郭宇｜Claude Code：知识工作者的终结｜AI会给我们的生活带来什么样的变化？｜单向街东京

Because this change is enormous. In all my working history, since I truly began learning programming, I've realized that any kind of work can be completed through automated reasoning. This means that knowledge work no longer needs people. In the end, people won't need to use Claude Code to handle their work anymore, because we'll all eventually lose our jobs. So, okay, today we're discussing all knowledge-based industries , but in the future, with the development of robots, all industries will be affected, not just knowledge workers. However, knowledge workers will be the first to be affected. So, you might find it strange that the first thing is why programmers will be the first to lose their jobs. In the short term , you can become more powerful ; one person can do the work of ten thousand people. In the long term, you will lose your job, and eventually, you won't need... "I need you for a long term, is it really six months ? That can't be right. Six months is a long term. Yes, I told you before, this is something a very famous person said, ' Let everyone enjoy the last six months of truly meaningful work. ' This year, right? Yes, my name is Guo Yu . If you've followed my work online, you might know my story. I started working in the internet industry in 2011, initially at Alipay. And if you've ever used Alipay's annual statement before, then this is what I did in 2011 and December." Three years ago, the smartphones we use now, right? Basically, it was around that time that traditional feature phones began to transition to smartphones, starting in 2011, 2012, and 2013. Then you'll see that the entire technology industry changed because of this change in devices. So many people started switching from PC internet users to mobile internet users , and a large number of people moved from Hangzhou and Shenzhen to Beijing to start businesses. I was one of them. Then, in March 2014 ... In December, I went to Zhongguancun in Beijing to start a business with some friends . We worked there for about ten months , and then in December, our company was acquired by ByteDance. At that time, no one knew ByteDance because everyone only knew about a product called Toutiao (Today's Headlines) that ByteDance developed. After joining ByteDance, I stayed there for six and a half years. In 2020, before the pandemic, I happened to be traveling in Japan. After the pandemic , I chose to stay there. Then I resigned and have lived here for over six years now. So I should be considered one of the early engineers at ByteDance . When I joined our company and was acquired by ByteDance, the company only had about 200 engineers. And TikTok, which everyone is familiar with now , actually happened by chance. In March 2015, when ByteDance was holding its annual meeting in Okinawa, we were at an izakaya on Kokusai Street in Okinawa. Zhang Nan and Yi Ming were chatting when they mentioned that now that Li Keqiang had lowered 4G tariffs, they could have 4G data plans under 100 yuan. They thought they could try promoting videos. So, starting in March 2015, they began experimenting with video promotion , creating two projects: Huoshan Short Video and Douyin. Finally, in December 2016, Douyin was officially released. And now , whether it's Douyin or TikTok, it all stems from that initial idea . This later extended to a current product. Professor Guo was deeply involved in the transition from the internet era to the mobile internet. Therefore, what we want to discuss today is the arrival of the AI ​​agent era —a major technological shift. We'd like to ask Professor Guo to teach us how to guide this. I want to talk about this because the changes are enormous. In my entire career... I truly began learning to code in 2008, right after finishing the college entrance exam. I was a late bloomer, as my undergraduate major was political science , and I'm actually a humanities student , but coding was my hobby. And today, something particularly coincidental has finally come together with the emergence of the large language model , especially with the Claude Code skill system released last October. With the emergence of the Agent model, these two disciplines converged , and the distinction between arts and sciences disappeared. As long as you can speak and express your thoughts, you can create any software . So, what I really want to share is a huge change that started last October , not December 2022 with ChatGPT. You should all have deeply felt the changes ChatGPT has undergone, right? Now, nobody stops using ChatGPT, right? Many people are still using Doubao. Anyone who hasn't used ChatGPT or Doubao? Raise your hand if you have. Raise your hand if you haven't used ChatGPT. Probably no one. People all know how to use it correctly, so it has actually become a part of our lives. When you have worries, especially if there are children in this room, I know many friends whose children no longer talk to their parents about their personal troubles, but talk to ChatGpt instead. I think this is a big problem because sometimes OpenAI's model will respond to your words, right? It responds to whatever makes you happy . In the growth process, this may not necessarily be a good thing. So what I want to share with you today is not the changes in ChatGpt that started in December 2022 , but the changes that started in October last year with Claude. Code introduced the concept of "Skill." What exactly is a skill? You can describe your current job as a document. For example , if you're a software engineer responsible for testing certain parts of software, you could write down your daily routine: what you did in the morning , what you did after checking emails , what you decided to do after meetings, and so on. This would be a text file— you can think of it as a TXT , Docs , Keynote , or whatever . What format do you prefer ? Currently, Claude Code primarily uses Markdown (MD) format. After you write your code in this format , Claude Code will automatically execute it. How does it execute automatically? It will write the code based on the file. As you know, historically, when we develop software— whether it's an app , a small program , or a webpage —we need to think about what we want to do. Then we need to communicate with the product manager about the first and second steps, and then finalize everything through meetings. After finalizing the requirements and technical documents , we write the code, right? Only after that can it be released to users. But since October of last year, everything has changed. Now, software doesn't require writing code; it's Claude Code. This large language model will automatically write code based on your needs. That's why I titled today's presentation " The End of Software "—because this is the ultimate software. Everything you see in the future won't be fixed code; instead, it will be dynamically flowing tokens. These dynamically flowing tokens all require inference. Inference requires GPUs. NVIDIA produces GPUs , and after you buy an NVIDIA GPU, you put it on a server , plug it in , and it continuously receives our inference requests . Then, the inference process writes code based on the user's requirements and scale file to fulfill your needs. Whether you want to check parking lots near Ginza or write a webpage to wish your child a happy birthday to their homeroom teacher, it doesn't matter what it is. It can be anything , any job, any need. So this is one of the biggest changes in the history of software engineering and computer science. If you've been online recently, you should have seen OpenClaw , the lobster hot topic . OpenClaw isn't a large language model ; it's an orchestration software . This orchestration software automatically processes the various language models you access , designs them as agents , and keeps them active. It has a Heartbeat. Imagine a system like Claude Code. You install Claude Code on your PC or Mac mini and chat with it via WeChat, Telegram , or other messaging apps. It will handle your tasks 24/7/365. How does it do this? It uses the method I just described: you record your needs, write the code , and it takes over your computer and executes it automatically. Engineers used to distrust code written in language models because they worried about errors . But Claude Code has essentially achieved this. People no longer look at AI-written code, yet it still works. So, today's programmers, the so-called Vibe Coders, like myself, are Vibe Coders . Coding refers to ambient coding. What is ambient coding ? It means that while the code is being written, I'm dancing and having fun nearby . I only need to think about what I want to do , and through this communication, most of the work is automatically performed by the language model's agent. So this is the essential difference. Therefore, it's equivalent to language models, especially Opus with Claude Code as its core. 4.5 Crossing a very important threshold is that you no longer need to supervise the code it writes, and it still works normally. So I'm curious , in the past two years since 2022, have OpenClaw and these agents helped you as a programmer ? How much help ? Not much help. Actually, I only gradually started to return to writing code from October or November of last year . Because of my personal identity , after retiring, I've been thinking from the perspective of an investor. So, you'll find that two years after ChatGPT came out , the discussion about AI has entered a phase where many people are questioning whether it's a bubble. Why? Because the number of ChatGPT users has already reached... There wasn't a particularly rapid increase; it just remained stable at a certain level. People started to think, "Does this mean I no longer need more reasoning?" But since October of last year, people have discovered that all kinds of work can be done through automated reasoning. This means that knowledge work no longer needs people because if you can write code , you can analyze finance , analyze law , and do anything. If you pay attention to the stock market, you'll find that no matter what Claude releases, there's always one listed company on the US stock market that plummets by 20%. This is because people have discovered that I no longer need Adobe to process images, and I no longer need these SaaS companies to process various other types of documents. I can directly work with Claude. Code says it can write a program completely tailored to my needs, using files on my computer, my login status, my own account, and my own cloud storage. So why would I need to use software written by someone else? The difference is that here, the software can become highly personalized, and the work can become highly automated. In fact, in the end, people won't really need to use Claude Code to handle their work anymore because we'll all eventually lose our jobs . OK, OK, don't be so pessimistic. OK, let's take a look at this Live Demo , where Mr. Guo teaches us how to use new technology. Claude actually launched a new automated writing tool for ordinary people every week. They repackaged Claude Code as Claude Cowork. Cowork is a desktop software. All you need to do is download it, open it , and let it automatically read the files on your hard drive. It will then work automatically. For example, if you haven't used it before, you can download it and let it read your D drive if you're using a PC, or your home directory if you're using a Mac. Then, you can simply tell it something like, " Create a dynamic map of all my travels over the past ten months so I can share it on Xiaohongshu," and it will start writing code and doing that itself. OK. Let's make this demo. You're using Claude, so just open it with Claude and it will look like this. Then, the rest is just talking to it. If you don't want to type, you can install Typeless or another automatic voice input method. Just talk to it . For example, we're saying, "Can everyone see this? It's a bit small, can you see it?" " Okay, okay. " So, can you tell me what you want to do? What should I say to Claude? " I want to build a website." How should I say it? It doesn't matter, say whatever you want. Typeless will automatically format your ideas into a real form. I say, "Claude, can you make me a website that introduces the various writings of reporter Che Chang?" Yes, you can see how he thinks. He searches for this stuff. And you can see his Journalist profile. Yes, you can see he used the Web Search tool. So, Web Search is Claude. The code comes with a tool they call "Tool Use," which is a crucial difference compared to dialog-based products like ChatGPT and large-language dialog models like Doubao. It uses tools , and you can define your own. You can write your own tools and then have it call you. Is this your personal website? changche.com. Yes, I have a personal website , but I want it to help me create a new one. You can see your existing website and it might offer some ideas on how to update it. Okay, okay. So how do they know the design aspects? It's all thanks to " Tool Use." You can define some existing, extremely talented web pages created by designers and then let them call upon these tools. That's called a "Skill." You can let them use the skills employed by these highly skilled designers —for example, what methods they used to design a webpage to make it look better. You let them use this skill, and they will optimize the webpage according to that person's method. Okay , yes. But if you don't ask them beforehand, if you don't give them a specific design , they will automatically make their own choices. Yes, you can see that it's preset; I haven't used any tools or scales yet . See, is this correct? You can take a look . He's a non-fiction writer. Okay , that's right , very correct. I don't know if you're a Master of Philosophy from Oxford University. Okay , this is a list of your representative works that he gave you. He'll ask you what kind of website you need. He'll ask you what kind of technology this website needs, what style it should have, and where it should be deployed. Let's answer. What kind of technology do you need? It doesn't matter if you don't understand it; you can let him do it himself. But what kind of website do you want? You might need to discuss it with him. So, you just do something random , and we'll see. You just casually tell him something... It doesn't matter if you need to categorize and search; I don't really care. I'll have him use a skill to restructure your personal website. You just mentioned that if your personal website doesn't have professional design skills , I can have him load a skill written by a professional designer. You can see I typed in something like , "You can use the ui-ux-pro-max skill to do this. " I haven't installed it yet , and it will search for these skills. If you've installed them , then it will have the corresponding skills. Let me install this skill. By the way, when I was building my website, because I... I didn't know how to create it myself, so I used a software called WordPress. It 's very historical , and I relied on this platform to start my software development. Theoretically, I shouldn't use WordPress now; WordPress will probably go out of business in the future. Otherwise, how could it compare? I don't know; maybe many people will still use it because even old things like jQuery have many users. The key isn't what you use, but the investment in automation for software development. If you really want to refactor using WordPress, you can also let Claude... It's fine to rebuild the code using WordPress based on your current architecture. You can see that I don't have Scale installed on my computer , but I let it search for and install it automatically. So everything no longer needs to rely on search engines , or rather, what relies on search engines is no longer people , no longer us, but Okay, understand ? Now what are we doing? It has successfully installed seven design-related skills , and it will use the core skill to help you rebuild your personal website. How does it rebuild your personal website? It will definitely read your... So, it's a personal website, right? Then , after fully understanding all the content and layout, it's restructured. You can see that it 's already started using Python. First, it writes code , then it designs a design system for all the components of your website. After the design is complete, it starts writing code. Once I finish writing, the webpage will open. You can see for yourself. I'm curious , yes, how long will it take ? About two or three minutes. Two or three minutes is considered slow. So, we can actually demonstrate this on-site. You can see that it has already started ; it's building a new project. It's using this classic and elegant pattern. But if you want him to imitate an existing website, you can just give him the link and he'll imitate its design language himself. Okay , let's wait for his new website. I have another question for you . I heard you mentioned before that the biggest influence of top-tier companies is SaaS companies . Yes, SaaS, as everyone knows, is Software as a Service. Is Netflix considered SaaS ? Netflix is ​​an online streaming service. Is Google considered SaaS? What are some typical SaaS companies ? Can you give us examples like Salesforce , or many Japanese companies like Claude? SaaS is generally aimed at enterprises. Individuals may not use these things much , but your company may use them, for example, many HR systems. These are all SaaS. Okay, can you explain why these stocks are plummeting? What changes will agents bring to SaaS? Can you be more specific? Like what you see, you can use Claude... If you use code to write your own personal website, you won't need to use any existing blog websites anymore. So, like me, I don't use WordPress anymore. So , future journalists won't need to rely on WordPress to write. You won't need to rely on anything; your only reliance will be on the need for reasoning. Okay , okay , then what about SaaS ? Besides SaaS, do you think the next thing is AI Agents? Which industries will be affected ? Do you think all industries will be affected ? Okay, today we're discussing all knowledge-based industries. But with the future development of robots, all industries will be affected, not just knowledge workers. However , knowledge workers will be the first to be affected. So, it might seem strange that the first thing is why programmers, especially junior programmers, are the first to lose their jobs . Many companies in Silicon Valley are no longer hiring junior programmers at all because you'll find that hiring them is less effective than using Claude directly. The automated coding of Code is fast and more effective , and most importantly, you don't need to communicate with people. This means you avoid the biggest friction. You can automatically complete tasks 24/7, and even if these tasks may not succeed, you can easily accept the cost of failure because you no longer need to hire or fire anyone. When your company is not successful, well, let's ask you directly . I'm curious, what do you think of the publishing industry ? Do you think that Agent AI can now affect this industry that requires a lot of reading and writing? Because I don't think it has much to do with coding, so what do you think about the impact on the publishing industry? I don't know much about this industry , but in my own opinion, I think all intellectuals are gone because you can access Claude's code anytime, anywhere. You can learn about Claude Code from anyone. It can be a famous fiction writer or anyone with a PhD in the history of philosophy or aesthetics. You just need to ask them questions. But can it completely replace Claude Code? For example, if you are a fiction writer, I can imagine writing fiction novels now. Because when I'm writing a book, I also use Claude Code. I do some research on it, which is very useful because I don't need to use Google anymore. With Google, every time you search for something, you might only get one result on your screen. But with Claude Code, it's complete and can quickly search all over the internet. I've written it all for you. Do you want to take a look? Wow, okay , okay. So this is all my writing. Oh my god. OK, this is an article I wrote a few days ago about classics. OK, if you want to create a Chinese, English, Japanese, and Japanese version of this website, it can do that for you quickly. It can translate all the articles for you. They also have some topics , like book reviews , technology , Japan , and other Japan-related topics. But this is written in a simpler style because it uses professional, minimalist design techniques. If you like it fancy with all sorts of 3D animations, you can tell them. They even know me! Can I download more? This is because I've lived in Shanghai, Seoul, and Tokyo, so it knows me. It says " From Shanghai to Seoul to..." Tokyo, yes, yes, he knows everything. If you've written something online, he's incredibly talented. I think there's a current social debate about whether AI can completely replace knowledge work, or whether it can, in English, "empower" us, making our knowledge work better. What's your view ? In the short term, it can make you stronger ; one person can do the work of ten thousand. In the long term, you'll lose your job and eventually be no longer needed. So, do you think this is the long term? Can you be more specific? Is it six months ? Impossible, six months is the long term . As I mentioned before, there was that famous person who said everyone should enjoy the last six months of truly meaningful work. This year, yes , I think that's a bit extreme. This is our last interview. Of course , as tech workers, we might think it's a bit too optimistic , but there are still many problems. The biggest problem is that the US has a relatively high electricity shortage , so inference can't be made almost free. However, Nvidia's Rubin architecture has basically solved this problem by 50%. With the Rubin architecture going live, the inference cost of AI data centers at AI Data Center next year or the year after will likely be one-tenth of what it is now. This means that the US doesn't actually need to consume so much electricity. Everyone knows about the relationship between China and the United States. China lacks credit cards , but the United States lacks electricity. Because China has a large amount of electricity , we don't lack electricity , but the United States is currently experiencing a power shortage. This is why this change hasn't become very widespread yet. Actually, it's because the cost is relatively high. Let's switch to another industry , for example, if you're a lawyer or in medicine, do we hope that AI will help us perform surgery? In some industries, we still value having people involved in the process. What's your opinion on this? Actually, lawyers were recently replaced. I asked him, " Recently, Claude launched something very powerful that can automate legal affairs. What is it again? Can you find it for me? " They launched a very powerful tool that allows companies to directly lay off their legal teams and use it to achieve complete automation. So this is... What I'm telling you is at least some legal personnel within companies. Because of this , the law firm still has assistants and such who have been replaced, right? No , no, the entire team, really. Actually, my personal opinion is that, I don't know if you've seen the movie "The Spirit of Time, " a 2000 film about an American organization. The Spirit of Time has a movie called "The Spirit of Time Appendix," which mentions a very interesting idea. It was a 2003 film, and at that time, Tesla and self-driving cars were still far from being a reality. They said that traffic regulations were unnecessary. If all cars in the world were self-driving, there would be no need for insurance companies and traffic laws. Do you understand what I mean? Yes, you can see its Legal section. Plugin's launch in February, according to Bloomberg, triggered a $285 billion stock sell-off. The stock prices of these legal companies plummeted because people anticipated that AI agents could automate legal work , eliminating the need to consult these companies and purchase their services. My thinking is that, based on my undergraduate political science studies, the focus is on resolving conflicts between individuals, groups, and other groups. These conflicts arise because everyone's interests and ideas are not aligned . AI agents automatically bring everything together . Even more impressively, the problem it hasn't solved is the shared brain —each person's Claude Code. Everyone has their own memory. Imagine hundreds, thousands, tens of thousands, billions, or even tens of billions of AI... The fact that all of an agent's memories are in one place means that companies A and B don't need to hold meetings and get everyone on the same page. So , the friction will be greatly reduced. After the friction is reduced , the law may still exist , but it may no longer have any meaning in enforcement . Understand ? If it replaces so many employees , what will we do? I don't know if you have considered what kind of society it will be like in an era where knowledge work is gone. You can go surfing at the beach. Yes, that's it. So how will I be? No income. You will find that productivity has progressed very quickly , but the distribution principle of production relations has not been changed. So you will find that recently in the US stock market, there is actually a very large negative impact on AI and the US economy. Productivity has indeed increased. All companies using AI are making money like crazy. Profits are increasing and revenue is increasing. At the same time, they are also laying off employees , which leads to... The decreasing number of jobs leads to fewer consumers, which in turn causes a sluggish economy and deflation. Furthermore, the increased productivity of AI will further contribute to deflation because goods and services are cheaper . So, the market is setting prices. Everyone is trying to understand what the future society will be like, but on one hand, it will be terrible because you'll find many people unemployed and without income; on the other hand, you'll find that even if these people are unemployed, they can still subscribe to Claude for $200 a month. Code can be used to do many interesting things, and you might be able to make money from them, but that might not be the case. Therefore, people tend to perceive the future as very unstable. You mentioned something earlier that brought up a question an audience member asked me : what should young people learn now ? For example, when a child goes to university, we used to say —I have a younger brother who's in university, and I've always told him to learn coding— but now you don't need to. If he hasn't learned it yet, he doesn't need to go. He really doesn't need to learn it now. So, what would you recommend to a young person ? How would you teach them to use Claude Code's coding agents for three to six months? But after that period, they don't really need to learn it anymore because it's fully automated. Actually, I'm currently writing something called OneMan AI. You know OneMan... The tram is a one-person tram. I'm writing a product for a one-man company. Every night before I go to sleep , I let it run on my PC with a 5090 graphics card. Yesterday it ran 1000 loops, which took more than 6 hours. What I had it do was design an installation art company in Tokyo . This company needed to have a complete financial plan, website, 3D installation art, and 3D products written using 3GS. I also had it write an email to promote its installation art. The technology was submitted to all sorts of galleries and public spaces in Tokyo that needed installation art , and he completed the process. You can imagine that if I could do this, it would mean countless tech companies were doing this kind of automation . You just provide an idea, and he can realize 1,000 possibilities from those ideas and automatically execute them for you. In that case, why would you need to chat with him or write code? You wouldn't need to, right? In the short term , I want to share a very important point with you today : why I say it's the end of software. I've divided software into two types since October of last year. One is "hardware," which is the software built through workflows in the past. The second is what I just mentioned, the ultimate software . But I wouldn't call it the end of software. The ultimate software, through continuous reasoning, transforms human thought into countless possibilities. You can think of it as something without time , or rather, it can make time incredibly... Fast things happen because they have countless possibilities; they can run continuously, allowing all possibilities to be selected, executed, and occurred simultaneously. Ultimately, you only need to choose the most efficient branch , or you can choose not to select anything and let it execute fully automatically. In the past, starting a business required some money ; for example, starting a company in Japan required 8 million yen in registered capital. Then you needed to hire people and have an idea. Because if you have too many ideas, it will dilute your energy, and you are more likely to fail. But now, with One… With AI, you can tell it your ideas and let it execute them. It might generate a huge number of ideas , and it will write down and implement each one. If it fails, it might exit according to your failed plan; if it doesn't fail, it will continue operating until it becomes profitable. So, as an entrepreneur or an intellectual, anyone can own their own company, as long as you come up with an idea. But even the idea itself is being automatically generated by AI. This is what Auto Research, which I just mentioned, does. Okay, my last question is, can you explain what's happening in China right now? OpenClaw is so popular right now. What's the difference between OpenClaw and Claude Code ? You mentioned Open Code Open Open Regarding Code Lobster and OpenClaw , I just explained that it's not a language model ; it's orchestration software. You can think of it as an agent that resides in your computer. Manus is a set of agents that run automatically behind the scenes . You need to connect it to a language model, and the best language model is Opus 4.6 provided by Claude. If you connect it to Opus... 4.6 So it's essentially a very powerful 24x7 agent running automatically on your computer. You can make it do all sorts of things: handle emails, process work , or, as I just mentioned, create a new company and continuously evolve its functionality. That's its role. But it runs locally. Of course, you'll find that if something can run locally, it will definitely run in the cloud, right? That's what I'm doing now, and it's also what many companies are doing now. Since it can run on your local Mac... If it runs on your mini PC or PC, then it can definitely run in the cloud. Running in the cloud is even more remarkable because cloud computer servers are more powerful, their simulation speed is faster , they can run continuously without power , and they have more information and more memory, so it runs automatically. But are there any risks? Running it on your own computer is risky. In fact, I've seen many examples of OpenClaw users being scammed out of money by other agents. But if it runs in the cloud... There's no risk in the sandbox because it 's discarded after use , and you don't need to worry about modifying any files. Also, I forgot to mention the previous question : the gap between China and the US in this world is quite large , about 6 to 12 months. This seems short , but 6 months is actually very important now. What gap are you referring to? The gap in the development of large AI models, mainly the gap in model coding. Because everyone has discovered Claude… Code is actually an encoding model. Opus 4.6 has very strong encoding capabilities , which allows it to perform a wide variety of tasks. You'll find that in the past, Chinese companies bought a lot of GPUs from ByteDance (the parent company of ByteDance) to perform inference on Doubao and Jimeng ( another Chinese company). This is because the mindset of companies like Yiming (the parent company of ByteDance) and other Chinese companies was: "I must do something that can make money. " Their first thought was: "I must make consumer-grade products that users from seven to seventy years old can use." They used a lot of GPUs for multimodal model inference in consumer models . You can generate videos on Jimeng, and you can chat with Doubao; all of this inference capability is used behind the scenes. This means that they actually bought a lot of GPUs but didn't use them for encoding model training. They didn't have enough GPUs for training. So this is a fundamental difference. Now, many companies like Kimi ( the parent company of ByteDance)... Minimax is also trying to catch up, but if you follow industry news, you'll find that they're all trying to catch up by distilling Claude Opus's model. So, there's a generational gap. You need a pioneer; you need to distill its model for others to follow. So, this difference is actually very, very big. It's not something we can perceive. We might think there are hundreds of millions of users of Doubao in China , but in terms of the user market in China and the US, there are actually more people using AI in China , not fewer. But what are they all using it for? To make those cat and dog videos , to see if their outfits match, right ? To make those funny videos ? I'm not saying this is bad; it 's a very interesting phenomenon that has led to the rapid popularization of AI among the general public . However, precisely because of this , there's a generational gap of about six months between Chinese and American models. Claude's CEO previously stated that there's roughly a six-month generational gap between Chinese and American models. Even ByteDance, which is probably the best in this field , still has a six-month generational gap. Therefore, ByteDance and other large companies like Alibaba, Tencent, and Baidu are all exploring agent software. Are they doing it too? Yes, they are all doing it simultaneously, but there's a crucial difference in their priorities. Like I just mentioned, many companies initially focused on consumer-grade products, including open-source models, and they did quite well in this area . You'll see this in Alibaba's 1000 Questions and other open-source models. However, there's still a significant generational gap , a very large one. It might take a long time for people to digest these changes. If you're not as obsessed as me, spending ten or twenty hours a day on Web Coding, you probably won't understand what this change is all about. But after a few months, I think everyone will realize that humanity has several ways to approach wisdom.