China’s Top VCs on Backing the Next Generation of AI Founders

During the recent AI Creator Carnival in China, Pan Luan, host of the popular podcast "Luanfoushu", sat down with four of the country’s most active venture capitalists to discuss the evolving landscape of AI entrepreneurship. The panel featured Liu Yuan, Partner at ZhenFund; Cao Wei, Partner at BlueRun Ventures; Zang Tianyu, Partner at Jinqiu Fund; and Chen Shi, Investment Partner at Frees Fund. They shared their unfiltered perspectives on what it takes to succeed in the current AI climate and the types of founders they are looking to back.

The conversation comes at a pivotal moment as the AI industry shifts its focus from the development of foundation models to the creation of practical applications. This transition is reshaping the startup ecosystem, lowering some barriers to entry while raising others, and forcing investors to recalibrate their strategies. The insights from these leading VCs offer a crucial look into how China's investment community is navigating a market defined by rapid technological change, intense competition from tech incumbents, and a new generation of founders entering the fray.

The following is a transcript of the conversation, edited for clarity without altering the original meaning.

Are "One-Person Companies" the Future?

Pan Luan: Let's dive right in. The prevailing narrative is that AI is lowering the barrier to entrepreneurship, giving rise to "super individuals," and even leading to AI engineers being headhunted for salaries comparable to superstar athletes, like one hundred million dollars. I'd like to ask you all, with the trend of technology democratization, what changes have occurred in the current startup and team ecosystem? To what extent has this altered the teams for both startups and investments? And how do you view this new generation of entrepreneurs?

Liu Yuan: From a causal perspective, many one-person companies have already emerged. As long as there's a single case that achieves a hundred million dollars in revenue, it proves the model is possible. The companies we are optimistic about now have fewer people, and the founders are younger. They don't necessarily have to be serial entrepreneurs. For instance, creators and designers can learn quickly and, in a very short time and with high efficiency, create very usable products that generate revenue and attract users. The barrier to entry has become more democratic. This is a very significant phenomenon.

Cao Wei: I have two points. First, I agree with what Liu Yuan just said. The capability of production tools has greatly improved, leading to significant changes in production efficiency and organizational structure. One person can now do much more. My second point is that there's also an element of hype, especially in China. Silicon Valley is a separate case, but in China, it's still a bit early for the one-person company. The complexity, the ability for the ecosystem to connect seamlessly, and the official interfaces being pushed by major MCPs (Multi-Channel Platforms)—creating a closed loop from the digital world to the physical world and to services—are all still works in progress. I think the general direction is correct. Organizational structures and production relations are changing due to shifts in productivity. However, from what we see now, it will take some more time in China to reach a definitive point for true one-person companies.

Zang Tianyu: Several independent developers we’ve backed still ended up building teams after starting their companies. At the very least, they needed people for product, engineering, and operations. It's hard to strictly define it as a one-person company. The team size required to complete a minimum viable business or validate a commercial closed loop has indeed shrunk. We've seen many teams of four or five people who have already validated a product's feasibility. Even companies with around 10 people can achieve monthly revenues of hundreds of thousands of dollars and sustain themselves at that scale for a longer period. Overall, organizations are becoming more condensed, but we haven't reached the stage of one-person companies yet.

Chen Shi: If it’s a truly great business model, a one-person company might work at the beginning, but it can't remain a one-person company later on. Today, the direction of one-person companies is correct; teams are indeed getting smaller. Midjourney, for example, had a very small team for a long time. But while it's okay to start this way, there are challenges as you scale. From an investment perspective, our approach can also change. For example, we can scout for these individuals earlier, not waiting until they have a company or even a product to invest. We can help them with resources, get them to make a single-point breakthrough, and then gradually add more investment to help them grow. This might be an insight for us as investors.

What Are Investors Backing in AI Today?

Pan Luan: Over the past two years, as foundation model capabilities have matured, AI investment has shifted from models toward application-layer opportunities. Since the second half of last year, there have been fewer early-stage investments in large models, with more focus on areas like Agents and embodied intelligence. What changes have you made to your investment strategies? With models stabilizing, what are the early-stage directions you're focused on? At this moment, what AI startup track are you most optimistic about, or most worried about?

Liu Yuan: ZhenFund has never had a habit of pre-setting directions. We never know what track we'll invest in next. For example, our investments in AI applications like Manus and Kimi were made two or three years ago, when investing in AI applications wasn't the "obvious" play it is today. Taking this as an example, there's nothing we definitively won't invest in today. To put it simply, for us, we are focused on what changes are happening with entrepreneurs. What essential elements for starting a business a decade ago have had their weights changed today? In terms of direction, at least at ZhenFund, we maintain an open mind. We could potentially invest in anything.

For example, we now place more importance on the product. In the past, we were often criticized for having an overly elitist investment style. Today, several founders of our recent investments are not from 985 universities [a group of China’s top universities]. This year, we've invested in many unique projects where we've seen founders demonstrate passion from a young age, with early practical experience and ability as creators. Some of these founders are in hardware, some in gaming, some in SaaS. Their resumes might not look special, but their work shows they've created very mature products at a very young age.

Cao Wei: We at BlueRun Ventures do have our own trade-offs based on changing scenarios. We are following major underlying directions like AI and embodied intelligence for the long term. For instance, we have been consistently investing in robotics since 2015-2016. We believe robotics is still in its very early stages.

For example, we break down robotics into four core capabilities. The first is navigation, the second is local motion, the third is interaction, and the fourth is what we call manipulation. Among these four, what's truly been unlocked? For example, navigation is basically a solved problem. For local motion—dancing and jumping without falling, and adapting to various environmental changes—we are about 40-50% of the way there. The third is interaction, which is still a bit clumsy. And manipulation is even worse; current robots have probably unlocked only 5% of their manipulation capabilities. So, we see embodied intelligence as a long-term area of focus, a potential investment opportunity in a super-long cycle, super-large track.

Finally, we see even bigger variables beyond the AI layer. When we look at a physical product, it has many stacks, for instance, on the materials side. Whether it's a wearable device or a robot, materials are a critical point in human-computer interaction. If the glasses you wear are a bit heavy, you won't like them. Or if they cause a slight skin allergy or have technical issues, you'll dislike them. These are many problems that have not been well-solved. In the long run, there are many pain points in materials science, robotics, and human-computer interaction. So, we are still investing in the more fundamental layers of robotics in the physical world.

Zang Tianyu: At Jinqiu Fund, we did indeed shift a significant portion of our focus to applications around mid-2024. By our estimates, about 60% of our investments in the second half of last year were in the application layer. By the first half of this year, with the improvement of models’ Agentic capabilities, including coding, it has further spurred relatively more complex Agent scenarios and applications. More recently, with multimodality, whether in images or video, many models like Nano Banana and others have undergone a major wave of iteration. This corresponds to the content experience, where you can further experiment in the direction of video or multimodality. So, we feel that along this trajectory, the opportunities in applications are becoming clearer and clearer.

Another point is that everyone is now talking about models entering the second half, the "Experience" era. This is naturally integrated with application scenarios because you need to create experiential scenes and environments for these Agents. The fact that Yao Shunyu left OpenAI is also a significant signal. A large part of our energy is indeed focused on applications, but that doesn't mean we don't invest in other areas; we still have our allocations there.

Chen Shi: I think the shift from models to applications is a great step forward. The year and a half from 2023 to 2024 was, frankly, a boring period for the investment industry. What I mean is that everyone was concentrated on barely three tracks: large models, computing chips, and embodied intelligence. The directions were highly concentrated, and so were the targets. This made investing feel like going to an auction market—a scientist would emerge, and everyone would rush in to invest. Of course, this was a necessary process. Starting from the second half of 2024, real application areas slowly began to have opportunities.

From the current perspective, first, entering the application stage is what will generate real revenue for the AI industry; otherwise, the preceding stages are all just a bubble. Second, only by entering the application stage can we see a "hundred flowers bloom" scenario of non-centralized, non-consensus startup and investment opportunities. So, we actually feel that entering this stage is very good for us early-stage investors. And as Liu Yuan just said, there will also be non-consensus founders. Entrepreneurs from outside the traditional mold of "elite schools and big tech firms" can now enter and start businesses in this area. We are quite optimistic about the opportunities in this second half.

AI Speeds Up Learning; Founder's Age Becomes Less Important

Pan Luan: The next part is about entrepreneurs. There has been a lot of heated discussion, even adoration, in the venture capital circle about the "Post-00s" generation [born after 2000]. Everyone feels that investing in young people is investing in the future, which is very sound logic. What are your thoughts on this phenomenon?

Liu Yuan: People born in the year 2000 are already 25 in 2025. Even in the PC era, if you look at China's first generation of internet entrepreneurs—Jack Ma, Pony Ma, William Ding—Ding started his business around the age of 26. The same goes for Pony Ma. Moving on to the mobile era, we look at Zhang Yiming and Wang Xing; they were also in their 20s. The time they achieved success, they were 28 or 29. They all started in their 20s. It's just that today, we're all older, so we say those born in 2000 are young. It's the same in the U.S. In the PC and mobile eras, Mark Zuckerberg and Bill Gates also started their companies in their 20s. History hasn't changed that much; it’s just the people with this perspective who have gotten older.

Cao Wei: I'll talk about this from another angle. First, the pathways for learning and the speed of knowledge acquisition are completely different from 10, 15, or 20 years ago when I first entered the industry. As long as you can ask good questions and want to do good things, you can ask ChatGPT, and it will tell you everything. There are a massive number of advanced productivity tools now that represent a qualitative change from 10 or 15 years ago. So I think age is not important. What’s most important is a team's own learning ability and person-job fit.

When we first talked to Zhihui Jun [a famous Chinese robotics prodigy], my first impression of him was his "super full-stack" ability. Whether it was software, hardware, or even down to joint modules—he would personally solder the hardware with a soldering iron—to the implementation of cutting-edge models and even his thinking on models. We asked him how he could have such a broad knowledge base. You have a vast number of tools at your disposal. Whatever you need to ask, whatever you need to learn, as long as you want to learn, you can interact with an intelligent terminal, learn from an AI model, and acquire a massive amount of knowledge very quickly. At this point in time, just talking about a founder's age is irrelevant. We see a trend where age will become less important. In southern China, many entrepreneurship academies are starting from high school, even middle school, instilling the idea of starting a business. By the time you get to university, you might have already started two businesses. This trend of youthfulness is unstoppable.

Second, getting things done well is a different matter from being young. Some things can't be learned from a model. For example, organizational ability—how to manage a team well. In the past, the ability to manage 500 people was what we called a "big tech" capability. Now, managing 50 people can get a company to unicorn status. We pay close attention to organizational ability, but now it seems its marginal importance is decreasing. So, we feel we need to dynamically look at the person-job fit and the changes in underlying tools.

Finally, it’s about whether the venture itself can bring about disruptive innovation. In the end, what everyone looks at is the essence of innovation, not the age behind it or a dimension based on age. We ourselves are getting older, so I feel this quite deeply.

Gaps for Young Entrepreneurs Under Hyper-Competitive Giants

Pan Luan: If we look at the mobile internet era in the U.S., because the previous era had so many giants, not many new large companies emerged. Today in China, the giants that developed during the mobile internet era are world-leading. Is it possible that in the AI era, there won't be many gaps left for young people, because we see that China's giants are extremely competitive?

Zang Tianyu: It's true that in overseas markets, where the internet was already very mature, going mobile was a natural extension for those giants. However, new companies like Uber did emerge in the sharing economy, just not in the areas where the original giants were strongest or most focused. If we look at the transition from mobile internet to AI today, it's not a completely natural extension. In that transition, new companies have also emerged. As long as you avoid the directions that the big companies will obviously pursue, there are still opportunities.

Pan Luan: Which directions are the ones big companies will obviously pursue that should be avoided?

Zang Tianyu: Right now, the big companies are first focusing on models and a continuous iteration of intelligence. On top of that, they build applications. Directions that are highly dependent on intelligence, like Coding or Agents, are things the big companies will do. Beyond that, many things are built on top of the model's intelligence, and many of these directions can be explored by startups.

Chen Shi: The reason China's mobile internet, or China's internet in general, grew bigger than that overseas is because the underlying traditional industries were not strong. To this day, I feel the mobile internet, combined with the earlier internet, has already blocked the channels for AI in both software and hardware. Today, their "+AI" [adding AI to existing businesses] is significantly more efficient than your "AI+" [building an AI-native business]. This is the challenge of the current "second half." The second half is for grassroots entrepreneurs, and their challenge is that today's opponents are all on high alert. China is also too competitive, ensuring that several major tech giants are working on all sorts of applications. Any application you can think of today, they have already built.

The second question is, what gaps are there? I think the gaps are: first, maintain a safe distance from large models. Second, you need to find a vertical track, whether it's a vertical for AI applications or applications within a vertical industry. Absolutely do not go for general-purpose applications. General-purpose, in theory, is the business of the large model companies today. For one, they have invested a huge amount of money in large models and are now under pressure to charge for applications directly. In this era, I feel that for large AI models, the advertising business model looks unfeasible.

Basically, there are three paths to monetization: first is advertising, second is e-commerce, and third is direct user payment. I think the advertising business model will probably not exist in the AI era, or it will undergo a major transformation. So these big companies, having invested so much money in models, will definitely build applications. It's unstoppable. What I really want to say today is that if you're starting a business, you must think carefully about what you want to do. If you can't figure it out, you should talk to these investors. Really, don't just jump in thinking you want to do something. For example, the Coding space—the big opportunities in China are likely not for startups. This is obvious, and everyone should think more about it.

Cao Wei: I disagree on one point. People should do what they want to do. When you're an entrepreneur, you should start with what you're passionate about. If you don't do it well, you'll pivot on your own. I think before talking to investors, you should still do what you want to do. It’s our job to find the reliable ones among you. For those who aren't, you just pay your tuition and learn your lesson.

The Hype of Fundraising vs. The Difficulty of Spending Money

Pan Luan: We've noticed that some highly sought-after founders are appearing in the market. They can secure a lot of funding even without a product. As investors, are you pre-pricing the founder's future potential? Also, amidst this huge AI wave, how do you judge whether a founder is genuinely passionate about what they're doing, or just chasing a trend?

Liu Yuan: This also goes back to the PC era. Back then, Jack Ma's three soul-searching questions were, "What do you have, what do you want, and what are you willing to give up?" When we talk to entrepreneurs now, we often ask them, "Why do you need to raise money?" Many times, financing isn't because money is the bottleneck or the moat for the project or business. It's because they feel that raising money provides a sense of security, gains them recognition, and gives them a vanity symbol for comparison. You mentioned the consensus-type star founders who raise a lot of money before they have anything.

On one hand, it's easy to understand why investors invest—the supply of such entrepreneurs is scarce, and the window of opportunity to invest in a great project is small. So everyone has this sort of frenzied mindset. But why do entrepreneurs raise money? Especially, why do they do multiple rounds of financing in a short period? Multiple rounds actually suggest you didn't plan well, you didn't raise enough in one go. And the price is still rising. It's like if your IPO stock price suddenly shoots up, it might mean you priced it too low.

Back in the day, there were bicycles of many colors on the streets, but everyone's bikes were the same. In the group-buying era, all the websites were the same. Everyone needed to raise money to subsidize users. I think this might be a kind of inertial legacy. I think the founder of Manus put it well. He originally thought that for a bad company, every financing round is precious. Later, he realized that for a good company, every financing round is precious because it might be the last round of financing he needs. So, I think the current habit of raising a lot of money before having anything is definitely not a good trend. One shouldn't unconsciously compete on fundraising valuations and use it as a safety cushion.

Cao Wei: I have a slightly different view from Liu Yuan. First, let me give an example. Take ZHIYUAN. When we invested, they had only about 20 people, and we gave them a valuation of 3.5 billion at its peak. At the time, the valuation was indeed very high. There was some internal pressure and some controversy in the industry. But what was the reason behind it? It was because we had done enough homework. We had talked to the team six or seven times. Based on their use of funds, their overall team planning, their product iteration speed, and why they needed so much money, we had a very detailed projection. Sometimes these are individual cases. Every "super team" we see that can raise funds continuously is likely an exception, not the rule. So I think using an exception to describe the general rule is not a very good perspective.

So I think for every excellent team, when the team is sufficiently outstanding, its appeal in the capital market, the talent market, and even the government resource market is completely different. This includes what we see with embodied intelligence and its appeal within the robotics industry. When we invest in these teams, we evaluate them from these four dimensions: what is your appeal in these four areas? Is money your biggest bottleneck, or will other problems remain unsolved even if I give you the money? So the pricing of good teams is still more market-oriented.

Based on what you said about the three soul-searching questions, that was from Jack Ma's era. We now have what we call the "security guard's three questions": Who are you, where are you from, and where are you going? In fact, we see a lot of problems now. Many teams raise money but don't have the ability to spend it. When investors are over-excited and the market is hot, everyone can raise money. But after you get the money, do you dare to spend it? And after you spend it, can you turn that money into performance, into value? This is actually the harder part.

So while we hope to see excellent teams and provide them with sufficient resources, a very important precondition is that the team has a clear idea of how to use the funds. I agree with Liu Yuan on one point: the rhythm must be managed well. You can't raise funds without organization or purpose. At the same time, the use of money must be effective. Only then can you form a virtuous cycle from capital injection to value creation, rather than just seeing money being poured in without seeing value being realized, which would lead to an even bigger bubble.

Pan Luan: In 2016, Zhang Yiming mentioned a metric to argue why Jinri Toutiao deserved more attention. He said that in terms of their valuation-to-funding ratio, they were the most efficient. They raised the least amount of money to create the highest valuation.

Zang Tianyu: As investors, who wouldn't want to invest in these people at a more reasonable price? It gives us more room for returns. On the other hand, it also shows that not many people immediately generate consensus and a bandwagon effect among investors. It's definitely a minority of founders with outstanding backgrounds who form consensus more quickly. Add to that the market sentiment, and this is the result. Actually, sometimes we get quite nervous about this. You'll find that in a certain direction, someone comes out and gets a lot of money. You might think this person is the "chosen one," but then you discover another project emerges, and another wave of consensus forms. So is the new person the "chosen one," or was it the previous one?

Pan Luan: If we don't talk about the "chosen ones," are there cases of companies that weren't noticed in the early rounds or, later on, developed well as non-consensus picks?

Chen Shi: I want to add to the case Pan Luan mentioned. I once met an investor who said he missed out on Jinri Toutiao. The biggest reason wasn't something macro. He said he got some non-public operating data from a channel and found that some metric—I can't remember if it was retention or something else—wasn't good. He went to ask Toutiao, and the leader didn't give a particularly good explanation, so he didn't invest. Investing is actually very difficult. If you focus too much on the details, you might not invest. But if you don't pay attention to the details, that's also not good. What you mentioned earlier about why someone can be invested in without a product—I think there's merit to it. One, the person is sufficiently outstanding. Two, the track must have sufficient opportunity.

No matter what, if you join an outstanding person in an outstanding track, you're bound to produce some big opportunities. I think that's essentially the bet. We also need some more detailed, but not overly detailed, investigations. For example, we'll ask them—and sometimes you might think VCs ask silly questions, like "What if the big tech companies enter the market?" A VC really will ask that. It's actually a thought experiment to test whether you've thought about it, whether you have deep thinking. I think deep thinking is very important. If I ask, "What if the big tech companies come in?" and his reaction is, "Why are you asking such a question?" it shows you haven't done your own mental thought experiments and deep thinking. This is something young entrepreneurs should avoid.

Product-Driven vs. Tech-Driven: Which Path Leads to Success?

Pan Luan: In today's startup era, open-source models and "wrapper" approaches have lowered the barrier to entry, allowing more people with non-technical backgrounds who understand the market and users to enter the game. Yet, we also see the barrier to AI entrepreneurship seemingly rising, because a deep understanding of the underlying technology is very important. When a founder with a slightly weaker technical background but a deep understanding of customer needs appears at the same time as a top-notch technical founder who is somewhat naive about the business ecosystem, who do you prefer? Who is more likely to succeed: a product-driven founder or a tech-driven one? And in your practical experience, what's a difference in their success rates?

Liu Yuan: I personally have an extreme preference for the former. Starting a business is about building your own company, making a product. The purpose is to serve users and create user value. A precise understanding of the question "What is the need?" is getting closer to the answer itself. Understanding the problem is the answer itself. Only by understanding the problem do you know where to look, what technology to use, and how it should manifest. Conversely, having the technology first and then looking for a need is, to some extent, completing a technology competition. This is a classic case of having a hammer and looking for a nail, which is always bad. Even today, when technology seems more important than ever, we still believe that between these two basic qualities, the understanding of needs and the pursuit of understanding needs is the more important quality. Pursuing an understanding of needs is more important and more fundamental than pursuing advanced technology, in our view.

Cao Wei: I have a different viewpoint. As the joke goes, only young people make choices; adults want everything. As mature investors, we hope for strength on both sides, because the current market environment has entered a state of extremely fierce competition. The entrepreneurs we're looking for, I call them "hexagonal warriors" [excelling in all aspects]. The more complex the technology stack chain, the more multi-dimensional information fusion and the richer and more diverse the product innovation will be.

On one hand, there's the product aspect: being close to users, understanding needs, and defining a good product. This is user-centric product definition. But when the technology chain is long enough, especially when it involves the model side—and I think the pre-trained model side is a bit difficult, often requiring billions of dollars in expenses, which is not something a startup team can do—but the ability for continuous training, post-training SFT, or reinforcement learning, to be able to create some fancy effects based on existing data and models, is crucial. Because the ultimate understanding is not about the model's definition or innovation itself, but about a more complex understanding of the system that supports your product's functions. Can you see the problem from a system level, rather than from a single technological point? This determines the cognitive height of the team and is key to whether they can supplement their multi-dimensional capabilities during execution.

As for how to view these two sides, in terms of importance, in the current era, especially in the era of embodied intelligence and AI Agents, I think both are important. And we see that whether in Silicon Valley or in China, the well-performing teams are rarely severely lacking in any corner. As investors, most of the time, we invest when we see a team has a particular strength. But in the subsequent investment process, we will provide suggestions based on their own cognitive awareness of their capabilities in various dimensions, helping the team to fill in its missing corners. This way, we can play the role of a co-pilot for early-stage investors.

Zang Tianyu: It's a bit similar to what Cao Wei just said. For an application company, understanding user needs is definitely the most important thing. But building model-based products now is quite complex. You need to build the user experience around the core underlying technology, and at the same time, you need to know the boundaries of the model's capabilities across different dimensions of the entire framework. So, the founders we see now, on one hand, have a good grasp of user needs, and on the other hand, also have a good sense of the changes in models and technology. We also look at people dynamically. How is the learning and iteration ability of both the product and business-oriented person, and the tech-oriented person? Can they make rapid progress? We look at it from that perspective.

Chen Shi: Essentially, it's a choice: do you choose technology or business and product? My view is that when the technological frontier has not yet converged and is still developing rapidly, it makes sense to bet on technology. But once the technology improvement curve gradually converges, you need to consider that AI is just a technology. I think the biggest problem with AI entrepreneurship is that the person needs to have a "real-world feel." If you don't have a real-world feel, you don't know how to manage people, you don't know how to compete, you don't know how to define your growth methods, or how to handle fundraising, social resources, and so on. You will also face challenges. I tend to invest in both, but if the technological frontier is not converging, I'll invest more in tech. But at a certain point, especially in the application era, I think the latter [business and product] is more important.

How to Avoid Becoming "Cannon Fodder" in the AI Startup Era

Pan Luan: The final topic. In today's entrepreneurial environment, some feel it's crowded, while others feel it's just beginning. Some also believe that building applications on top of current models is likely to result in becoming "cannon fodder." Once the underlying technology is innovated, they will be eliminated. For entrepreneurs, regardless of their age, how can they avoid becoming cannon fodder in the AI startup era?

Chen Shi: I'll make three points. First, you still need to think ahead about your business model and moat, even if you can't necessarily build it right away. Second, on the technology side, you need to build a good scaffold, so you're not hindered by model development. That way, when the model advances, you can quickly adopt it. Third, you need to balance present survival with the future vision. In the investment industry, we call this "generating revenue along the way." This is very important. The goal is the long-term vision, but you need to survive. These are the three main points.

Zang Tianyu: Because information is highly symmetrical now, a cognitive lead only lasts one or two months. It's more about being fast and iterating quickly. The second point is, when building a product, don't bet your moat on model improvements. Focus more on solving specific problems, leveraging the know-how and business flow within that domain, or building C-side user network effects or community culture and atmosphere. Build the moat outside the model.

Cao Wei: I think this question is very difficult to answer because the speed of change in model technology is far beyond everyone's imagination. If you're building applications on top of a base model, the model's capabilities extend everywhere. I think there is only one reason you might become cannon fodder: speed. I think nothing else is as fundamental. Because everything else, whether it's product innovation or data flywheels, the underlying data is shared with the large models. The model's capabilities run all the way from the base model to the application layer. So I think only speed matters.

Liu Yuan: Hands-on practice is very important; it's fundamental. As an entrepreneur, you shouldn't position yourself as—the reason I can succeed is because I can be faster than others. Don't expect to be faster than everyone, although you can push yourself to be as fast as possible. Look for specific needs, find what's different, what's new, what others haven't done, and what you can try. Build it as quickly as possible, get feedback from customers and users, and from practice. This is the timeless principle, from the PC era to the mobile era, it's all been like this.

Pan Luan: I hope everyone can avoid becoming cannon fodder. Due to time, our discussion ends here. I hope this has brought you all some food for thought.