One sentence is used to generate an individual-specific teaching/discussion video for K12 education Agent product VideoTutur, which today announces the completion of seed wheel financing of $11 million。
Original title: Zhao Chang Peng voted for a Chinese college student, $11 million Seed Wheel, Education Argentina
Source Park
Three Chinese students in college, $11 million in seed wheels, and Silicon Valley students are currently the most financed product。
One sentence is used to generate an individual-specific teaching/discussion video for K12 education Agent product VideoTutur, which today announces the completion of seed wheel financing of $11 million. The current round of financing was led by YZi Labs, with the joint participation of a wide range of investors, the King Autumn Fund, Amino Capital, BridgeOne Capital and several prominent investors。
This is also the first AI product company to invest in YZi Labs。
The founder, Kai Zhao (Zhao Kai), stated that VideoTutor was recognized and supported by the CZ and YZi Labs investment team and was eventually financed by YZi Labs. They got more than 10 TS, and they finally chose them。
On 14 May, the first version of the product, which was launched at the Founder Park product market, was approved by the market and validated by the PMF, and this $11 million seeder financing was completed in less than five months。
In Kai, it seems that the central reason why they can get this funding is that, in the right direction, the "genius team" has addressed the pain of learning in the U.S. advanced examinations on K12。
"this field is more suitable for young people, with very good engineering and hands-on skills, and very good insight and experience, and very quick implementation. i don't know
Not only are they, Cursor, Mercor, Pika, GPTZero, etc., students in Silicon Valley, using an innovatively financed AI product, to refresh the perception of AI entrepreneurship。
THE AI ERA OF ENTREPRENEURSHIP IS REALLY DIFFERENT。
We talked to these young people in Video Tutor, and we wanted to know why they had access to this seed ship financing, and what was happening in Silicon Valley today. And why do they want to hire workers in big factories in the country。
Interview guests: CEO Kai Zhao, CTO James Zhan。
Interview & Editor
Below is an interview, edited by Faunder Park。
Founder Park: So many institutions are looking out for you, and in your opinion, what's at the core that moves them
Kai:I think the first thing is the right direction. AI, education is a very promising and promising track, and we're going into the U.S. S.A. SAT, AP. The target population is K12 high school students, and we and this user population have a very small, basically no generation. We went through the whole pre-test learning cycle, and we knew where the pain of the exams and the pre-tests lay, and we could make a product that really addressed the pain of this population。
Second, the team is very good. James from Gemini, at Google is the core engineer for AI engineering and algorithms. I've had three years of teaching entrepreneurship, starting from the beginning, and participating in the creation of MathGPTPro during the second year of the year. There are experiences of successful education products。
Thirdly, we're doing the AI education field, the core of which is the animation engine, and we're the core developer of VidioTutur, the most knowledgeable team of core technologies that can make the animation engine very precise。
the team itself has very good marketing genes, knowing how to spread。
VideoTutur is very much in line with an investment consensus in the mainstream VC in the United States, called the "Team of Genius" , which means that the field is better suited for young people, with very good engineering skills and very good insight and experience. I think this is a reason for consensus that all investors can appreciate。
Video Tutor at YZi Labs EASY Residency Demo Day
Fourer Park: Which core issues do your products want to address in education
Kai:Current learning products available on the market can be grouped into two categories: active learning products and passive learning products. Passive learning products, such as byte Gauth, Chegg, AnswersAi, cover what we call Homework Help, and the learning links are very short, mainly for student fees。
And Video Tutor covers active learning, and we don't have to consider students' motivations because they have to study, they have to take exams, like the U.S. S.S.A. SAT, AP. In this scene, there's a lot of visualized pain demand, and 80 percent of the U.S. highs are related to functions, calculus, etc. knowledge that requires complex image rendering. VidioTutur's animation engine can solve this very well。
And, in this field, it's very expensive. In the United States, an average of 2.6 million students take the SAT examination each year, and there is a high demand for fees. Underline SAT courses are expensive, not on a set, but on an hourly basis, starting at an average of $150 per hour, mostly at $230. Many students and parents pay for their studies. But VideoTutur is a good way to smooth out and even replace teacher training, because at this stage the video produced by AI is almost identical to the content of teacher training. This way, students can have their own AI personalized teacher at the lowest cost。
Founder Park: What was your chance to make this product
Kai:Actually, Stanford had a team that did it before us. They also wanted to do visual studies. I was aware of the impact of that direction. On previous occasions, education products were basically connected to the API of GPT, similar to a ChatGPT Wrapper product. But we find that such products have ceilings based only on words and questions. As you can see, business like Chegg and Gauth is falling, and much of the scene has been replaced by ChatGPT, because 20 dollars for students can solve a lot of problems with ChatGPT。
PRODUCTS BASED ON API CASINGS AND OPTIMIZED HAVE REACHED THE CEILING。
However, multi-modular visualization has great promise because there is a very high degree of visual learning in the United States of America. Unfortunately, Gatekeep had a good start, but it didn't go on because it was a little early, the basic model programming capacity was not yet mature, and the GPS-4 had not been released. Plus the mathematical animation engine involved rendering and algorithms, they didn't attack. But our team mastered all the core development of the animation engine, solved the problem and made the video very accurate。
FOODER PARK: You were working with several schools when the product went online. In your opinion, when or what function did it make you feel, "I'm doing the right product, I'm finding the right product," and feel like I found the PMF
Kai:Three dimensions。
First of all, VideoTutor has so far received an API request from 1,000 businesses, including all the major and well-known educational institutions in the United States, even domestic institutions. In addition, many schools want to buy services. C The end-user ' s intentions are more direct, with a student ' s parent and investor who, after experience with the product, gives it to all relatives and friends for trial and all are willing to pay for it. Then he got my phone from somewhere and texted me to vote for us. C End-users have a very strong willingness to pay。
The second point is from the user demand level. Why is Tutor education so hard on the American line? Parents are willing to pay for this because they feel that one-on-one teaching works well. Multi-modular AI technology is now humanized to deliver one-on-one teaching, and answers the question. And, on the American line, video lessons from one teacher to another are no different from the video produced by AI. That's what I'm talking about, "Necessity smoothing," the video course that students buy at great cost, which is no different from the one I.A. created. Less expensive and better teaching。
We have received very positive feedback from many students, and many teachers are willing to disseminate the product, with a particularly good completion rate for the previous period and the length of its use. The 200 seed users we've screened now are all early accumulations。
Third, it's a product of taste and sense. When you keep doing it, from the progress of the entire education industry to the core demand point where students and parents pay, to the evolution of the product itself, the whole logic is closed. So from these three dimensions, you think PMF is enough. At the core is the very, very strong willingness to pay。
WE MADE A DEAL WITH FIZ
Founder Park: Many users have offered to pay, and others have contacted you to invest。
Kai:RIGHT. SAT, AP, IN THIS AREA, THERE'S A STRONG WILLINGNESS TO PAY. IT'S WORTH 100 TO 200 DOLLARS TO GET STARTED IN THIS FIELD, AND IT'S MORE EXPENSIVE TO TAKE CLASSES OFF-LINE, PROBABLY $800. IN THE UNITED STATES, THERE ARE 2.6 MILLION STUDENTS GOING TO SAT, 37 PER CENT OF WHOM WILL PAY AT THEIR OWN INITIATIVE, A MARKET WHERE THE WILL TO PAY AND DEMAND ARE STRONG. OUR PRODUCTS CAN MAKE A VERY GOOD DEMAND SHIFT。
From Park: SAT, this track, for candidates, a real teacher and an AI, will he trust AI
Kai:Now AI answers questions at levels like U.S. S.S.S.S.S.S.S.S.A.S.S.A., AP, and it's not very likely that there will be a factual error. In that case, why is it better than below the line? One is cheap, and the other two are students who can ask questions without fear that teachers who ask stupid questions will have opinions or impatience to learn at any time and at any time in 24 hours。
And the market can be smoothed, and after the American market, we can be smoothed to Canada, the British A-Level examinations, etc., and the demand for payments is huge。
From Park: What do you think about paying for this
Kai:WE'RE A MONTHLY SUBSCRIPTION, AND ONE IS TO PAY FOR THE RESULTS. I THINK NOW AI CAN ALREADY PAY THE RESULTS. WE MIGHT LAUNCH A SET, LIKE YOU PAY $799, AND WE'LL MAKE SURE YOUR KID'S SAT MATH SCORES。
Founder Park: But isn't it true that paying for the test results depends on the individual motivation of the student
Kai:THIS MAY NOT BE POSSIBLE AT THE NATIONAL LEVEL, BECAUSE THERE ARE A LARGE NUMBER OF EXAMINATIONS, IN THOUSANDS. BUT THE U.S. S.S. S.S.S.S. HAS ONLY 62 POINTS, 50 OF WHICH ARE REGULAR, MOST OF THE STUDENTS ARE FINE, AND THE REMAINING 12 ARE BASICALLY AVAILABLE. UNLESS THERE IS A REAL PROBLEM WITH THE STUDENT ' S LOGICAL LEVEL, THERE WILL BE LITTLE CHANCE OF LEARNING. AND THE EFFECT OF AI IS OBVIOUS。
In fact, a lot of U.S. online tutus have this service, and you pay teachers $1,800, teachers tutoring children, and the success rate is almost 100 percent, because SAT points are fixed. As long as the students have a normal IQ, they're basically fine. But I can't get it up in the short term. And there's a problem with the need to open the score gap at the national level, but there's no absolute problem at the American level because it's more about whether you have a knowledge point。
Fees on the basis of results are also the model that was already used by the teaching assistant, and this is a prerequisite。
Founder Park: Will model costs be a problem in your pricing? High percentage
Kai:we have a very high unit price in this area, 69 dollars a month, and model costs are very cheap and not a problem. education is not like the field of coding, and everybody's in roll prices because coding needs to support a long context。
FOODER PARK: Remember the last time you said that your first version took almost two months. What were the full development cycle at that time, such as division of labour, determining what functions to do and what functions not to do
Kai:The consensus of all of us in the team is that it is fast because it is fast to get feedback from early users。
The first version, which was posted on Twitter, generated a great deal of excitement and a great number of users. But many of these users are programmers, investors, or tech lovers, and we can call them "technology tasters." At that stage, the feedback received from them was fragmented and of marginal value. It is also important to screen the true core of seed users, namely high-quality high-school students, from so many a wide a range of users, and to obtain useful feedback through counselling。
THE CORE FEEDBACK THAT WE GET IS THAT THE EXACTNESS OF VIDEO RENDERING MUST BE 100 PERCENT, WHICH IS THE HIGHEST PRIORITY THAT NEEDS TO BE OPTIMIZED. WHETHER UI LOOKS GOOD OR SUPPORTS DIFFERENT TTS SOUND SELECTIONS, WE CUT THEM OFF. BACK TO THE CORE OF THE PRODUCT: WHAT WE DO IS TO LEARN FROM THE SCIENCE SCENE, AND THE ACCURACY OF GRAPHIC RENDERING IS THE CORE。
Founder Park: What was the trade-off at the time of generation
Kai: The peak was about six minutes. The main consideration at the time was that the presentation of ordinary topics should not exceed six minutes. But in the follow-up feedback, we found that there were students who were less capable of learning and wanted to be slower and deeper. We are aware that time should not be limited or more dependent on the ability of users to learn。
For how long
Kai:It should last up to an hour to break the casserole. It is generated in real time by means of communication, although this feature is recent and the first version is not available。
Founder Park: Is there a function that you wanted to do and then found it less important not to do first
Kai:For example, App. The idea was that the app would be developed quickly, but then it turned out that most of the students in the United States were basically studying with Laptop or iPad, and that most of the K12 schools in the United States would send students a Cromebook computer, which was very popular, and that their work was done on computers. High school students basically have one computer, and mobile phones account for less than 5 percent of the study scene, and very little。
Fourer Park: So if it's a major educational or student-group product, the end of the page is the first one to do, and App is not that important。
Kai:Yes, it was already known, after all, years of schooling in the United States. And then we dug up 100 students from the early tens of thousands of users, and more than 90 out of 100 students have computers, so we're more convinced。
Founder Park: Did you also target K12 when you went to the first version
Kai:Yeah, and then it's targeting this group. We don't compete with Gauth, we do test training. A large number of high school students in the United States will themselves choose an off-line training or online learning platform, and Vidio Tutor has very well shifted this demand。
Founder Park: Would K12 be your core user group for at least a year
Kai:It should be a core indicator within two years。
Founder Park: How about a brief introduction to your current technology realization program? Video Tutor did a lot better than any other video-generation model, even when many models couldn't produce exactly text。
James:The video that we produce is both text and pattern. The probably production process is to let the large-language model generate text and the corresponding animation instructions, and then the animation instructions go through our animation engine and are eventually shown on video。
The text is relatively simple, and we let the large-language model generate the text and then render it directly. But the animated part was created by one of our own mathematical animation engines. It has the advantage of being very precise in reproducing coordinates, geometrics, and so on, which is our core technology。
now the big-language model is just text, and this one we're making is like giving the big-language model a piece of paper and a pen to paint the appropriate teaching animation that it imagines. that part of the painting is all our technology。
Fourer Park: How does the final synthesis of the whole video, including audio and video, work
James:In the beginning, the user will enter a prompt, like "What's the charade of stock?" In the first step, we let the big-language model deduce all the scenes, and we usually set three to five scenarios, depending on the difficulty of the problem. The model will then generate an approximate script for each scene. A second reasoning is then made on the basis of the scripts of each scene, producing the text in the scene, the corresponding pattern and the text of the human voice. The human voice text is synthesized with TTS。
And finally, we put all the scenes together and made up a complete video。
Fourer Park: I understand the first version of this program. Is the process of generation changing now that it is interactive
James:There has been a change. Now, in order for the user to see the content as quickly as possible, we're going to have the first scene for the user to see, and the next scene continues to be played backstage. When the user asks questions, we turn his human voice into text, and then hand it over to the large-language model, along with the content of all the previous scenes, so that it can plan the next course. The next scene is the same as before。
From Park: If the user hears a question for a minute, he will ask the question directly. When you have been asked, you return to the model with the user's questions and previous statements. In the process, after the user questions have been completed, will the animation continue or will it stop
James:We've been delayed in 20, 30 seconds from the start, under five. On the interactive side, we're going to make some transitions so that users don't pay too much attention to these five seconds, and the whole process is going to be smoother. In four to five seconds, he can see a whole new story based on his question。
THE DESIGN AT THIS STAGE IS THAT THE TEACHER AT AI WOULD SAY, "WELL, I'LL THINK ABOUT IT," AND THEN WIPE THE BLACKBOARD, LIKE A REAL MODEL TEACHER. YOU THINK THERE'S SOMETHING WRONG, AND I'LL WIPE IT OUT AND WRITE IT OVER AGAIN. IT'LL FEEL MORE NATURAL。
And not only are we waiting passively for questions from users, we'll do Quiz in the middle. We'll reason with Quiz's feedback and user questions. And we're not completely free wheat, but we need users to turn on the microphone and have an action to turn it on and off。
Founder Park: So based on such a mechanism, it can generate a speech for about an hour。
James:There was no limit to the point, and if he had problems, he could keep asking。
Kai:Yes, there are no predetermined limits. In fact, VidioTutur is doing this in a way that we are not creating demand, but better meeting existing demand, as multi-model AI advances. Why would American parents pay so much money? Because the American training industry is more one-on-one, starting at $100 an hour. Just because a teacher under the line can lead a question, I can see where you're not going, and then I can ask you. VideoTutur is also trying to achieve this true teacher's performance, so that every child can interact and teach in real time。
Founder Park: Will students ask to turn on the camera during class
Kai:Not really. Whether a student has access to a camera depends largely on the United States Privacy Act. It's not very easy to design mandatory openings in products, and it depends on the students' will. The main interaction was also through questions and voice feedback。
Founder Park: Technically, are you using small models and cloud-sized models or what
Kai:It's a collaboration. We have a data set inside, and now we have more than 100,000 video data. The better data are manually re-marked and then used to train fine-tuning models. For example, we have over 800 SAT sample training data. These fine-tuned little models will be done with cloud-based generic commercial models like Claude, Gemini。
Founder Park: Does the use of Claude, Gemini or GPT affect the core performance of the product
Kai:We mainly deal with the K12 field, and the base model is already at a sufficient level. But to make sure that 100% is right, we'll call two models at the same time, and if the two models have the same answers, then they're basically not wrong. In terms of code generation, it's more about Claude, and it's better。
Founder Park: What are the technological bottlenecks in the product now? Modelling capacity or code generation
Kai:MODELLING CAPACITY IS ONE OF THEM. AND THEN THERE'S THE RENDERING, AND IT'S NOW UNDER FIVE SECONDS, AND AS THE GPU DEPLOYS MORE QUICKLY. THE OTHER IS LONG-TERM MEMORY. WE NEED TO ACCUMULATE LONG-TERM LEARNING BEHAVIORAL DATA ON THE STUDENT, KNOWING WHAT HE DOES NOT UNDERSTAND, FOR EXAMPLE, THAT HE FORGETS A MONTH OF PRE-SCHOOL, AND CAN BE REMINDED AGAIN。
James:We actually put a lot of effort into rendering time, and we've been doing technological breakthroughs, from 2 minutes to 1 minute at the beginning, to 10 seconds now. Our ultimate goal is to achieve a largely undelayed rendering, and as soon as the user asks, the reasoning comes to an end. This is a difficult task for our team at present, but we have found a new direction。
Founder Park: How can core indicators of products be measured at this stage? How do you judge a video is useful to users
Kai:one of the core indicators is the examination. in the new version, after you read the video, there's a quiz at the end, and you do the right thing, and you don't know it。
The results of the study cannot be seen only in the coverage rate, and some students may read half. Give him a test when he's halfway through, pass it, and the rest is not needed. The core indicator of our products is how many students have increased their scores here。
Finder Park: But his final exam was done in another scene. How did you get this result that he passed
Kai:This is to say that the product culture in the United States is that when the product is used by the user, good results are obtained and there is a spontaneous sharing. Many students will come to share their experiences and achievements after using VideoTutor. We will also make them campus ambassadors for secondary dissemination。
We have 20 high school students made up of campus ambassadors. You see, Mercor was very successful in the early days, using a typical user success story. Mercor helped a lot of Indian programmers find jobs in the United States in the early days, and then they contact the users and shoot them a user story about how to get jobs with Mercor. This creates a good reputation for dissemination. Video Tutor is also the reason that we want more students to use the product to achieve very good results, and then share their experiences with user story。
Founder Park: Where are the main channels for student sharing
Kai:Students are mainly in TikTok, parents in Facebook groups。
Fourer Park: What are the ways in which you plan to grow your products if you put your time at half a year or a year's time dimension
Kai:In essence, I think the Vidio Tutor core is still a C end-user product, and it's very important to spread it. Many of the successful AI applications were initially based on seed users' reputations, such as the designer's feeling good, and they spread. For us, the core indicator is how many SAT candidates have used the product and passed it to other children and parents. Parents mainly use Facebook and Instagram, students use TikTok, and we'll spread on these platforms. When such a consensus is formed, it is natural for school teachers to realize it. We were known to so many schools at an early stage because many teachers used it well and recommended it to the school procurement manager. So, the core is the dissemination of C-end user slogans, and it's the key indicator of how many children have used it, and the increase in scores。
Fourer Park: What is the approximate status of the new version and the timing of its launch
Kai:We want to be official within two months as soon as possible. And by then, students will be able to answer with a very low delay, and the graphic rendering of science scenes will be 100% accurate. Of course, for the time being, we will not cover competition scenes or complex university knowledge like linear algebras, more or more, K12。
Founder Park: Video Tutor what are the barriers or moats now
Kai:I think there are a few points. The first is the data flyer. The video is backed by code, and good video data generated by users can be retrained in fine-tuning models after re-marking. The more data the video works. What is more is learning behavioral data, and we know which points of knowledge are weak for different students, and we can build data flyers, and the more people use them, the better the product is for students. The second is the leading technological advantage, such as the algorithm for animation engines. Although algorithms are not in themselves the most central advantages, the advantages will become more obvious as we evolve rapidly。
Thirdly, the brand, Video Tutor, has become a head brand in education in North America, and the trust of parents is an invisible barrier。
Founder Park: Three to five years later, what kind of product did you expect VidioTutur to grow into
Kai:We want Video Tutor to become an AI teacher for everyone to learn science. We only do science. I think it's going to be more than neighbors in the future. Multi-neighborship is a world-class language learning product, but in STEM science scenes, world-class products have never appeared in the past because science requires too much graphic rendering. Now that basic model technology is ready, I think the science scene will be the next multi-neighborship。
Founder Park: How many business experiences have you had before
Kai:I'm a senior now. In freshman year, we started a business with James to make educational products, with $200,000 of angel investment. Despite that failure, valuable lessons have been learned: you can't fall into a homogenous competition. The app we made at that time, there were a lot of similar products on the market, and it was hard to charge for it at an early stage。
For the second time, I joined another team, MathGPTPro, as a co-founder, for months. At that stage, I learned how to look at product indicators, how to build products, how to expand users. And that's when I came to the conclusion that text-based solution-type educational products are at the end. Because it's no different from ChatGPT, and it's been replaced by the editorial capacity of big models by a structured knowledge base that used to be done at great cost by the helpers. So for the third time, I knew visualization was the inevitable trend。
Zhao Kai's photo of Sam Altman Pitch at Harvard
Fourer Park: What does it help you to do VideoTutor now, in addition to making you aware of the limitations of text-based products, in a team or otherwise
Kai:Very helpful。
First, better judgement of direction and future of products. I will judge the evolution of the whole product by looking at the competition's web traffic, the revenue。
The second point, product creation, is better able to judge the pace of product development, including product design, back-end interface, and what indicators to look at。
Third, team management and organizational culture. I have put in place a more complete management system that includes a division of labour, incentives and options for each student. And I learned how to finance it. This round of $10 million in financing, we're done in 20 days。
From Park: How many people are on your team now
Kai:Six people, everybody lives together。
Founder Park: How did the team start
Kai:Me and James have started business twice. We both graduated from school, and we made an App in freshman year. When I was a sophomore, I started a business with two other people. And when we realize that this technology can lead to a very big product vision, we contact teams to do this. We were all alumni, including another partner in the team, Nick, and my college roommate。
Founder Park: What kind of man are you going to hire now
Kai:Our main sources of experience are back end, front end, large language model and UI/UX. Because we've now passed the trial and error phase, and we've entered the rapid build-up of products, and we need experienced people to help us grow。
Founder Park: Experienced engineers, product managers and growth managers are needed to get products from 1 to 10, or even from 10。
Kai:Yeah, that's the stage. We expect to expand the team to nine to ten people, and the core is still focused on hiring engineers。
this call could be domestic, so it's in-person and remotely mixed。
Founder Park: What is the image of this man
Kai:We'd rather he'd gone through it in some big factories, like bytes, missions. Because the byte is a high-speed, comparable organizational culture that values young people. Those who have been trained in bytes have a better methodology and ability to join us to bring in these successful experiences for integrated learning。
People who want to fight hard in large factories in the country and have fast-track experience. We've been through the student start-up phase, not much to recruit new recruits, much more experienced, but not the kind of full-blown "business-employer." It's not possible because old-timers can't take care of their families. So the middle level, the young and able to roll。
WE ARE WILLING TO GIVE OPTIONS TO GOOD TALENT. WE MELTED $11 MILLION, BUT WHY DIDN'T WE HIRE ENGINEERS IN AMERICA? IT'S BECAUSE WE THINK THE DOMESTIC CAPACITY FOR PRODUCTS AND ENGINEERING IS REALLY GOOD. THIS WAVE OF 100 PERCENT WILL HAVE CHINESE-RUN TEAMS MAKING GREAT PRODUCTS AND COMING OUT INTERNATIONALLY. MANY OF THE AI APPLICATIONS ARE NOW MADE BY CHINESE PEOPLE, AND THE COUNTRY'S ENGINEERING CAPACITY IS REALLY GREAT. IT IS ALSO OUR ADVANTAGE TO TAKE ADVANTAGE OF THE ADVANTAGES BETWEEN CHINA AND THE UNITED STATES。
Founder Park: The trend towards university entrepreneurship is particularly evident, especially in Silicon Valley, and what is the state you see
Kai:Look at the fact that Mercor, the owner of AI-recruited company, has completed a new financing of over $300 million and has already been valued at $10 billion; and Cursor is already a nailed $10 billion valuation. There's GPTZero, Pika, and so on. These are university start-up projects, especially the founders of Cursor and Mercor, who are freshmen。
This wave of youth entrepreneurship is characterized by highly differentiated competition. They have focused on very narrow areas and have not done anything common. Mercor, for example, did the AI recruitment, starting with the Indian programmer recruitment。
The second point is the environment. The capital environment of Silicon Valley as a whole and the bottom-up innovation, like Stanford, YC, Peter Thiel's fund, are supporting university students at the earliest stage of starting a business, whether you have mature ideas or not, and are willing to support you and provide a powerful human network。
Third, I think it's the quality of these college students. These students, whether we or from Silicon Valley, have a very brave spirit of adventure and a great ability to learn. Many students in the country may not have such a brave spirit. Because in Silicon Valley there are many examples of peer success that inspire you, and the capital environment is willing to trust young people。
For me, costs and benefits were also compared. If I choose to finish a university and find a job, I may not be able to afford the cost of studying at home, nor will I be able to have a significant return. But if I choose to start a business, I can go mad at the youngest, and my life has unlimited possibilities. I've wanted to start a great company since I was a kid。
From Park: Why does this generation of university students start a multi-billion-dollar company, and it's amazing if they could sell 12 million dollars? Is there an AI heat and foam factor in this
Kai:I don't think it's all foam. Cursor has a real $450 million collection, which is very reliable. Behind this is the methodology and recognition of this generation of young teams that matter. You see these teams, they have excellent backgrounds, they have great learning skills。
Cursor was an early student programmer with a high level of acceptance of AI and strong feedback. The founders themselves were talented engineers who had a deep understanding of the users and a strong engineering trajectories, and at an early stage four people dried up the product. And when they're done with the product, they create a user reputation, and investors are afraid to miss the next Mark Zuckerberg, so capital helps。
The bottom line is that many of the technologies in AI are new and young people learn fast, practical, reliable and daring, so that there is an extremely high degree of user understanding and super-quick iterative speed to defeat traditional products. Before Cursor, for example, GitHub Copilot did well, but why didn't he? It's because of user experience and speed of implementation。
Founder Park: Is it possible to say that because AI is a new technology, much of product recognition also needs to be seen in a new perspective
Kai:YES, THE YOUNGER GENERATION HAS A DEEPER COGNITIVE PERSPECTIVE THAN THE PREVIOUS GENERATION OF ENTREPRENEURS AND CAN BE CLOSER TO USERS. NOW THE MAINSTREAM AI USERS ARE AFTER 0000, AND THEY LEARN AND GIVE FEEDBACK AT AN ITERATIVE RATE AND ARE MORE INCLUSIVE THAN PREVIOUS ENTREPRENEURS。
So, cognitive trajectories are at the core. In the era of mobile Internet, technology overlaps are annual or quarterly, but in the AI era, technology overlaps may be sky-based. You have to learn fast, as a baseer, while young people stay up late and work harder。
Fourer Park: What do you think
Kai:i'm around some white entrepreneur friends, a lot of money, too. they're like us, renting a big house, all living together. i think 996 is more environmental, and now silicon valley is a little bit of a gold rush, and no one wants to fall behind, but it's faster than the product, and it has to stay up late. this is an environment that compels people to do so。
Finder Park: Are these students in Silicon Valley starting their own businesses
Kai:I think that there's a tendency for everyone, whether we're doing education or anyone else, to start a business in their comfort zone. Comfortable circles mean you have a good understanding of the field and the users. Cursor's founders know very well about coding, and we do education because we know this population well. Young people are now more likely to start businesses in their already existing cognitive comfort circles and no longer jump into an unknown field. Because that's how you get feedback from users that's fast enough to be correct。
There's a cognitive overlap. We've been doing education three times, and my perception is constantly superimposed. These college kids don't do what they've never done in the past. They're trying to do better. They have a new generation of ways of thinking, evolving in their own cognitive circles, and courageously creating opportunities。
There's also a brave spirit that doesn't take care of what you think about me. Behind that is the culture of "high-speed experiments," and I know that my product is not ready yet, but I don't care, fast online, fast iterative, quick feedback。
Founder Park: When did this wave begin
Kai:I think it is a consensus success. When you see projects like GPTZero, growing up from dormitories, evolving over time and gaining capital and user recognition, there are many successful cases of rapid error and rapid explosion。
"Better done than perfect." And people aren't too concerned about competition, and a lot of people in Silicon Valley are willing to talk about their product ideas, and they're not afraid of you copying them. I think this wave of young people has a good ability to tell stories that are not false, but based on realism and their own vision of the future。
Founder Park: Marketing yourself first。
Kai:Right. I think the bottom line is the spirit of adventure and extreme self-confidence. Driven by this, they continue to try and make mistakes, without fear of saying wrong things. It is a big mistake to speak out about its product ideas and to implement them. This culture of fear of error has contributed to the enthusiasm and success of this wave of university students。
VCS FROM THE U.S. ALSO LOOK AT COLLEGE STUDENTS' PROJECTS, AND YC REGULARLY INVESTS IN COLLEGE STUDENTS' PROJECTS。
Finder Park: What do you suggest to yourself if you go back to being Video Tutor? Is there anything better to do
Kai:I think it should be faster. And there's team formation. Video Tutor's team is grinding through multiple wheels. If I'd known, I'd have done better to build a team based on the skills required by the product. I think it's crucial to get back to business. I'll spend more time on organizational skills: choosing people, knowing people, using good people。
Now the team is fit to grow from 0 to 1, but to do much more with Video Tutor, or need someone with more work experience to bring their good experience and abilities to the team and help the team grow together。
Finder Park: In the next six months, what kind of product or technical difficulties do you think VideoTutor might have
Kai:I think one is retrofitting, and to get down to a real zero delay, we need a breakthrough. The second is growth, and I think it's product taste, and there's a lot of stuff behind it, like UI, whether interactive design is smooth, whether functional interaction is not bug, whether visual layout is beautiful, etc. These are all tests for us。
James:I think at first we were targeting Vidio Tutor for visual tutoring in all disciplines, but then we were doing it very vertically, only in mathematics, because that's what we do best. Our mathematical rendering engine is the most specialized. What's next is a breakthrough, possibly horizontal expansion. For example, how can visualization be brought to a literary scene? For example, it's an explanation for the "sweet day, sweat and earth." That's what we're going to do technically。
Foreman Park: Will there be problems with subsequent expansion due to the founder's background
Kai:Not really. Actually, there's a lot of big VCs looking for us, like a16z, and they don't do it too early, but they do it when the team has a sign of success, so they know that the investment won't fail. We have a good relationship with many big VCs。
Finance is the last thing Video Tutor needs to worry about, the last thing to worry about is about user ecology and products。