Star projects are lining up to sell themselves, is it too late for AI to start a business? 丨Vote in the spit conference

At this point in time in the second half of 2024, when talking about AI entrepreneurship, the overall mood will actually be very "frustrating". Although in the news, the AI track is still making money, Cohere's latest round of valuation has reached $5.5 billion, and Anthropic is said to be close to $20 billion in the new round of valuation. But the other side of the news is that you've heard that all the AI teams with heads and faces are selling themselves:

Stability AI, a large-scale model star enterprise with nearly 30 billion yuan, has been repeatedly hovering on the edge of a broken capital chain, seeking to sell itself for a while and seeking a merger for a while; Adept, which has received the double buffs of Nvidia and Microsoft, is discussing the acquisition with Meta; Humane, which is known for its innovative AI hardware Pin, was revealed to be in contact with a financial advisor looking for a potential buyer, with a target price of $7.5-1 billion, and Character.AI made a big move, the founder sold himself and the core team to Google, leaving Character.AI for the licensing fee for the technology and model.

If you think about it carefully, this seems to be the sentence that I talked to many large model investors about when ChatGPT was first born a year ago: it may be difficult for artificial intelligence to bring a new wave of entrepreneurship, and it will even bring a new round of "entrepreneurial hegemony".

So is it too late to start an AI startup now? The txyz.ai is an interesting sample - it has a very small incision, and it is a vertical application that many people think will "sooner or later be replaced by the pedestal model"; Its founder is a popular science V with millions of fans on the whole network, and he has his own halo; Its founding team, which spans China and the United States, is full of scientific research and is waiting for commercialization – what is the answer to this question in their eyes?

Seminar members: Yan Bojun, a well-known popular science author and AI application txyz.ai entrepreneur; AI Wild Evangelist, LinkedIn Expert of the Year Yu Yi

How does an AI application startup project take shape? （00:43）

Pu Fan: The teacher Yan Bojun I was familiar with before was a popular science V and a popular science writer, but this time the identity of the dialogue was "AI application entrepreneur". My topics are also set around this identity. For example, what is the formation process of an AI startup project at the application level?

Yan Bojun: Actually, the starting point of our thinking is not entrepreneurship. The background of the matter is like this, it was the second half of 2022, I was planning to go to Stanford University to study for a PhD in physics, and return to academic research.

As a result, in the process, the large language model was born, and everyone felt the power of the large language model. Of course, at this time, our thinking is still very low-level, and we are more focused on analyzing the difference between large language models and previous innovations, and focusing more on the shock of philosophical views. And then after the discussion, we began to realize that it was too inefficient to rely on people to do scientific research project by project - as I often say, people should not do physics, people should do AI, and then let AI do physics.

There is no doubt that when we think about it at this stage, it has the business perspective that entrepreneurship requires. We believed that this path could be pushed forward, so we started to do it.

Many investors have asked me the same question before, and my answer is that I don't have the "choice to start a business". Our path is still based on scientific research. To put it a little more lofty, what we seek is to help human beings expand the boundaries of cognition more efficiently, and we are in the passion, which is the direction we are interested in and the pursuit of life.

Pu Fan: I can understand that what you are doing is actually the inevitable result of the development of an industry to the present stage. But what about the timing? Just like the background I mentioned in the outline, there have been many star projects in the past six months, because the high arms race cannot be withheld, either going out of business or selling out, and the patience of the venture capital industry is slowly running out.

In addition, the pedestal model is becoming more and more powerful. Many people have told me that the AI era will be an "era of entrepreneurial hegemony", and any small creative product will be quickly wiped out by the pedestal model. So I'm curious, don't you think it's a bit late to do this at this point in time?

Yan Bojun: No. First of all, our products are started in the second half of 2023 and have been running for a while.

Secondly, the value brought by artificial intelligence technology is actually not the same thing as the value of the Internet and mobile Internet, which we summarize as "reducing the complexity of information processed by the human brain". In any previous era of technology, there is no tool that can do this.

Dig deeper, what has the emergence of AI changed? Our understanding is that it changes the complexity of the computer's output information. In the old era of technology, the complexity of any computer output information was very low, and it could basically be divided into three types: copy, paste, and sort. No matter how complex the process is, the final result is relatively simple, it comes from the existing data, and it is still essentially human beings who produce the content. So this also explains why this generation of AI is called generative AI, because it is a huge increase in the complexity of computer-generated content under the hood.

From this point of view, I think it is normal for some traditional Internet and mobile Internet projects to be smoothed out by the underlying model because it is not complex enough. Especially if an industry can have a super Internet APP, it means that the main contradiction and pain point of this industry is "information asymmetry", and the main role of the Internet is to "break information asymmetry". On the other hand, if a problem of 1 billion can be solved with an APP, then the complexity of this industry must not be too high, and it must not be a particularly complex problem.

The ability of AI is precisely reflected in the complexity, which is not possible with traditional Internet products. And according to some of the latest research, it has been found that the current general model has begun to appear "whack-a-mole phenomenon", and if you increase ability A, the previous ability B may decrease.

What industries do we think are the most promising for AI applications? It should be those fields that cannot replace humans on a large scale even if the Internet comes, and require a high concentration of human intelligence. For example, in the fields of law, finance, consulting, etc., even in the era of the Internet, the value of people is still irreplaceable, and their income and wages are very high, I think this is where AI can really exert its power.

So we don't feel like our products will be smoothed out by the big model. Because if you're really going to grind it out, then I'm sorry, you're going to have to me straight to AGI in one go.

Why is it difficult for current AI applications to reflect "differentiation"? （10:25）

Yu Yi: In the past two years, I have tried a lot of AI products at home and abroad, and the use time may be close to 1,800 hours. Including Mr. Yan's products, I also tried them very seriously. So my curiosity comes more from the perspective of products, for example, why did Mr. Yan choose the current product form?

In addition, I actually recommended a similar popular science AI product on the spot, which was Amine made by the Zhipu team. In addition, there is also a product called Consensus abroad, which has just received financing, and its positioning is also related to popular science search. When I use them, I can clearly feel that they are obviously different from existing AI products, while Mr. Yan's Txyz is not so different from perplexity and secret tower. So I'm also curious, what kind of standard is Mr. Yan for the choice of product form?

Yan Bojun: Actually, there are no trade-offs and no decisions. The main reason why you see the product in its current form is that the current function is simple and easy to do. Later, when more features come up, you will find that the current product is very different. For example, we will launch the AI writing function next month, professional writing.

All of the features we're currently offering are free. Regarding whether users are willing to pay, we have a very simple criterion, that is, whether your corresponding user needs are charged in traditional scenarios. If the traditional scenario does not charge, and you just give him a better experience, then the user's paying mentality is not so strong. For example, the functions of searching and reading are rarely charged in traditional usage scenarios, so our products will not choose to charge at present.

But professional writing, in the traditional scenario, you find someone else to write for you, I don't care if you are looking for a gunman, ghostwriter, or your secretary, the payment mentality is very strong, so in our plan, the product needs to have a writing function before charging.

But that's still not the vision of our product. Again, we chose to use the reading, searching, and writing functions first because it is relatively easy to do and can attract a certain amount of accurate users in the short term. Our product is a technology that we call a self-organization network, which can be understood as a "self-organizing intelligent network".

Its form, simply described as a natural language, is that each user can create some professional work flow to solve the problem through natural language, and can share it, and then form a marketplace. At present, we have implemented a showcase that successfully uses AI to predict quantum materials. In the traditional scenario, a scientific researcher's workload for a week can basically be completed in 5 minutes by talking to the AI.

This is our ultimate goal, and we hope that AI can really participate in the front line, do deeper and more professional specialties, such as predicting quantum materials, quantum computing, and other practical aspects to play a role, and eventually evolve into a community. We target professional users who have a lot of know-how in their heads. We hope that through a set of technologies, users can hand over the know how in their heads to AI automation through natural language. At that time, it was no longer "what content does the user want - we will develop it for him", but users can solve their own problems through natural language, skipping this stage of writing the program.

AI applications are still not "popular" in the domestic market (15:54)

Yu Yi: Mr. Yan mentioned a set of data before, and at present, txyz has accumulated more than 500 papers and 1 million users, so will the current user portrait have very prominent characteristics? The reason why I care about this is because it directly affects the strategy of some of the products behind them. For example, the aforementioned Consensus, their main users have two obvious portraits, one is students, and the other is the medical industry, so they need to cooperate more with medical and health-related datasets, and there are many designs that will also focus on students and medical groups.

Yan Bojun: Of course, I won't hinder the public from using it, but our goal is to change majors more vertically, our largest field is also medicine, the second largest field is computer, and the third field is materials, engineering, physics, etc. Overall, 99% of users are basically professionals, and students are not the largest group.

In addition, there is another set of data that the average user usage time is more than 20 minutes, so the stickiness of users is relatively good at present.

Pu Fan: When it comes to user portraits, I'm actually more curious about geographical distribution.

Yan Bojun: Domestic users account for more than 30%, and most of the rest are international users. Among international users, in addition to English-speaking countries, Spain is also our main user group.

Pu Fan: In my impression, the direction of AI entrepreneurship is very influenced by the local market. For example, in the United States, pan-GPT products have actually been stationery for middle school students, and it has been popularized that middle school students have begun to use GPT for homework, so entrepreneurs are still interested in "vertical application" entrepreneurship. However, in China, the core product drive of the current AI application is to play, or to create a certain "effect".

For example, the first time I heard about Minimax, it was because their AI conversation product Glow is very famous in the "elementary school language C culture" - "language C" means language cosplay - of course, this may not be a user scenario they actively planned, but more the result of users' spontaneous exploration in the later stage.

In particular, I would like to ask you about the differences in AI applications between the Chinese and European and American markets at this point in time today (August 2024). And why is there such a difference?

Yan Bojun: A large part of the reason for this problem is the difference in model capabilities. Of course, the gap is certainly not as big as at the beginning, but because our needs are concentrated in academic research and the ability to demand models is very high, we try it out as a whole and find that the current domestic model is not able to meet the needs of more in-depth scientific research.

If we go deeper and talk about why there is a certain gap in model capability from a technical point of view, this has something to do with Chinese corpus. There are a lot of corpora that can be used in English, and even in 2024, the frequency of overseas users using browsers and websites is still not low, and due to the deep popularization of mobile Internet in our country, a large amount of data is privately accumulated in different APPs, and the corresponding accumulation in the browser and the data that can be picked up will be much less.

One of the most typical scenes is when I introduce txyz to my fans, many people's first reaction is "where is this app". Many young people may have forgotten that there is a thing called a browser, and there is a thing called a website in the browser. This difference between the Internet ecosystems will lead to differences in the capabilities of the final presented models.

Yu Yi: Actually, I still have a say in this issue. Because I call myself an AI evangelist, I have not only shared a lot of information within the company, but also communicated with many companies externally, such as how to implement AI at the enterprise business level. I've even helped others sort out practical application solutions.

From my observations, it is true that the domestic market has not yet reached the level of mass application. When I was sharing at the beginning of this year, I also went to check the WeChat index and found that although the popularity of AI in the somatosensory is very high, at the WeChat index level, it was not until Sam Altman's "Gong Dou" incident that the index of AI, generative AI, and ChatGPT surpassed that of the metaverse. In other words, at the level of public awareness, the popularity of AI is not as high as imagined.

Including I took down the 2023 financial reports of many listed companies some time ago to see if they mentioned generative AI applications, and what they are used for. The good thing is that where there is a global layout, business related to going overseas, it is necessary to be able to communicate across languages, and the work needs to frequently involve translation, production of local publicity, etc., in fact, it has been implemented very widely, and even the entire workflow has been completely changed. There are many applied AI products, such as ElevenLabs, that have established themselves as master suppliers.

But the level of being a "digital employee" at work is still far from what everyone expects. The reason here is that, as Mr. Yan just mentioned, the model ability is still insufficient, and it is not enough to deal with the huge workflow. At present, what AI can do is to complete some tasks well, but this process is difficult to detach from the relevant processing and help of humans.

Of course, from a personal point of view, if you evaluate the value of AI, on the one hand, you can evaluate it from the perspective of productivity, you need to consider ROI, and on the other hand, you can also think about how to improve organizational efficiency and how to reduce employees' work pain through AI. There are obviously many cases of this piece of AI. Even in these cases, although it did not perform this function 100 per cent, it provided at least 20 per cent of the assistance.

So internally, I also "brainwash" the HR team, saying that you may need to assume that AI becomes an employee earlier, think about how your organization needs to change, how to identify the best practices of AI applications at this stage, and promote some very good software or cases when you encounter them, etc.

However, there is another particularly obvious feeling this year, that is, everyone has become tired. Whether it is a new product at the toC level or an industry-oriented product, even among the groups that use AI heavily, everyone seems to have a sense of fatigue to some extent. Many companies are also rethinking what role AI should play for my business, my state, and my company's operations.

In the overseas market, my own observation may be better than the domestic ecology, and many of the products I pay attention to, even through subscription, the current development is not bad. For example, Consensus, which we mentioned earlier, has just received a round of funding, and its current ARR is about $1.5 million, accumulating 400,000 users - this is a relatively healthy number.

The core users of AI are already "tired" (25:52)

Pu Fan: Can you expand on the "tiredness" you just mentioned?

Yu Yi: Let's take Consensus as an example, how did they promote the wave of AI application products last year and how did they get users? In fact, there is the first wave of dividends, everyone has seen the pie drawn by AI, and seen the AI's depiction of the future, which activates everyone's freshness. A lot of people, including me, have been trying out new products like crazy in the last year. As long as it can be launched, describe to me what new features there are, and release some beautiful videos that I can't imagine, I'm willing to open it and try it, at least add it to my waiting list.

My wait list has collected fifty or sixty products, and so far more than 20 have been realized, and more than 30 don't know if it's because the company is gone or for some other reason, and now there is no follow-up. In short, in the last year, we have seen a very strong willingness to try the new product, coupled with the traffic generated by various news reports, which has led to a particularly high degree of tolerance for new products.

But this year, taking myself as an example, I've had some very regular habits through various experiments. I already have my own main supplier, main product, and I will no longer give new product opportunities as before, unless you have a very strong reason in front of me.

In addition, there is a certain threshold for the use of AI. It doesn't have a clear use case like the original mobile Internet application, and you will find that it needs more exploration from you. So for the public, if they have the habit of using software in the past, or just use AI as a tool, they don't have the willingness to "accumulate experience together and explore ways to use it". On the other hand, some users who are purely new to the market will gradually pay less attention to AI application products because it takes time to get used to the fact that your relationship with AI should be "collaboration" rather than "use".

Enterprises do not need to say more, everyone has generally felt the impact of the capital environment, and secondly, they have encountered bottlenecks in terms of payment rate and user growth. As I mentioned the day before yesterday, a lot of ordinary users have started to think about how they can really use AI instead of watching news, but there are many frustrations in the process. So how to make users continue to pay after the first wave of freshness has passed, I think this year has encountered a greater challenge than last year.

Will the "fatigue" of the market make entrepreneurs anxious? （29:00）

Pu Fan: It sounds tiring, especially for a teacher like Yan Bojun, under the premise of the identity change from "scientific researcher" to "entrepreneur", it seems that there are more issues to consider.

Yan Bojun: We have actually thought about this question very clearly, that is, we will not choose to roll on the same function.

I think that if we want to apply AI and give full play to its core value, the most important thing is to jump out of the traditional mobile Internet mindset, or the "curse of knowledge".

I'm often asked when I talk to investors, "Is your user base big?" How's your coverage? In fact, there is an implicit assumption in these questions, that is, any product can only have enough value if you cover enough people and a wide enough audience. This formula is valid in the era of mobile Internet, because as I just said, the main core value of the Internet is to break the asymmetry of information, and the value of "breaking the asymmetry of information" itself is not so high.

In other words, investors want to invest in a product that covers a very large number of people, preferably for everyone, and implicitly assumes that the investor subconsciously believes that this product can't actually make a lot of money from each user. A product that can be used by a billion people and can earn $1,000 per person, which is impossible.

So if you still pursue a large crowd coverage, in the case of many products competing with each other, think about the difference between the good and bad of the same function, think about how I push it, how I can reach him, how I persuade him to change, this is actually the way of playing the mobile Internet, there are a lot of know how in this is very mature, (not the way we are good at competition).

And this wave of AI has provided us with a lot of non-consensus opportunities. As I mentioned the day before yesterday, the core of the greatest value of AI is to reduce complexity, and it will replace those fields in traditional scenarios that must be highly involved by human intelligence. In such a situation, we don't have to pursue 1 billion users, it should not be competing at the level of "art", it should be competing at the level of "Tao".

Specifically, we know very well that what we have to do is to help you predict a quantum material and help you predict the folding of a protein in his actual scientific research. In many practical scientific research, production and research, those problems that can only be solved by very professional scientific researchers in the past. In this matter, there will definitely be a lot of valuable entrepreneurial projects in it. Just like there are tens of millions of researchers in the world, and everyone is different, and each AI also needs to be responsible for the deeper direction of a single field, I think this is OK.

Pu Fan: But even if investors agree with your statement, I think they will still ask you a very practical question, which is how did you get your first 1 million users? Because this is a problem that must be more serious at the business level.

Yan Bojun: Part of the first 1 million users comes from my own influence. Users in overseas markets are more likely to grow naturally.

Pu Fan: What is the feedback so far?

Yan Bojun: The retention rate is still good. We have a data - generally AI products have functional likes and likes - we have more likes than clicks, which is actually rare. Of course, this may be our focus on scientific research. The like-like ratio of general-purpose large models is generally relatively low, because they are used by many people and are miscellaneous. We've also set up a wish list so that users can come up and say what they want.

Pu Fan: I always have the feeling of exchanging space for time, that is, these functions will buy time for subsequent development?

Yan Bojun: Yes.

Pu Fan: As an entrepreneur, Yan Bojun, do you feel that time is running out? Because we have just listened to a large discussion, the consensus is that both investors and core owners have shown fatigue, and core users have established their own moats at such a stage, that is to say, the market has left you with less patience for the next product implementation. In this case, will you be in a hurry?

Yan Bojun: I think it's okay, because we know better who our users are, and these users obviously belong to the kind of "grandpa doesn't hurt, grandma doesn't love" but has a very high personal value. I think there's still a sense of urgency, but it's not to the point where "a lot of people are doing the same thing."

Will the "influencer boss" be an obstacle in the process of starting a business? （36:25）

Pu Fan: I also prepared a more offensive question. You have a very external identity that is difficult for everyone to avoid, and that is "Internet celebrity". "Influencer" means that you are influential and have private domain traffic. To put it more abstractly, you have a certain amount of social influence.

In my personal experience, it's hard to get along with such a colleague or boss. First, he is more likely to be skewed attention and resources, creating a substantially unfair gaming environment. The second point is that as a long-term "favored" existence, he will have a corresponding ego, we Chinese are more about "face", then raising objections has become a more dangerous thing.

And when it comes to entrepreneurship, we all talk about result-oriented. No matter how fierce you quarreled before, as long as your result is good, I can recognize that your insistence is partly right. So from the current point of view, being able to become a top Internet celebrity in the field means that you are the one who has got the results, so at least from an external perspective, the conclusions you make may be more informative.

So I'm thinking, will your "Internet celebrity halo" bring a certain misjudgment to the business management level? Does your presence create a certain illusion for the team? For example, why did I ask you just now how to accumulate 1 million users, that is, because I start from the stereotype, I will guess that the current activity of txyz is not real, it may not match the current product power, and there will be a layer of hidden worries in it.

Yan Bojun: The problem you mentioned must exist objectively, so within the team, I pay great attention to my words and deeds. And as an influential manager, you have to be especially careful not to talk nonsense, especially you need to be very clear about what you don't understand, and avoid paternalistic management.

When it comes to technical solutions within our team and so on, I don't make any decisions, I leave it to our CTO to make decisions, unless he asks me what I think.

In addition, our team is relatively far, and it is not a purely Chinese team. So not everyone will agree with my Internet celebrity buff and influence buff. And the partners of our team, each co-founder, have very strong achievements in their own fields. For example, our chief scientist, he is one of the youngest tenured professors at Stanford, and he has also won many international awards in scientific research, so who do you say has the strongest buff in this field?

In general, we realized this problem very early on, and then adopted a management style where everyone trusts each other, and the CEO doesn't want you to decide everything. In this case, that's where the team is at the moment, and it's actually pretty good.

Do you trust professional managers, the biggest shortcoming of scientific researchers-turned-entrepreneurs? （39:58）

Pu Fan: This seems to support a speculation from the outside world, that is, compared with the domestic venture capital industry, foreign researchers are more relaxed in the transformation of entrepreneurs?

Yan Bojun: Yes, this may be a cultural difference, in the foreign entrepreneurial viewpoint, "professionals do professional things" is more popular. When a scientist turns into a boss, the first thing he thinks of is finding a manager who is good at managing to work with you. Even in the early days of Google's startups, Sergey Brin and Lwarence Page have not improved for a while, and investors have suggested finding a professional manager. Everyone should be clear about what they are good at, and everyone should have a sense of boundaries.

Pu Fan: And there may also be a question about the founder of Ego. I've heard from some investors who find it difficult to convince their portfolio companies to sell their companies. Perhaps in the traditional Chinese concept of "Zhishi", they will default to why the rivers and mountains that Lao Tzu fought down should give it to you. Will Mr. Yan Bojun have two souls fighting in his heart? Will there be a clash between the traditional Chinese concept of "Zhishi" and the garage culture of Silicon Valley?

Yan Bojun: No, I'm very flexible. You say that your company is reluctant to sell and other issues, my only judgment point is whether the act of "selling" is better for the future development of the company, and if it is better, it will be sold, and there is no entanglement in this. I agree with what you just discussed, which is that the problems that Chinese and American entrepreneurs will encounter are different. In Silicon Valley, investors will never ask, "What would you do if Google made this thing?" or "What would you do if a big company made this thing"?

The reason behind this is that the overseas financial environment is relatively mature and liquid. Google plans a product with the same positioning, and the first choice must be to buy you.

Pu Fan: Is it better to buy than to build, and it is better to rent than to buy?

Ono-chan: Why is China's venture capital circle keen on "I want to do it myself"?

Yan Bojun: Maybe it's because domestic costs are relatively low, there are not so many financing channels, and the financial environment is not so liquid.

Another big reason is the difference in traditional culture. There are a lot of companies like us in Silicon Valley, where there are many co-founders, and everyone will assume that the companies are co-created. Correspondingly, if your mindset is ownership, you will inevitably think about having some ambitions, and you will think about how this is how I have laid a foundation.

Pu Fan: In this way, Mr. Yan Bojun's experience is to be friends with the "co-founders" and start a business with friends. And the lesson from "Chinese Partner" is that don't start a business with your brother.

Ono-chan: I think it's probably due to the influence of "educational background". The senior in the prototype of "Chinese Partner", he has gone through the college entrance examination three times, and his understanding of many things is not so light, but thicker. In the cultural atmosphere in which Mr. Yan lives, his thinking mode is more professional managers.

Pu Fan: Yes, many people have mentioned to me the idea of "professional manager". An investor once told me that he went to United Kingdom to do an incubator project, and found that the venture capital there was mainly composed of three forces, one is universities and venture funds, one is entrepreneurs, and the other is professional managers who are constantly circulating from the world's top 500 - and this group is the most important factor that can form a difference from China's entrepreneurial atmosphere in his opinion.

He is now starting to persuade the founders of the invested companies to recognize the shortcomings of their companies, accept this model, and let the "professional managers" take turns to run out one project after another.

Yan Bojun: We can even raise it a little higher. The first attribute of all living organisms is to "live", but none of us can live forever. So the second best thing is that since the physical body can't live forever, what can live forever? Your work, your thoughts, can live forever.

Shouldn't your highest goal be the thing, this product, and this company that I built, as long as everyone knows from history that I built it, it can run forever with or without me, and that's my greatest success. If it doesn't work well without me, then this thing will inevitably disappear with you, so you still haven't reached the "quest for eternal life".

So in other words, the entrepreneur gets to a certain point, hires a professional manager, and gradually fades out and lets it run on its own, which is a greater success.

What is the goal of "breaking even"? （47:56）

Pu Fan: Have you ever set a goal for yourself? What do you think TXYZ should do, even if the business is small?

Yan Bojun: Break even, just survive.

Pu Fan: How far are we from this goal?

Yan Bojun: I think next year will be about the same, because we have started to have some income now, and the cost is not too big, and we don't need many people in this field.

Pu Fan: This goal seems to be a bit too Buddha, which makes me very curious - you said that you meet and chat with many investors, is it the investor you are actively looking for, or is it the investor who is actively looking for you?

Because I think for entrepreneurs, investors will only be in line with your goals in the short term. If you get along for a long time, your goals will definitely diverge. This raises two questions: first, can the development goal of "such a Buddha" convince investors? Second, if you start a business with such a "Buddha" goal, theoretically you don't need to have too much contact with the capital market.

Yan Bojun: "Balance of payments" is not a Buddha, because it involves a problem, what should be the relationship between the three things of the enterprise to become bigger, stronger and longer?

In my personal opinion, the goal of the enterprise is to stay long, that is, to survive. And if you want to live, you have to constantly and consistently generate value, which is a high demand. So being long is your goal, being strong is your means, and being big is your result - I have to keep getting stronger to survive, and then I live long enough and then I roll everyone else to death, and I naturally get big - that's how I think it. The so-called "too big to fail" does not actually mean that I want to be too big to fail, I just want to "not fail", and the inevitable result in the end is that I will become "big".

Pu Fan: So the next question is, is the investor looking for you, or are you actively looking for an investor?

Yan Bojun: Most of them are investors who come to me.

Pu Fan: They took the initiative to ask you, do they want to invest in an AI project or a star entrepreneur?

Yan Bojun: I think one of the main reasons why they find it interesting is because a lot of investors watch my show. The second point may be that our field is relatively niche, and everyone wants to see what is going on.

If you talk about investing, the last round of seed people did vote, and I voted after talking about it once. But he didn't vote for me alone, and we have a lot of partners in the team, such as our chief AI officer, and our CTO has a very strong background.

To what extent is the desire of foreign entrepreneurs to share? （54:13）

Pu Fan: Finally, I would like to ask Mr. Yu Yi, through today's exchange, do you feel the difference between Chinese and American AI entrepreneurs?

Yu Yi: Yes. In addition, I usually listen to the podcasts of some overseas entrepreneurs, and I find that whether it is for the purpose of creating personal IP or wanting to share their own experience about products, foreign entrepreneurs do a better job of "sharing". For some new entrepreneurs, just listening to those shared words, you can get a lot of very specific and detailed answers, which can guide you to do some follow-up entrepreneurial actions on your own, which can provide reference. Including myself, I like to read their blog, which will have a lot of information about product development, product marketing methodology, and even practical experience in user acquisition.

This is a very significant aspect that I have noticed in the long time since I observed the overseas developer and entrepreneur community. Whether from code or personal experience, everyone has an "open source" attitude.

Although there will be a lot of sharing from this group of AI entrepreneurs in China, on the whole, there will be fewer than abroad, which is also a little regrettable for me.

Including now, although everyone will say that everyone is an entrepreneur in the AI era, everyone can do some of their own development and application, including myself, I will actually try to write some small tools for my own use. Occasionally you'll want to try some small promotions and see how you can get to the market level. However, relatively speaking, China's venture capital circle still lacks a lot in terms of developer ecology at home and abroad, especially in terms of some help in the early days.

Including I have observed very new entrepreneurs, now everyone will not be particularly limited to the Chinese market, but in a global market to distribute and layout, including I also asked that there are actually some overseas users, that is, to do overseas markets, you can indeed see and learn from the sages of the things that are indeed much more than in China.

Pu Fan: I was talking about entrepreneurs just now, do you feel the difference between domestic AI investors and foreign AI investors?

Yu Yi: Do you need to talk about the difference between investors?