Dancing with Love, Zhang Huaiting: Staying a Calm "Value Builder" Amid the AI Hype

This article is republished with authorization from LatePost (ID: postlate). Author: Honghao Gao.

Many people call Huaiting Zhang "Big Brother."

"Big Brother" isn't a title earned through seniority. It means you've fought battles, taken losses, and turned things around. When others needed you most, you were out front.

In 2005, Huaiting Zhang joined Baidu and helped build Fengchao — Baidu's ad system — from scratch. The modeling team he led produced Wenyuan Dai, Hua Su, Dong Zhang, Mu Li, Bohai Yang, Yuqiang Chen, Shixi Chen, and Lin Wang — top scientists and star entrepreneurs alike.

By then, Zhang was already someone who could "lead a team."

Nine years later, he co-founded GSX (later renamed Gaotu). When the company hit its hardest times, he pulled money from his personal account to pay year-end bonuses for the team. Before going public, his stake dropped from 22% to 6%, with shares gradually distributed to the team. He rarely brings these things up.

Another nine years passed. When the entire industry was at its most pessimistic, its most demoralized, he started his second company, choosing AI education. "If no one else is doing it, I'll do it. That's the best opportunity." In 2023, at age 46, he founded AI for Dance. A few phone calls later, his old deputies all returned. Today, the company is valued at nearly $1 billion.

His angel investor, Yusen Dai, managing partner at ZhenFund, summarized that "Big Brother" entrepreneurs are most prone to two pitfalls: biting off more than they can chew, and carrying too much ego, always wanting to unveil some grand masterpiece. "But Huaiting avoids both."

Over the past three years, AI for Dance raised four rounds, all from top-tier funds, but totaling only $150 million — less than what Moonshot AI or MiniMax raised in a single round. During the angel round, investors fought to give him $80 million; Zhang kept only $25 million. "Couldn't take that much."

When competitors threw out seven-figure salaries to poach AI talent, he told them, "Sorry, I really can't offer that." Zhang said he's started a company before and knows that survival matters more than anything. He takes zero salary and continues to co-invest in every round his company raises.

He didn't have the company build large models from day one. Instead, he entered through proven businesses like live-streamed large-class courses. Without stable data sources, clear usage scenarios, and a viable business model, building large models "is death."

"Survive" was never just a slogan. It's experience paid for with his body.

Zhang has been injured. His back still holds two steel plates and six screws. During rehabilitation, his calf atrophied, and he had to relearn walking like an infant. These experiences make him clearer than most about what resilience in entrepreneurship means.

Our interview ran from 7:30 p.m. to 2 a.m., until the entire office building went dark. He said "peak of stupidity" no fewer than twenty times — more often as a reminder to himself not to be pushed forward by past experience.

He misses the simplicity and purity of the old internet world, yet doesn't fear the noise of the new AI era. When hearing about which star founder gave themselves a million-dollar salary, or which hot AI company claimed its people could leave at 5 p.m., the Big Brother had only one thing to say:

"In ten paces, one man falls; a thousand leagues, no trace remains. The deed done, I brush off my clothes and go, hiding my name and my fame."

---

"I can do it, I'm the best fit, and no one else is doing it — so I'll do it"

Q: You started your company in March 2023, less than two years after the "double reduction" policy took effect. It was the market's most pessimistic moment — several leading online education companies had seen their stock prices fall over 90% from their peaks. Where did you find the courage to re-enter this industry?

Huaiting Zhang: On a day in late February 2023, I suddenly received a news link from a friend — Huiwen Wang announcing his $50 million startup with capital in hand. He asked me, "Huaiting, don't you have any ideas?" I said, "What could I be thinking?" He said, "I think you could do an AI startup too. If you jump in, I'll be your first angel investor." He knew I wasn't in great shape then, searching for my next chapter in life.

So I called Mu Li, my former Baidu colleague (former Amazon principal scientist, co-founder of BosonAI), to confirm two things: Are the model capabilities there yet? Is the combination of AI and education worth doing?

If yes, do it immediately!

Looking back, if I hadn't decisively entered at that moment, and waited another six months, some resources might not have been available.

Q: Did anyone advise you that AI was fine, but don't do education?

Huaiting Zhang: Many investors asked me, why choose something so hard, so grueling? I said the reason is simple: one, the social value is enormous; two, I can do this; three, I'm the best person to do it. Since no one else is doing it, I'll do it. That's the best opportunity.

Actually, entrepreneurship is always a march toward death from day one. It's not that good environments guarantee success, or bad environments guarantee failure.

Q: Did you get some information that policies would change?

Huaiting Zhang: I judged that the space for AI education would open up. What does great power competition rely on? AI. AI competition is talent competition. Talent competition is competition over training paradigms and training efficiency. And AI can precisely break the education industry's impossible triangle — large scale, high quality, low cost.

Q: I heard your first round, investors offered $80 million, but you only took $25 million.

Huaiting Zhang: Fundraising was originally supposed to take two phone calls. First I spoke with Chang Chen (founding partner of Gaorong Capital), who said Gaorong must lead. Then I called Gu Kai (founding partner of Ferryman), who said whatever I was doing, save him allocation. Adding my own investment, that was basically enough.

Yusen (Dai, managing partner at ZhenFund) had been discussing AI + education opportunities with me all along. When he learned I was starting up, he immediately decided to invest. Now the allocation was oversubscribed. Then Jui (Chan, managing partner at BlueRun Ventures) and Terry (Zhu, managing partner at BlueRun Ventures) suddenly invited me to dinner, saying they wanted to lead. I said I was truly sorry, but this round could only go to Gaorong, an old supporter. Runxin (Yang, investment VP at K2VC), who had been following Liu Wei's entrepreneurial moves, also insisted on participating. Soon Qingsheng (Zheng, partner at HSG) heard the news too. I had participated in Ostrich Club run by HSG and ZhenFund, so I suggested they join in the second round since the first was too small. But HSG was insistent — as co-organizers of Ostrich Club, per agreement they were to split allocation equally with ZhenFund.

Q: And your first round procedures weren't even finished before you started raising the second round.

Huaiting Zhang: Right. Xi Cao (founding partner of Lishi Capital) heard we were starting up, talked with me for over an hour, and gave us an offer with post-money valuation doubled. So we just signed both round one and round two agreements simultaneously.

Q: Why did so many investors fight to give you money?

Huaiting Zhang: Because investors had genuinely made money on our previous ventures. Gaorong and Qifu, for example, both saw roughly 100x returns on their GSX investments.

Q: Why didn't you think about taking more?

Huaiting Zhang: How do you value $80 million? Our founding team was only willing to dilute 20%. If we took $80 million, we'd have to be valued at $400 million, and we had nothing at the time — that sounded ridiculous, and we weren't that stupid.

Actually the third round was oversubscribed too. Alex (Zhou, managing partner at Qiming Venture Partners), Qi Hu (executive director at Qiming), and Dingzheng (Li, VP at Qiming) came to the company, decided to lead; total offers including existing shareholders were around $120 million, but we only took $60 million. The fourth round was also led by Qiming. AI for Dance is one of the largest investments in their history — very grateful for that.

Q: You've raised only $150 million over nearly three years. Moonshot AI and MiniMax each raised over $1.5 billion.

Huaiting Zhang: When I first started GSX, we didn't know which direction to go. Xiao Wang (founding partner of Unity Ventures) gave this advice: prepare enough for three shots, because entrepreneurship is like shooting — if the first shot misses, take a second.

At that time, Xiaomi held China's Series A funding record at $41 million. We ended up taking $50 million precisely to have room for trial and error.

This time, starting over, we knew exactly what we wanted to do from the start. The question now isn't having extra shots prepared, but how quickly this one shot can produce results.

---

There's no better scenario for AI落地 than education

Q: Some investors describe you as a traditional education company in AI clothing, because right after founding you started with traditional online live large-class courses.

Huaiting Zhang: If they understood, they wouldn't say that. This is just a phase.

On day one, all the co-founders wanted to build large models. Only I said do online education first. Everyone shook their heads — they'd done online education, it was too grueling! But I believed this was the only viable path: build the application first, then倒逼 the algorithm. Otherwise going straight to AI was death. You don't do real business, don't roll up your sleeves and get into the mud, you have no idea how to cross that mountain.

Q: Sounds like you didn't give them room to object?

Huaiting Zhang: I said if you want me as number one, then this is how it's going to be.

Our HR lead told me at one point that everyone was working too hard, that AI entrepreneurship shouldn't be like this. I said I don't know how to explain this to you today, because you haven't experienced it, but remember this: entrepreneurship is being swept forward.

Q: So you're not optimistic about startups building large models directly?

Huaiting Zhang: I've just been constantly推演 what core advantages startups have in building large models.

Data matters most, and truly high-quality, unique data is basically all in big companies' hands. Commercially, do you have clear applications and scenarios? If not, what does the model make money from? On compute and capital, big companies already have servers and can sell cloud services, while startups buy compute with raised funds. As for algorithms, I've done algorithms — two or three months, half a year, and you're caught up, plus abundant open-source projects make long-term leadership very difficult. Though perhaps my own understanding is still insufficient.

Q: So you also don't believe general-purpose large models can replace vertical models?

Huaiting Zhang: At least not in education.

Large language models are probabilistic models, but the underlying logic of education is "guarantee." That's why photo-based homework help has been so hard to monetize — it doesn't guarantee correctness, let alone effectiveness. If you had a child, would you let ChatGPT teach them directly?

What we're building is an "educator large model." Getting the answer right matters less than understanding pedagogy: adjusting pace based on student responses, teaching to the individual, and converging AI's uncertainty into a deterministic experience.

Q: If large model capabilities continue to leap forward over the next three years, will this path still work?

Huaiting Zhang: What's special about education is this: the same content, for different students, varies enormously in whether they can absorb it and how to teach it effectively. If a model can interact with a student long-term, understanding their level, interests, and learning efficiency, it can continuously adjust content and methods, forming a true data flywheel (the model grows smarter through human interaction), achieving true personalized teaching.

By comparison, data flywheels are actually harder in AI healthcare and AI law. Diagnostic rules and legal standards are relatively consistent for most people — individual variation isn't as large, and decent models can be trained on public data.

Q: Why are you so convinced education is the best scenario for AI落地?

Huaiting Zhang: Because it simultaneously satisfies three things: data flywheel, high-frequency usage, and commercial闭环.

At its core, large models do associative reasoning across knowledge networks, which inherently resembles泛教育 services; education covers all demographics and the entire lifecycle; learning is high-frequency, continuous, and resistant to突变 behavior, making data dense and experiences easier to stabilize; education has both strong willingness to pay and a high commercial ceiling.

Few industries can satisfy all these conditions at once.

Q: Since you're thinking so clearly, why not build both applications and models simultaneously?

Huaiting Zhang: Too idealistic. Where do the people come from? Top AI talent was commanding several million or even over ten million RMB in annual salary, and at that time large model companies were springing up everywhere, with ByteDance and Alibaba throwing money around — what did we have to attract reliable people?

Q: What do you have now?

Huaiting Zhang: Now we have business, data, and growth. But back then we had nothing.

Q: What truly remained from these three years of dual-teacher large classes?

Huaiting Zhang: First, users. We have real paying users and complete teaching relationships. When pivoting to AI, we weren't starting from zero on user acquisition — we were directly switching how they use the product. Second, teaching methodology. We now call it "star teacher AI one-on-one," and there's a reason "star teacher" comes first.

In a cold start, if you push an AI teacher directly, why should parents trust it? Only when this teacher has actually taught enough classes and developed a stable teaching method will parents accept its AI digital twin.

We now have 1 million users who have experienced our AI one-on-one courses, with a completion rate over 92% — higher than live online classes; single-session accuracy rates improved from 59% to 83%. We continuously do user回访, and satisfaction has remained fairly stable overall.

Q: When do you expect to completely abandon traditional business?

Huaiting Zhang: The first batch of users started experiencing AI one-on-one in February this year, we began scaling in July, and now quite a few users are actively choosing AI tutors. Our expectation is that within one to three years, most users will shift to "AI one-on-one tutors."

Q: If you had to summarize in one sentence, what will AI turn the education industry into?

Huaiting Zhang: Turning a service industry into manufacturing. That is, providing personalized services at scale while guaranteeing quality — with the potential to break the long-standing impossible triangle in education: large scale, high quality, low cost.

Q: Your company name is quite unusual. Why "Yu Ai Wei Wu" (Dancing with Love)?

Huaiting Zhang: Education is great love; AI can also be read as "love." We hope love and AI dance in harmony, with humanities and technology complementing each other.

We are a native AI technology company, hoping to use AI to achieve tech for good and education for all, erasing what may be the most unequal thing among humans — the cognitive gap. But we also know technology is a double-edged sword; AI needs "great love" too, like martial arts practitioners in wuxia novels who recite scriptures to purge the violence in themselves. Including in the future, whether we can enable carbon-based life and silicon-based life to coexist harmoniously, because silicon-based life will surpass carbon-based life in many aspects.

How Do You Recruit Top Talent?

Sorry, I Really Can't Pay That

Q: Everyone is scrambling for top technical talent now — some major companies can offer annual salaries exceeding 100 million RMB. But I've heard many technical talents who come to you actually have to take pay cuts, from tens of millions a year down to hundreds of thousands?

Huaiting Zhang: How could I possibly afford that?

I've founded companies before; I know surviving matters more than anything. If I spent all my investors' money today and the company disappeared tomorrow, wouldn't that be humiliating?

Q: If there were a truly brilliant expert willing to join, and they wanted a 10 million salary, you still wouldn't give it?

Huaiting Zhang: Sorry, really can't pay that. How much have I even raised? Ten million for one person a year, ten people is 100 million, a hundred people is a billion — I've only raised about a billion total.

If you were making 100,000 a month at a major company and I also give you 100,000, then shouldn't I also give you equity? Theoretically, no. But startups are obviously riskier. If you're unwilling to trade lower cash for higher uncertain returns, that means your risk expectations are already calibrated to a large company's level.

What I can offer is: take less today, in exchange for tomorrow's outcome. If you can't accept that, it's a stage mismatch. I've rejected many such people. Gotta make sure the company survives first.

Q: When salary is clearly not your advantage, how do you compete with large model companies for talent?

Huaiting Zhang: We're not competing on the same dimension. They're building general-purpose large models; we're doing vertical industry, with proprietary interactive data in education and high-frequency application scenarios, where models' self-feedback loops are faster.

Also, after the shakeout of the past two years, many people have a clearer picture of the competitive intensity and uncertainty at large model companies. In AI education, we have a chance to reach a globally leading position.

Q: Do you actually say this to them?

Huaiting Zhang: I do. For some top candidates we've spent over half a year, with cumulative communication exceeding dozens of hours, until they finally couldn't resist and joined. Recently, a well-known foundation model lead from a major company joined us.

Q: We've interviewed Zhilin Yang and Junjie Yan — they represent two different talent philosophies in the industry. Yang believes you must recruit geniuses; Yan believes you recruit 80th-percentile people and manage them well. Which are you?

Huaiting Zhang: Closer to what Yan said. Maybe I'm just an ordinary person, not a genius. What's most important in a team isn't individual heroics, but people having each other's backs.

Q: During your time at Baidu's Fengchao, you worked alongside technical talents like Dong Zhang (founder of AiMi Learning), Hua Su (co-founder of Kuaishou), and Wenyuan Dai (founder of 4Paradigm). What's your management experience?

Huaiting Zhang: First clarify the goals, then try not to interfere with them. Truly capable people need space — don't micromanage.

We once had a genius engineer who liked sleeping during the day and working at night, but could crack the hardest problems. So we let him work on his own rhythm. Management can't be one-size-fits-all; it has to vary by person.

Q: Why couldn't Baidu retain these people, or you?

Huaiting Zhang: Similar reasons — saw opportunities, but couldn't execute them internally.

The day I resigned, Yiming Zhang asked to meet and posed one question: "How many organic results between commercial ads is optimal?" I said, one in four is already the limit — that was实战 experience.

In 2013, mobile data was still relatively expensive. During those years I pushed a project internally at Baidu called "Butterfly Plan," hoping to partner with carriers so users could browse news in the Baidu App without data charges, with us covering costs through ads. News is a刚需; once habits formed, we'd own the mobile entry point. But the project was ultimately rejected. I even told Yiming at the time, if Baidu actually did this, Toutiao would feel pressure.

Q: What's the easiest pitfall in managing technical talent?

Huaiting Zhang: Managing everyone the same way will definitely cause problems.

Engineering teams have high certainty and can be managed more tightly; people doing strategy and algorithms have low certainty and must be given space, but can't be left completely without pressure. For top algorithmists like Bohai (Yang Bohai, 2005 ACM-ICPC World Finals champion) and Shixi (Chen Shixi, who together with Tiancheng Lou was known as "Tiancheng in the North, Shixi in the South"), the most important thing is giving enough space — don't get in their way.

On Management: Hire Slowly, Move People Fast, Use People Hard, Treat People Well

Q: You're the only AI company we've seen that scrolls messages on conference room screens saying "Please turn off lights when leaving," "Push chairs back in," and "Take your trash with you."

Huaiting Zhang: I personally lead everyone in pushing chairs back after every meeting (laughs).

Many people consider themselves high-level talent, but don't do high-level things. I want everyone in the company to become someone who "thinks of others," not just someone who says I'm awesome, look at me. Our values are "love yourself, love your partners, love the world," and the first rule of loving your partners is: kindness matters more than cleverness.

Q: We've talked to quite a few AI companies, especially technical founders, who feel a bit embarrassed discussing culture and values at an early stage — but you seem completely unbothered.

Huaiting Zhang: Give it time.

What truly binds people together in a company is the soil, and the soil is your culture — including mission, vision, and values. If you want to build a great company, you must have your own culture.

We started researching mission, vision, and values at the same time we began discussing our entrepreneurial direction. Give it time and you'll know its power.

Q: We once interviewed a star AI founder who, when asked why his company had no co-founders, replied, "Do I need co-founders?" But you not only have four co-founders, but two founders (Huaiting Zhang, Wei Liu)?

Huaiting Zhang: I'm just an ordinary person, and entrepreneurship is a march toward death — you need a group of partners with high caliber, complementary abilities, and aligned values working together to have a chance.

Q: How were they all recruited?

Huaiting Zhang: We'd worked together for years, so the trust was already there — basically, one phone call and they were in. For Wei Liu, I called him for ten minutes. Then we called Huiyan (Wang Huiyan, head of product) and Guixing (Liang Guixing, head of growth) separately. Together, we got Wang Lin (head of technology), who was far away in Hangzhou, to come back to Beijing for a chat.

At the time, Jianbiao (Ji Jianbiao, head of operations) had just joined a company valued at 2 billion yuan as a partner two weeks prior, with an annual cash income in the millions and 3-5% equity. When he learned we were starting something, after a brief conversation, he didn't hesitate to give up everything and join us.

Q: You've shared before that your standards for hiring are: hire slowly, reassign quickly, use people intensely, treat people well. Which of these is hardest to put into practice?

Huaiting Zhang: None of them are easy. "Hire slowly," for instance, means considering factors comprehensively enough — not literally moving slowly. If you hire the wrong person, the cost is very high. "Reassign quickly" means separating personal feelings from positional decisions, and having the courage to make swift adjustments when you spot problems.

Actually, these four phrases need to be followed by two more: "set rules" and "build consensus." Before making any management decision, you need to define the rules clearly and establish shared understanding. Then there are two more: "high targets" and "frequent coaching." Targets must be set high, and coaching must be constant.

Q: What's the most important management lesson you learned from building Gaotu?

Huaiting Zhang: Partners must be top-tier. Only then can the organization stay healthy, and only then can the founding team manage it as it scales. If partners are limited in capability, on one hand it's hard to attract excellent talent — you end up with B-grade managers hiring C-grade people, and talent density collapses.

The ability to lead ten people, a hundred, a thousand, or ten thousand are completely different things. When an organization grows large, if you lack experience leading teams at that scale, serious problems will arise. By then, if you need to replace them with more senior people, redistributing equity becomes another major challenge.

Q: How do you prevent partners from becoming organizational bottlenecks?

Huaiting Zhang: Our current co-founders have all led teams from several hundred to tens of thousands of people, so they won't be bottlenecks today. Besides, AI companies don't need tens of thousands of employees.

On Competition: Watch the Sharks, But Also the Hungry Wolves

Q: Breaking through in AI entrepreneurship today feels harder than in the mobile internet era — first, AI is consensus, not non-consensus, so the giants are running alongside you from the start; second, you're building from zero on people, capital, resources, brand, and business.

Huaiting Zhang: We used to say that when starting a company, you need to watch both the sharks and the hungry wolves. Sharks are the giants; hungry wolves are fellow entrepreneurs like yourself. But today, building in education is particularly difficult.

Q: Who do you see as the most competitive shark?

Huaiting Zhang: For now, it's still those major online education incumbents. Of course, we have some differentiated competitive advantages. In the AI education track, there are very few who understand both models and education and have actually delivered results — AI Weiwu is relatively rare in that regard.

Q: Several leading online education companies started laying out AI very early, but progress hasn't been particularly fast. Why?

Huaiting Zhang: That's not for me to comment on others. For a startup like us, the only path is to be sufficiently focused and sufficiently fast — there are no other options.

We're a native AI technology company. At the very beginning, I first judged whether the AI technology inflection point had arrived, and only then considered which industry would be most suitable — that's how we landed on AI + education. At the technology level, we benchmark against top technology companies, not traditional education companies.

In China, driving bottom-layer technology evolution through concrete business applications tends to be more realistic. ByteDance's recommendation system, Alibaba Cloud, Baidu's Kunlun chips — they all emerged this way. We also hope to gradually accumulate end-to-end multimodal capabilities through real teaching scenarios.

Q: Then what's the fundamental difference between traditional education companies + AI versus native AI education companies?

Huaiting Zhang: The difference isn't actually in attitude toward AI, but in system architecture.

Traditional companies rely on stable processes and prefer to hire "experienced people." AI-native companies are the opposite — old processes are failing, and every link from growth, conversion, service to renewal needs to be rebuilt.

Take another example: in traditional internet companies, the product workflow is linear and waterfall-style — research, development, validation, the whole process is long, trial-and-error costs are high, and product managers get bogged down in documentation and inefficient communication. But now with AI, you can complete deep research in minutes, generate an interactive MVP in 1-2 hours, and eliminate most problems through rapid iteration before development even begins; what gets handed to engineering is a code repository with working logic, so engineers don't have to build from scratch and can launch in 1-2 weeks; product managers also transform from "document writers" into "product builders" focused on creativity and product logic.

Q: What core technology does AI education truly need? Where are you now?

Huaiting Zhang: Essentially it's a full-stack capability built around teaching scenarios.

It's not single-point technology, but deep integration of digital humans, voice, large models, and engineering: digital humans need to approach real human interaction; voice needs to adapt to real classrooms with multi-emotional expression; models need to cover education verticals across multiple subjects; on the engineering side, you need to support large-scale concurrency with extremely low latency.

Q: There are already quite a few top voice model teams on the market, like Doubao, Minimax — what makes you think you can do better?

Huaiting Zhang: Because we're doing voice for education scenarios. We have dedicated data training models to understand subject-specific terminology, judge whether phonetics and words are pronounced correctly, and handle cases where children's pronunciation is incomplete. Generalized models don't need to consider these things.

Q: When do you think AI education will explode, and what will be the sign?

Huaiting Zhang: I don't think it's far. The sign is when it gets discussed and used at scale.

Veteran Entrepreneurs Don't Take Detours, But They Have Blind Spots

Q: How do you see the difference between the internet world of a decade ago and today's AI world?

Huaiting Zhang: The world back then was probably more simple and pure. Now it's very noisy — I've even heard of entrepreneurs who, after raising funding, pay themselves monthly salaries of 80,000 or 100,000 yuan. For every round we've raised, I've quietly invested my own money alongside.

Q: Some investors have described you as a typical "big brother" entrepreneur — wealthy, well-resourced, charismatic, and also seasoned with experience.

Huaiting Zhang: I'd say I'm a veteran entrepreneur. And even with all that experience, no one can guarantee every venture will succeed.

Q: What's the biggest difference between "big brother" entrepreneurship and "prodigy" entrepreneurship?

Huaiting Zhang: Most likely, you won't take too many detours. What happens to most companies after two and a half years? Direction needs adjustment, path needs adjustment, most likely many early employees will leave, co-founders may fall apart. AI Weiwu has been around for two and a half years, the co-founder team has been very stable, and neither direction nor path has changed.

Q: In mid-2024, the narrative that "large model progress is slowing" became popular. Under external pressure, startups started buying traffic, building more product types. Of course now they've all collectively reflected and returned to technology.

Huaiting Zhang: They took a lot of investors' money, and were spending it quickly too, so they may have faced pressure from investors. Our investment terms have no valuation adjustment mechanisms, no repurchase clauses, and our investors haven't asked me to do anything — they trust our judgment.

Of course, saying this I may be standing at the peak of Mount Stupid.

Q: You've said "peak of Mount Stupid" many times today.

Huaiting Zhang: Every time something happens, I immediately reflect on whether I did it right or wrong.

Q: You've been through two entrepreneurship cycles. What's the biggest difference between entrepreneurs today and your generation?

Huaiting Zhang: Theoretically, entrepreneurs should get stronger generation by generation, but the difficulty keeps rising too.

Mobile internet-era founders fought their way through an environment already crowded with giants. AI-era founders need to understand both models and industry; find balance between model uncertainty and business fault tolerance; and build entirely new AI-native organizations.

More practically, AI entrepreneurs start out comprehensively behind giants on brand, resources, capital, talent, and scenarios. Competing under these conditions, the difficulty is compounded.

Q: You met Yiming Zhang in 2014, and later competed against him in online education. How do you see his evolution?

Huaiting Zhang: When I first met him in 2014, we talked for three hours — he was mainly asking how to do advertising, and he was still quite green then. But it's like machine learning: where you start doesn't matter, what matters is how quickly you can find the direction of gradient descent, and the speed of that descent. The ultimate difference between people lies in which path you choose, and the acceleration that path itself gives you.

When I saw Yiming again in 2020, his level was already far above mine. He'd been through so much in China — competing with Tencent, competing with Baidu, and also facing international challenges.

Q: War can make a person grow rapidly.

Huaiting Zhang: People are forged through battle. My partner Wei Liu, when he joined Genshuixue in 2015, was just a frontline colleague responsible for business analysis. But then he was called up in a crisis to explore the online large-class model, built Gaotu Classroom from 0 to 1, grew revenue from 20 million to 6.4 billion in four years, and expanded the team from 7 people to 25,000. He then went through the 2020 online education war and the 2021 "double reduction" policy — his growth speed was witnessed by the entire industry.

Q: Among all the successful entrepreneurs and investors you've met, who do you want to become?

Huaiting Zhang: I still want to build enterprises. As a child I loved reading novels by Gu Long, Jin Yong, Liang Yusheng, Wen Ruian — if I'd lived in ancient times I probably would have practiced martial arts; today, the tech industry is another jianghu, and technology is the martial art. I very much love what Li Bai wrote in "The Knight-Errant": "One man killed within ten steps, leaving no trace for a thousand li — when the deed is done, he brushes off his clothes, hiding both deed and name." I think that's a wonderful realm.

Q: Veteran entrepreneurs don't take detours, so what might be the blind spots?

Huaiting Zhang: The blind spot is when you treat your previous successful path as a dependency, you become unable to see potential problems.

For example, we used to believe that the organizational capability to manage large teams was very important — this was Gaotu's core competitiveness. But in the AI era, organizational capability probably isn't about managing large teams, but about how to use AI to enable human-machine collaboration, and operating the enterprise as a self-evolving intelligent agent.

From Baidu to GSX: What's Fake? What's Real?

Q: Every time you've faced a career crossroads, you've chosen the non-mainstream path. In 2005, for instance, you turned down Microsoft and IBM for Baidu — the lowest-paying offer, at a company of just a few hundred people.

Zhang Huaiting: Baidu was the only pure internet company among my offers, and the interview scene there left a deep impression on me.

Baidu's office had large expanses of transparent glass. I was in the middle of my written test when I looked up and saw Mengqiu (Wang Mengqiu, now founding partner of Qingliu Capital, then Baidu's senior technical director) with short hair, wearing slippers, leading a group of engineers clattering past. My immediate thought was: "Wow, so relaxed, so tech — I like this place!"

Q: A decade later, Baidu had reached its peak, yet you quit with nothing lined up.

Zhang Huaiting: I'd started several internal projects, each of which I handed off halfway through. This happened three or four times, until I finally thought: forget it, I'm done, I'll go out on my own.

In March 2014, I took my team to Baidu's US research center in Silicon Valley, and dropped by Google, LinkedIn, Apple, and Microsoft to catch up with friends. I was 37 that year, and found everyone talking about new things, many of them starting companies. I asked myself: why am I still at Baidu? Because the pay is good, the income stable, or because Baidu can carry what I want to do with my future? Two weeks after returning from Silicon Valley, I handed in my resignation.

At the time I didn't know what I'd do next. I just told myself: if you don't cut off your retreat, you'll never set out.

Q: I heard Yiming Zhang came to see you the very day you resigned from Baidu?

Zhang Huaiting: Yiming's purpose was simple. ByteDance wanted to build an ad recommendation system, and Phoenix Nest was the most advanced at the time, so Zhenyuan (Yang Zhenyuan, ByteDance VP of Engineering) recommended me to him for the job. The target was 5 billion yuan in annual ad revenue.

Q: In 2014, Baidu's ad revenue was roughly 40 billion yuan, while ByteDance's commercialization team had just five people at the beginning of that year. What was your first reaction to Yiming Zhang's 5 billion?

Zhang Huaiting: I thought it was doable.

Q: Why didn't you join ByteDance?

Zhang Huaiting: For me, wouldn't that just be going back to the same thing?

Later Larry (Larry Chen, founder of Gaotu) approached me, hoping we'd do online education together. For me, starting from zero was genuinely challenging; also, after years of traffic monetization, I wanted to try something different, and education was deeply meaningful. In the end, I heard it was Yuqiang (Chen Yuqiang, co-founder of 4Paradigm) who built the first version of ByteDance's ad recommendation system.

Q: Do you remember how you felt when your own entrepreneurial moment finally arrived?

Zhang Huaiting: Only after leaving did I realize that past success came from the platform's power, not my own abilities. Many people at big companies feel they're quite capable — try starting from zero and you'll find out.

Q: After leaving GSX in 2019, you spent time investing and attending various EMBA programs. How was that period?

Zhang Huaiting: I didn't invest to make money, but to maintain sensitivity to the era and the industry.

Those years were a painful process of cognitive iteration. We'd taken GSX public in five years — making a ten-billion-dollar company didn't seem that hard. Then "double reduction" hit, followed by COVID, and suddenly I felt my understanding of the world was far off, with no direction for the future.

I went to many programs, trying to understand how the world works. Once we had to do a secondary market case analysis, only to discover there was no point — at that moment, whatever company you analyzed, the stock was falling. It was quite funny: the previous cohort's classmates would research a company and its stock would rise. Later I realized that in the face of major trends, the fundamentals of any specific company don't matter much.

Q: What was the most important lesson from everything you experienced in those years?

Zhang Huaiting: I came to understand "borrowing the fake to cultivate the real." What's fake? What's real? Reputation, power, status, money — these are fake. People chase them, but they can vanish in an instant. Take reputation: public persona collapses happen almost daily, and many founders go from golden child to dishonored debtor. The CEO has the most power in a company, but perform poorly and the board can remove you.

Q: Then what's "real"?

Zhang Huaiting: In the pursuit of these desires, you undergo much tempering. Your cognition, capabilities, and values are elevated and formed, along with the true friends who stay with you through all the hardship — these are what's real.

Entrepreneurs Never Fail

Q: Why are you here talking with us until 2 a.m. instead of in New Zealand or Singapore?

Zhang Huaiting: Why would I be in New Zealand or Singapore?

Q: You're retired, and wealthy.

Zhang Huaiting: Clearly I've returned, not retired. Former Israeli Prime Minister Shimon Peres said: if what you're going to do tomorrow doesn't excite you more than what you did yesterday, you're old.

Q: Are you that afraid of getting old?

Zhang Huaiting: The core of fearing old age is whether you're actually living with energy each day. I don't want to live like a zombie.

Q: You're financially secure. You could be experiencing and enjoying a rich, colorful life. Why do you think entrepreneurship, working 15–16 hours a day, keeps you young?

Zhang Huaiting: I think enjoying life leads to gradual complacency. From a life sciences perspective, if you're in deep thought every day, your brain cells keep generating new ones — this is what keeps people young.

Q: What's been your biggest takeaway from your second entrepreneurial venture so far?

Zhang Huaiting: I often tell new hires: are you here working for the boss, for the team? No. You're borrowing this era, this company, this organization to elevate your cognition and build your own capabilities. So whether it's the people or the experiences, everything ultimately cultivates you — the elevation of cognition and ability is what's real.

I'm the same. Building this company is, in a sense, like playing a game. You face enormous monsters one after another, and figure out how to defeat them.

Q: Today, isn't the monster facing every big and small company ByteDance?

Zhang Huaiting: In the online education era, ByteDance also launched Dali Education, yet GSX, then five years old, still listed on the NYSE.

Q: So what's your next monster?

Zhang Huaiting: I think the next monster is whether we can make AI education truly world-class.

The internet era was high-frequency beating low-frequency. The AI era is high intelligence directly crushing low intelligence — like when DeepSeek emerged, many large model companies pivoted. I'm now wondering: could some company suddenly emerge with AI education that far surpasses us in cognition, pedagogy, personalization, reliability, safety, and experience?

Q: You said many entrepreneurs have their obsessions. What's yours?

Zhang Huaiting: My obsession is this company — I want to make it work. I tell the team: if we're not the global #1 AI education company, I at least hope we can make top three.

Q: From our conversation, I'm certain you want to do something big, but I don't seem to sense any mission-driven passion for education itself?

Zhang Huaiting: How does one demonstrate mission — does it require a moving story? Does actually getting things done count?

Q: Your colleagues' highest praise for you is "resilient." What do you think entrepreneurial resilience means?

Zhang Huaiting: Make decisions facing toward the goal, not toward the difficulties.

When we first founded GSX, I held 22% of shares, but only 6% at IPO. Those dozen-plus percentage points all went as incentives to later partners — from a human nature perspective, who would do that willingly, without complaint? But back then the company had limited cash, yet needed to motivate partners to keep going, while not wanting investors to take losses. There was no other choice but to give away my own shares.

On the eve of Spring Festival 2017, the founding team discussed that we couldn't use the company's limited cash for year-end bonuses. After much deliberation, I went to the bank at noon without telling my family, withdrew my own savings in cash, carried it back to the office, and stuffed 1,000, 3,000, 5,000 yuan into red envelopes to hand out at the afternoon year-end meeting. At the time I just felt: if these teams don't get bonuses, they'll fall apart.

Q: Why can you do what most people can't?

Zhang Huaiting: Probably resilience built from being an athlete. As a child I was physically weak — gymnastics at 4, swimming at 6, long-distance running and soccer from 10. I even became a national third-class athlete in swimming, receiving government subsidies.

In 2002, the year before my graduate studies ended, I injured my back and had major surgery — two steel plates and six screws put in my body. After discharge I needed eight months of bed rest, unable to even turn over, drinking water only through a straw. To graduate on time, I had to write my thesis lying face down every day. My parents worked, so no one was home during the day — I'd lie prone from 8 a.m. to 6 p.m., bathroom trips inconvenient, so I drank as little water as possible. When I could finally get up for recovery, my leg muscles had completely atrophied; I had to relearn walking from zero like an infant. My back muscles had also adhered to the steel plates — every time I sat down and stood up, it tore, then healed, then tore again. I got through it all in the end.

Q: What do you most want to achieve in your second venture?

Zhang Huaiting: Education for all without discrimination, teaching according to aptitude — completely transforming an educational paradigm thousands of years old.

I've always believed the greatest inequality between people is the cognitive gap, and the greatest charity is leveling everyone's cognition. But for thousands of years, our education has been either one-to-one or one-to-many, either unscalable or poor quality when scaled. Changing the educational paradigm for our descendants — how monumental is that?

Q: You said your first venture was to prove your capabilities. What do you want to prove this time?

Zhang Huaiting: Nothing. I think I'll retire once this company is doing well — best if nobody pays attention, and I go do other things. That would be most comfortable.

Q: What counts as "having made it"?

Zhang Huaiting: If you go by the世俗 definition — going public, making money, growing the company's influence — that's one way to say you've made it. But to me, what's more important is the self-understanding, awareness, and personal growth you gain along the way.

Q: Are you afraid of failing at entrepreneurship?

Zhang Huaiting: Entrepreneurs never truly fail, because everything they need is already grown into them through the process of building something.