciting and addictive is because AI could predict what they want to see and keep showing them things that are attractive and addictive to them. And this began naturally with a Chinese company that mastered its use of AI in the face of huge number of users and huge amount of data.

So let me trace back to transformers. Researchers got super excited about transformers in China as did those in the US, but interestingly none, very few of the giants were willing to invest at that time because it seemed transformers were this cool technology that use tons of GPUs and cost a lot to do research. So going back to AI Superpowers, Chinese companies are very pragmatic. So when they see this really cool breakthrough technology that doesn't have use cases or profit, they are hesitant to jump in. And you know companies like Google and shortly afterwards OpenAI and of course Microsoft and others were much more willing to make bets on this breakthrough yet commercially unproven technology and China fell behind. So when ChatGPT came out, I would say that moment the Chinese companies woke up. If you actually read my other book AI 2041 which was written in 2020, I forecasted that a ChatGPT would emerge and what it would do, the benefits and the harms. So being American trained I saw the trend and I felt frustrated that the Chinese giants those days were unwilling to invest in something where I saw the writing was on the wall. It's a little faint but it's on the wall so someone needs to go buy those GPUs. So I was really on a crusade in China suggesting companies to do that because as a VC we can't really invest in a company that requires so many GPUs but large companies can afford to. But anyway that didn't quite come to pass. Although there was a government entity called BAI, Beijing AI Institute or something like that. They were the first to buy a thousand GPUs and they kind of incubated a group of generative AI researchers who are now populating places. So anyway, China was behind because companies like Baidu and Alibaba did have researchers in generative AI but they didn't get the resources they needed, you know, a thousand GPUs or more. But BAI did. So there were researchers who believe in the mission, the value, but the companies and the VCs, the VCs couldn't afford to fund it and the companies wouldn't fund it because of unclear commercialization. So when ChatGPT came out then everybody woke up. Large companies started making big investments and entrepreneurs were coming out to build large language models. So China basically started exactly three years ago when ChatGPT came out. So China really started to catch up the day ChatGPT came out and it took a while, maybe a year for China to get to the second tier of the global best models and in two years China was at the bottom of the first tier. And then today I think it's closely behind the top American models. So in three years, China more or less caught up. The performance of the Chinese models are definitely approaching that of the top American models.

We're definitely seeing agentic AI very big in China as it is in the US. I think some of the differences in the application is that US has a much healthier enterprise market where people are willing to pay a lot of money for software that helps their work whether it's for coding or something else, agents that create value. So the virtuous loop is happening in the US where enterprises are paying a lot of money to the tools which pay a lot of money to the APIs and that fuels a virtuous cycle that has not yet happened in China. So China is behind. All the examples I gave in AI Superpowers were intentionally consumer not enterprise because that market hasn't taken off. So that's a huge disadvantage at the moment for China. Consumer apps, big consumer markets are leading. That's interesting. The consumer market China usually leads. I think it's going to come out again because the product managers and researchers and developers who were trained in the mobile and AI 1.0 era are now able to build new apps in China. So I think the entrepreneurship for building apps in China in consumer will do a great job and also large companies are interesting. The difference between large consumer Chinese companies like ByteDance as an example is much more ferocious, hardworking, and willing to disrupt itself. So I would envision that ByteDance will become the most powerful consumer app company in the world as a result. And the Chinese startups are very fearful of them and I think globally they will spread their tentacles to a lot of areas. They're one of the very few companies that from scratch built super successful multiple super successful apps. Most American and Chinese companies rest on the laurels of one app and then building things around it maybe acquiring but they just kept building new apps. So I think consumer app will see a lot of stuff from China that will be exciting. But one most interesting area to compare US and China is that I think what drives the American generative AI companies is that each of the four big ones, OpenAI, Anthropic, Google and xAI, want to be the one that cracks AGI and they think with AGI comes enormous power and the ability to not only win but effectively cause everyone else to lose in a business sense. But of course they're not doing it for the purpose of making everyone lose. But it's an outcome of inventing a super intelligent AI. So they are pursuing this dream and their actions are in accordance with this. They build closed source models because you don't want anyone to copy your AGI and they raise a ton of money. They believe more compute, more smart people, more experiments will make them win. And each of the four believes that to a different extent. I would say OpenAI probably believes it the most. But that's all very logical and understandable. I see the logic. I respect the work they do. But I think the Chinese companies even though they talk about AGI sometimes really do not feel for the AGI. They kind of see everybody's learning from everyone else and everybody will advance together, some before others. American companies before Chinese companies because they're the ones who created disruptive technologies. But once the technologies are released even if they're not published, most other people will more or less figure out how it's done. Maybe they'll figure out the first principles, maybe they'll find a different way to arrive at comparable outcome, or maybe they'll just distill the models. But whatever they do, pretty much everybody keeps up with the leader. And we see that happening right now. When OpenAI did a bunch of great things, Google said okay code red and they came out with Gemini 3.0. Now OpenAI is going okay we need to go code red. So undoubtedly these super smart companies will overtake each other and as they overtake each other the Chinese companies are following right along. So the Chinese model of AGI if you want to use that term is that some company will arrive first but soon another. So maybe Google will arrive first, OpenAI next, Anthropic third, DeepSeek fourth, xAI fifth, and Alibaba sixth, and so on. But pretty much everybody co-arrives if you consider a window of 6 to 12 months. If everybody co-arrives, what's the point of Stargate? What's the point of spending a trillion dollars to do experiments to learn? Let somebody else spend the money and learn from them. So the American model is like a genius who wants to win the Nobel Prize and four geniuses competing to win. The Chinese companies are more of a study group. They compete with each other, but they know they're learning from each other. They're willing to let each other learn from each other because they figure everybody will get there. And the winner will be the one who develops a business model that can make money. So it's a complete continuity of what I said in AI Superpowers. The Chinese companies are practical. They see spending a trillion dollars to do experiments to get to AGI seems not a great return on investment. Why not spend a billion dollars, not a trillion, and learn from the people who spend a trillion and learn from each other and build everything open source and let everybody share and learn and grow. And as everybody reaches AGI or even if it's not AGI, it's a sub AGI or a valuable super valuable technology, the next generation, then let's compete on business model and let's spend a billion to have the ticket to play the business model game. Let's not spend a trillion. So you don't see Alibaba and Baidu producing massive losses. They are practical public companies. It's not so much just Chinese. If you look at Amazon or Microsoft, these are also pragmatic companies. So pragmatism is not a Chinese trademark. Any company can be pragmatic, just Chinese companies tend to be more pragmatic, less visionary.

So 01.ai is a very Chinese style AI company. We started trying to build the best AI large language model we could and in about 6 months from founding we built the best open-source model and in about 14 months from founding we built a closed source model that was the highest performing model in China. But I made a very unusual decision at the time because even though we did achieve momentarily the best model in China, I think the writing was on the wall that the Americans were going towards trillion dollar training and the Chinese were going to the billion dollars training and I saw no prospect of raising a billion dollars. And also saw the Chinese models were all getting very good. So being momentarily number one didn't really mean much. And we already saw DeepSeek guys were really smart. So I made a very unusual decision to stop building large pioneer foundation models at the time. We could pick it up later, but we paused it. So we then focused on what I thought was a much more practical Chinese thing to do, which is to go into building applications. So we're primarily building B2B applications, agents, and we're currently in the B2B business selling to businesses both in China and outside China who want to deploy AI. The two unique things about how we do agents are: first, we want to build agents that manage critical functions in the business. We don't want to do call center, customer support, improve documentation, that sort of thing. We want to, if you're a retail business, improve your revenue. If you're in a mining business, maximize your throughput. If you're in an investment firm, improve your profits. So we want to do that. And then the issue with traditional companies is they need help to do it. They can't do it by themselves. The tech companies, software companies, internet companies, they can all do this because they're naturally strong in IT and AI, but traditional companies are not. And we also can't just build a product and sell it to an investment bank and say here is an app, use it and it will improve this or that. We don't know their businesses. So the way to do it is I think more of a Palantir style. So we have forward deployment engineers, a platform to build apps quickly. We go into the business, keep their data on their site, and build applications that create value in hopefully a critical chain of their core business. And we get paid for that. We don't get paid a lot unless we help them reach a powerful business goal. So that is a very unique part of what we do, focused on forward deployment engineer at customer site working with them to build an app that's important to their core business function. The other thing we hope for is that after doing a number of deals in different sectors, we will have a product that emerges as a platform. It may have a combination of smaller models more targeted for enterprise needs, agents tailored, and can be customized at each customer site. And then a platform that makes the development much easier. So we don't always have to send forward deployment engineers. We might just have a customer support and we'll see, maybe consulting companies like Accenture can take over this task as we move to become more of a product company partnering with the consultants rather than doing consulting ourselves. So that's the roadmap that we've outlined. It's very similar to what Palantir has achieved and it's inspired by what Palantir has achieved. It's a much longer timeline, a lot of hard work. This is the kind of work American AI companies don't want to do. Sending your best engineers to go to a foreign country, stay there for half of each month, that's not something that a great engineer in Silicon Valley or in many Chinese companies wants to do. But I think hard work is what ends up building the moats. So we're giving this approach a try.

Yes, I think the part that makes what we do appealing in China is that the CEO, many CEOs get it. They want to see AI applied in the core function. They're not sure it can be done, but we're basically saying cover our costs. If we don't achieve it, don't pay anymore. If we achieve it, then pay us a lot of money. That's basically our business model. It's value based. It doesn't always work. So on the other hand, we're a small company. We can only take on four or five customers. So we don't need a hundred. If we have four or five customers who believe in us and if we achieve great success with half of them, then we'll have a very nice balance sheet. I think in the typical CEO's mind, especially the Chinese CEO's mind, they're still hesitant to pay a lot for software or for outcomes. So that takes a lot of convincing. It's a very big long shot. They are very interested in things that give them insight and control in the company. It's a very Chinese top-down company style nature. So they like that. Meeting summary you mentioned is one of them. But imagine if every meeting in the company were recorded faithfully and then you can get summaries of the meetings that you don't participate. You can find out who are the superstars. You can find out who really came up with the brilliant ideas. And of course, the opposite is also true, who are obstructing progress. The ability to monitor all of the company, gain insight, and make strategic decisions. That's highly appealing to the Chinese CEO. Some Western CEO would be a bit concerned about whether that's eavesdropping too much. But typically not as much in China. So that's kind of one interesting insight. The other one that's very unique in China is the lack of adoption of cloud. Cloud is growing rapidly but it's still trailing the US by six or seven years I would say. And what that means is many companies are not willing to use cloud. They're concerned about sending corporate data even to secure cloud or virtual private cloud. So they want to keep all their data on premise. So that's to our advantage. So that's the mixture of feedback. We also have some non-Chinese customers who exhibit some different behavior. They want the data, the analytics, the insights. They're sometimes open to recording and capturing more data. Others are trying to keep the environment more privacy aware. So that's a mixture. They're not as concerned about keeping things on premise. And they're much more willing to pay a larger subscription or fee. And perhaps the amount they pay is substantially larger than the willingness of the Chinese companies. So that's the trade-off we have to deal with.

I think China and US will end up building very different things. I think if we look at it as a collaborative world together, whether China and US companies collaborate or not, they're each working towards their approach and for their users and learning from each other. So I think the division of labor will be US will do the breakthroughs largely and then China will do the open sourcing to make it pervasive and accessible, especially to developing countries. Because these open source models are greatly helping developing countries who want the technology, don't demand the best, and don't have the means to pay for the expensive cost of American model APIs or subscription fees. So they will help this technology develop. Basically US and China will not collaborate but end up resulting in a world where the best of each will benefit the whole world. Breakthroughs by Americans, open source by Chinese, and then the rest of the world benefits. Especially if there's enough difference in the approach and in the focus that people can learn from each other but each have their value creation. And I think the American companies will show the world how enterprises are transformed in the fastest way through the most powerful virtuous cycle. And I think Chinese companies will show a larger variety of cool consumer apps that everybody will use. And then hardware is another area where I think China will play a bigger role because of the more powerful supply chain and the ability to iterate products more quickly. And also we see VC ecosystems evolving and changing. A typical Chinese VC is much more willing to invest in the hardware gadget, AI gadget, than an American because the Chinese companies can build these gadgets much more cheaply. So it's a good return on investment. The American investors have a more appetite for building great models and also building great enterprise apps. So they will each show the world different set of very cool things and that will each make different contributions to benefiting the world.

I think the information channel needs to be more accessed. Particularly I think there could be better ways that an American VC or startup or large company can see what's happening in China and learn from it. The learning from American companies has always been the case. Silicon Valley has been the mecca of China for many many decades. Only recently does China feel like okay we're now close second. But before that a lot of learning happened and I think China has learned a lot from the US and I think US can now learn a bit from China. And that means not just reading the research papers but also looking at what hardware gadgets are happening. Look at what apps ByteDance is building that may not be launched in the US. If you don't look, you won't know the cool ideas that these companies may have.

If I hadn't published the book five years ago, I wouldn't publish the book anymore because I think I published the book for people to have openness and that a very different future is coming because most people did not. Whether it served as a warning that your job may be affected or as a technology you might want to go into or as a business opportunity that you didn't see. So I wanted to paint that before there was a clear consensus. I think there is now a consensus. So that purpose is no longer needed. In terms of looking back what content was correct, I would say that I'm quite proud of the foundation model section which more or less forecasted exactly what happened. It didn't catch every little detail. We didn't quite name agents but the idea of building a foundation model, fine-tuning it, and the creation of more and more human likeness and approaching AGI, those things were all predicted even though there was far from ChatGPT at the time. And I think a lot of the potential issues that AI could cause were raised and they were not fully answered. I think those still can be valuable. If the purpose of showing people AI matters is no longer needed, the ideas about how do we deal with these problems? How do we deal with deep fakes? How do we deal with AI induced biases? How do we deal with job losses? How do we deal with a world where AI understands you better than yourself and your data is captured and used by powerful large companies? Those are the thoughtful issues that I think are still useful. The questions are still there. I wanted to provide thought-provoking example answers that may be wrong answers but at least it provoked people to think better ideas and I think that is still needed. I think people are now still in the mode of either believing AI is panacea or believing AI leads to dystopia. Which comes out to be the outcome depends on what we actually do as participants, not watching passively as super optimistic or super pessimistic observers.

China's most important industry is manufacturing. So getting AI into manufacturing is basically what every Chinese person, company and government wants to accomplish. It is a very difficult problem. Not a lot of progress, not nearly as much progress has been made there because of the complex working environments. Manufacturing not being a homogeneous industry. Manufacturing a car is very different from manufacturing a screwdriver. And before a lot of it focused on robotic arms and computer vision but I think now with embodied AI and also a lot of ability to learn and adapt in environments, I think that is still a very difficult but a promising area. So it's an area that we as Sinovation wants to invest in. We invested in a company called AI Innovation which is the world's largest AI for manufacturing company and we're working with a lot of companies on that. I think it's going to take a while a little bit longer and it's very hard but I think that's the biggest promise, the ability to manufacture goods at a much lower price such as Tesla has done for their cars can be done in more industries using these technologies. So I think manufacturing is a big one. I think consumer apps, I talked about that, clever new consumer apps. And I also think Chinese giants will in general inject AI and disrupt their businesses and improve their businesses faster. So Alibaba will put it in their e-commerce and payment. Tencent will put it in their social and gaming and so on. The speed at which they put in AI will probably be faster than the American companies. So these are all things to watch for. Enterprise on the other hand, I think US will lead for some time. But all the other areas and of course the gadgets, as much as I'm excited about embodied AI, the gadgets are here. The best glasses are made in China. And they're so much cheaper. There are glasses now where you can basically see the words other people are talking. These are translating glasses and they're meeting summary glasses. Without projecting, these are basically meta glasses without any need for metaverse type of things. It's basically AI glasses. So I think they're going to be made thinner, smaller, more battery life and more practical, more AI focused the Chinese way. And then you can extrapolate that to wristbands and pins and watches. So I think a future that's very important to keep in mind is that the phone is absolutely the wrong device for AI just like the PC was absolutely the wrong device for Uber. Because the mobile era came, you now need a device and apps that knows where you are. So now with AI, what do we think the AI interface will be like? It will clearly be speech driven because that's how we communicate. In order to make it efficiently speech driven, it needs to be always on and always listening, not push to talk, not open an app. So if it's always listening and has an infinite memory of what you did, what you said and how you reacted, and it becomes smaller eventually invisible, that leads to a very powerful future where basically we will continue to live posthumously because now you've captured everything. That's both exciting and scary. And also what happens about privacy. But there's no doubt this device will arrive. And I think we are trying a lot of permutations just like there were many tries at Blackberry, iPhone. We're all tries. Eventually the world settled on multi-touch. And now the glasses, the watches, the pins, a bunch of people are trying and we'll end up with an AI first device that will displace the phone as the most commonly used device with which we use AI.

China industry is alive and well. So that's kind of an answer in itself. I think there's sort of two parts of AI workload. One is training. And the American giants believe in training with the largest cluster possible so you can gain that little extra advantage with the model that you train. Spend a trillion dollar proposition and that China cannot do. But that happens to be not the Chinese approach to AI anyway. And I think the companies want to spend a lot less and make it open source as opposed to spend a ton of money and be the largest and the best. So it's just a different mentality. So it's kind of sanctioning something that Chinese companies aren't likely to do anyway and probably can't afford to do anyway. That's the largest model. Of course, it's a nuisance for people who want to train with a billion dollars, not a trillion dollars, because maybe you can't get the chips. So for those companies, it's a nuisance. But some of them are multinationals. So ByteDance can easily train in another country because it has subsidiaries. TikTok is an American company so it can train. So it doesn't really cause some nuisance for some companies but it projects an American thinking that the best company should want to spend a trillion dollars training into a market that doesn't have that desire to begin with. There's the other workload is inference. Inference is when you use LLMs and they don't require super high-end GPUs. Chips. And in fact, many of the Chinese semiconductor companies that come out are coming out with chips and they're going public. Their chips are not nearly as good as Nvidia's. They can't really be used very well for training except perhaps Huawei's chips. But even then they're well behind Nvidia. So forget training. But on inference they're all perfectly capable. They're not very cost effective. They don't use the best semiconductor technologies. Many of them are special purpose ASICs. They don't have general purpose GPU capabilities, but they do get the job done of running DeepSeek. And that's the growing workload. And I have no doubt the Chinese chip companies, there are probably seven or eight startups that are now going public at 5 to 10 billion dollar valuation and they will get the cash infusion to improve their technologies so that they are at least cost competitive with Nvidia if not with some advantage. So I think the Chinese market is perfectly fine and well prepared for the future of AI inference which is the much larger of the two workloads.

The first thing you should do is buy AI 2041 because that outlines the scenarios and the future of jobs. But I'll give you a summary. I think one could say a lot of things but basically many of the answers are not helpful to the average person. I think this is a good question because it asks the average person. So the answer that a great CEO will still be needed, a great expert in mergers and acquisitions will still be needed, that's not a helpful answer because that's such a small number at the top of the pyramid. So I'm going to skip that part. The part I think that is not going to be replaceable by AI is really what makes us human. That is we have empathy, feelings, connections with people. It's incredibly important. For people who have worked in the industry to know that the soft skills are arguably, I would say definitely, more important than the hard skills. The people who do well in companies are those who exhibit empathy, who are well-liked, who are trusted because they do what they say and they communicate well. They can persuade, they're good at teamwork. So I think these are the skills that will set people apart even more than before. So the good news is any average person can work on these. The other part is in terms of jobs, I think as AI does more of the quantitative and now the analytical work, agentic work, what is not going to be affected in fact may grow as a market are the service industry, especially the human centric service industry. So that's both existing professions. Sales is basically still a human to human touch area. So sales would be one of them. Teachers, and there are many other professions like masseuse, concierge, tourist guide. And there will be new jobs emergent that will have to do with people-to-people connection because people at least in the next one or two generations are unlikely to develop greater trust for AI than people. But eventually that may also change. But I think someone looking in the job market today should feel that in his or her likely working life, developing and going into service industry, building soft skills is probably the best advice I would give.

Absolutely yes, but not very fast and not all at once. So we're already seeing. For example, search engine is a great research tool and it's just a tool. A chatbot can give you an answer but you probably still have to ask the question many times to search engines and chatbots and then piece together the answer after you read a bunch of websites. But then look what came out. Deep research actually decomposes your goal of getting a detailed answer and it calls search engines and reads papers and puts everything together for you. So that's an example. It's almost become the overseer but it's still the doer I guess but it's no longer the humans. I think we're in currently a revolution where the humans are going to stop doing specific tasks themselves but just giving a high-level order of what they want done and then the AI performs what is being asked. So that is the step before the overseer. And then the question is, if you think about the corporate hierarchy, the search engine is sort of a gopher level job and a chatbot may be more individual contributor job and deep research may be a researcher kind of job. You can keep going up on the hierarchy. What would a director ask? Can AI do that? What would a VP ask? Can AI do that? So there's no doubt we're going to march up on the hierarchy over the years and that will happen gradually. And there are already many things that people just don't have the ability to oversee because there's too much minute detail. I'll give you an example. AI quantitative trading. We invest in a company called Attention Research and it's the top fund performing in China today. What it differs from other quant funds is other quant funds have human AI symbiosis. The human learns from data, decides what algorithm to use, and then AI gets more data and makes automatic trades but at the supervision of a human. But what Attention Research does is something totally different. It has only one model. The model does everything. So the human engineer's job is to go build the model. Then the model does all the work. So it's kind of at the almost at the overseer level. It's a kind of job where you don't have AI human symbiosis. The human super smart people create the AI then the AI runs itself. I think that is an outcome we expect to see in more and more industries. So starting from of course the industries where the outcome is most automatic. Quantitative trading is an area basically you tweak the algorithm and it prints money. So it's the most straightforward application to go after. And it was the first when I read transformer paper, it was the first investment I made into this company which took many years to finally get algorithms that work well. So coming back finally to your question, what stops AI from taking over more and more? Well, it's really regulations compliance. AI is not going to be running a hospital. AI is not going to be running a bank or an insurance company. There are regulations today required to protect the consumer, the patient, and so on. So that's not going to go away. So the human will still be in the loop. And also there are some cases where it's not just about the data. And in that case the human will still be playing a role. But certainly I see AI marching up to become the overseer in more and more areas. But not all.

I didn't know that. That's interesting. But I would say that's a temporary phenomenon because the inference cost is going down five times a year. So it may be let's say it's two times more expensive than a human hypothetically, next year it will be 2.5 times cheaper. So it's going to be solved with better hardware and algorithms.

I think right now we've only just started the agent revolution. So getting agents to be more dependable, less costly and more pervasive is the next big thing. That's very clear. I think everybody sees that. And I think coming up with the killer app for consumer is going to be a very exciting future. People think about AI infusing e-commerce and everything. Sure that's fine but I think having an AI first application and an AI first device as I mentioned, always listening, always on, becoming invisible and infinitely remembering. I think that kind of a device will be revolutionary. And there are many others in the book. There will be digital immortality, there will be virtual reality will finally work because AI will enable it. They will be able to create the content. And I think AI will transform education. That's probably the biggest transformation I can foresee. The biggest ones tend to be the hardest. Education is so obvious. We need to get out of the one teacher 50 student or 30 student kind of mode. But it's so hard to do because it's so deeply embraced by the culture and education is so difficult to change because people got it right, humans got it right and now you have to redo it. So there's hesitance. The world has changed around it. So I think what will happen is there will be apps that will be more effective than schools. Eventually the schools will be forced to change but it will be a struggle and a long process. I think medicine is absolutely ready for change but it's not going to change so easily because the doctors are in charge. So first I talk about the technology changes, but then a lot of the industry changes are going to be gated not by technology not being good enough but by existing status quo and people who are unable to think differently and regulatory and compliance requirements.

It's always impossible to predict the stock market because it always works as a pendulum. It either swings too far in over optimistic or to over pessimistic. So at a controversial topic like AI, it will undoubtedly swing from over optimistic to over pessimistic at some point but that doesn't mean AI is a bubble at all. So in that sense it's very different from the internet bubble. The internet was truly not ready for those things and AI is creating value today. So the question really is, is OpenAI worth a trillion dollars or half a trillion dollars? The argument an OpenAI person is likely to make is that yes, we make 14 billion this year and spend 60 billion. I don't know if I have the exact numbers, but let's say 14 billion and 60 billion. To an ordinary finance expert, that looks like a massive loss and the company headed for bankruptcy. But if you think about where's the 60 billion being spent, hypothetically if 10 billion was for this year's revenue, but 50 billion is investing for the next two years. And if the revenue triples every year for the next two years, then we're investing for the future. We'll make more money in the next two years than the 50 billion we spend because the revenue grows three times per year. So that kind of a math equation should suggest that at least directionally this is a company creating value, not creating a bubble. Whether the numbers are three times a year or 60 billion divided 50 to 10, I think you have to work at the numbers. They may not be as pretty as I described. But you really do have to think about a unique industry that is growing three times a year, at least for now, and a unique industry that requires massive spend in order to retain and grow your users. So those statements are definitely true. Is it worth a Stargate? Is it worth a trillion dollars? I think that's the ultimate question. None of us can answer for sure, but it's very possible OpenAI is currently overpriced by two times or three times, but more likely than not, it will make up for that gap in the next year or two. So in a rational market, it may maybe should be worth a bit less, but even if it's not, it will kind of self-correct. But in an irrational market where people swing the pendulum, then who knows if the stock of AI companies will drop tremendously in the next two years. There's a good likelihood it would, but it's not because AI is a bubble and it's not because of the massive losses. It's because of the irrationality of the market to begin with and people who take advantage of that.

It took time in the US also. So I think it will eventually develop like the American market. And probably it will be a little bit slower because many Chinese CEOs who are resistant to the cloud are traditional companies and they will tend to take a little longer. So my guess is if China is six or seven years behind the US, perhaps in 7 to 10 years China will be where US is at today in terms of cloud adoption for traditional businesses. Now that said, all the non-traditional businesses are all on the cloud. If you look at all the software companies, tech companies, they're all on the cloud already. So it's not like China behaves as one market. It's really the traditional companies that will tend to embrace things a little more slowly and they'll want to see some more evidence and guarantee of safety and there will be some peer pressure to move. Everyone's moving so I should move too. So I think it would take a little bit longer than the speed at which the American traditional companies embrace the cloud.

The smart city is a very powerful mechanism. I think that can make people's lives a lot easier. But the parts of smart city, the vision and belief that traffic lights and smart roads will make autonomous vehicle happen seems to be not a proven value proposition worldwide. So that's wait and see. But I will say that I am incredibly impressed by Huawei's approach to autonomous vehicle. Because there are basically two ways of approaching autonomous vehicle. You will of course want as many vehicles as you can. So getting more vehicles is important because that gets you the data. And the other is you want the best integrated experience possible. The latter is the Waymo approach. The former is the Tesla approach. So I think US is very lucky to have these two companies, one going deep, Waymo, and one going broad, lots of data. It's exactly the two experiments that every country needs to experiment with. So China obviously has a bunch of companies going deep, all the EVs and so on. But Huawei has done something quite incredible which is they are the first company that has developed an EV plus autonomous vehicle OS, so to speak. I mean, it's not exactly an OS, but they went after all the companies that didn't embrace EV early enough and say, 'Here's a platform. Change your company. You don't have time to do it yourself.' And they're really embracing this. The first company to have embraced this, the stock went up massively after their car became a huge hit. So as a result, I would predict Huawei will be the Tesla that has that data. And it will be interesting to see how this develops. And I think the Huawei OS is probably, it has to deal with many different types of cars and partners. And it might be able to actually gather more data. China has a lot more cars than the US and Tesla is one kind of car. But how fast can one company sell cars? But Huawei is enabling many many companies. So it's a different business model. A lot of companies have tried to build the car OS. Baidu tried it with Apollo but it looks like Huawei's figured it out. It's very impressive. So Huawei won even over Baidu's Apollo. There were Huawei executives who wanted to build a Huawei car but chairman Ren basically said no we're building the OS. And I think that turned out. They'll make a lot less per car but they'll get a lot more cars and then they'll get a lot more data. That may enable a very powerful future version of the OS. So it's a very impressive company that made a very unusual decision for a hardware company to decide not to build a hardware product but to build a platform. Very impressive.

The most straightforward is watch for every AI technology and use it extensively, especially those pertaining to areas that you're going to work on. Because it's like, can you imagine being a journalist without using word or photography without using Photoshop? So that's the obvious feedback. The other is pursue your passion, do what you love and that's how you'll differentiate yourself when in the future you have to compete not just against other people but against AI. So pick something you love because that's the area that will give you the biggest chance. The third is think about EQ and human to human connection and the service industry. And then the last advice is really don't obsess on the job issue. The idea that we are in this world not just for the job that we take even though most people think their reason for being is their job. But it's not. If you go back a thousand, two thousand years at the age of Confucius and Plato, people are many things. They are apprentices. They are able to spend time thinking about poetry and building families and family values. So it's really the industrial revolution that has made it necessary, whether it's the American dream or the Chinese dream, for the capitalists to tell people, 'Hey, your job is terrible, it's very routine, you hate it, but if you keep working hard, you'll get a house and a car and your family's future will be better.' So it was then that people became more and more obsessed with the job. Well, in the future, maybe there will be universal basic income. Maybe there will be a 3-day week. Maybe people will be paid for the hobbies they love. So I think go beyond thinking, obsessing, 'I got to beat everyone else to get a job because that's what defines myself.' Because there's more to life than that.

Thank you. Bye-bye.