An Interview on Building AI Frameworks and Life Choices | Original
Source: chatgpt.com
A: Zhiwei Li, Software Engineer working at the intersection of AI and finance.
B: Shawn Shao, Software Engineer working on AI frameworks, agents, and MCP at an international bank.
The following conversations were transcribed using the Google Cloud Transcription v2 API, then refined and organized with the help of AI tools. The original conversations were in Chinese.
Table of Contents
- Recording, Archiving, and Views on AI Media
- Archive conversations for future reflection
- Video and image AI have more potential
- People prefer video over text consumption
- Personal example of monetizing content creation
- Career Beginnings, Certifications, and Early Struggles
- Interest drives continuous learning and curiosity
- Pursued rare GCP certifications hands-on
- Certifications helped secure international bank role
- Overcame academic challenges through self-learning
- Work Experience, Family, and Career Choices
- Promoted multiple levels at international bank
- Career decisions consider family and roles
- Prefers stability over geographical mobility
- Values new knowledge impacting daily perspective
- AIPlayer, Agents, and Value Creation
- Building company-level intelligent agent platform
- Focus on solving real business problems
- Creating value through practical AI applications
- Not pursuing fan traffic or monetization
- Bosses value problem-solving over external fame
- Critical Thinking, Yin Wang’s Influence, and Worldview
- Adopted critical worldview from Yin Wang
- Questions established scientific narratives
- Believes major historical events are fabricated
- Learning should be joyful and process-oriented
- Time management reshaped by AI efficiency
Recording, Archiving, and Views on AI Media
A: Can I record it and then anonymize it later? We’ll organize the names a bit, then send it to you for review. After you review it, you can delete the parts that are okay with you. Posting it to my blog is just one consideration, but it will definitely go through your review first. Actually, my main purpose is to archive this content.
I’ve already done some initial organization using Google Cloud. Last time I visited an American PhD friend, it took five hours to sort it out with Google Cloud and Claude Code, but now thatit only takes 10 to 20 minutes. Since that’s the case, we might chat about this again in a few years. What I mean is, looking back at these conversations later, we might discover which predictions about the future were correct.
B: I wouldn’t dare to predict the future.
A: But regarding AI, I chatted with a friend from Tsinghua, and he thinks text has reached its peak, while images and videos have more room for development. Because images and videos require more computation, and unlike text, they’re not as easy to generate autoregressively. For videos, things like lip-sync and audio, for example with ffmpeg—modern ffmpeg has many steps and points to explore. For images, various shadows and details, and the products will be richer too.
Generally speaking, videos are more like TikTok or Douyin; people are increasingly impatient with reading text and prefer scrolling news in their familiar language. For example, my mom started a work account and recently earned 900 yuan in a month with 3 million reads. She’s been writing on WeChat public account for 8 years, persisting with a lot of things.
B: What you just said, I think it’s really true. In the past, you couldn’t change those things. You think that for many things, just buy the current best or make the current best decision. Any decision can only be based on your current understanding.
Career Beginnings, Certifications, and Early Struggles
A: What I’m thinking is, you’ve definitely experienced some painful moments, like taking the college entrance exam at 18; this decade must have had some tough times. But you’ve steadily made it to where you are now, with strong beliefs. At your age, many people are still doing outsourcing work.
Besides your AIPlayer, we could use MCP to integrate the company’s Jira and Confluence interfaces. It doesn’t seem that impressive, but there are many tutorials, and many people are doing it. Then you implemented this at the international bank, and the information you brought is really useful to everyone. Catching this wave is important. If you went to do a very Java-based microservices framework for the company, it might not be as popular.
B: So what’s the question?
A: The question is, have you experienced some tough moments? Like from Chinasoft to a unicorn Yeahmobi, then to the international bank. These past five or six years, you’ve been working very hard too. When you joined Chinasoft, you led a team of thirty people, especially right after graduating from Jiangsu University of Science and Technology. We’re not from 985 schools; especially in the first five years of your career, you must have worked much harder than before. For example, back when we graduated, we might have gone to Guangzhou to play for a few hours. Now, however, we really cherish the time we have to talk here in the hotel. I feel like why are you more diligent now than before?
B: First of all, interest is the best teacher. Whether you agree or not, I think interest is the best teacher. The root of interest is that you want to know what new things are, and seeing new knowledge gives you an impulse to learn about it. So no matter which company I’m in, I’m full of curiosity.
For example, the unicorn Yeahmobi in Xi’an. The company really encouraged everyone to pass GCP; many colleagues said GCP can’t be accessed in China without VPN. The previous company wanted to cooperate with Google. For Google’s cooperation, you need a certain number of professional certifications. Many colleagues weren’t willing to take the exams, but I was. I think it’s new knowledge, why not learn it? Google’s technical framework and capabilities are there. So I went to learn it and passed. The company even paid for everyone to take them; I took three.
A: You’re born in ‘91?
B: ‘90.
A: OK, so in 2019 you were 29, I’m 30 now, ‘95.
A: How to put it? Because half of high school was spent on competitions, I didn’t get recommended admission through competitions, but didn’t do well on the college entrance exam. I wanted to go to 985 like Beihang or Sun Yet-sen university, so it led to one wrong step after another. Later I went to a startup, joined a startup company, and dropped out. Sigh, dropping out—at the time I thought I could make a living with coding skills, a bit unaware of how society works. Later, many Chinese companies require degrees. OK, turn back to you, then you went to take exams; you took a lot of certifications, Coursera ones, and for company topics, you practiced very seriously.
B: Yeah, at that time I didn’t even have any practice questions; I just followed the official documents and did one experiment after another. I did it seriously without copying from online materials. There were no real questions or question banks back then; I had no idea about those rote-memorization problems. So I really ran through the GCP products hands-on. I closed myself off with the official docs, reading them carefully one by one.
Now with AI, it’s very convenient, and many people are faking it in this area, I know. Even when the company was recruiting for Xi’an office of the international bank. When they saw I had three GCP certifications at once, the interviewer was stunned. Because GCP was just starting to push then, not many people had taken it; it was rare. I had three at once, so the interviewer definitely suspected if they were real, but when grilling me on details, I answered well, so they kept me. Later I joined the international bank.
Work Experience, Family, and Career Choices
A: At the international bank, you got promoted for several levels, three levels, right? Before that, you started as Senior Engineer. Then the question came; I feel like around 2014-2015, you did a lot of risk control work.
B: Before, I was doing these risk controls, writing some fixed rules to check things—simple, right? Because that’s all there is to it; our big data stuff still has technical depth, can optimize networks for risk control.
A: I’ll definitely send that to you later; in the future, if that young guy wants to consult you, you can share these things with him. Now today, I think everyone has their own different experiences. This year, have you felt particularly busy? Because of AIPlayer or whatever, do you feel strongly needed, that kind of feeling?
B: That does happen, because some colleagues aren’t familiar with AI and often come to you if they think you’ve done well; they’ll want to ask. Sometimes there are so many questions that we built this group, built two of them.
A: Do you have plans to go abroad in the future? Someone as excellent as you, have you thought about going overseas?
B: Currently no plans; I think first look at the company’s policies. If there’s an opportunity, of course it would be good to explore, right? After all, paying attention to the different points of the world.
A: But you’ve been in Xi’an all along; university was in Jiangsu, you’ve been in Xi’an for about a dozen years now, and most of your family and friends are there. Family and friends. So some people definitely say come to Guangzhou, Beijing, the north, Shanghai, Guangzhou—why not go?
B: It’s multifaceted, because the more roles you have on you, the less you can just consider your own things as one person. For example, as a father, it affects you; as a husband, you definitely consider what your wife wants, her feelings. Everyone encounters temptations and faces many choices. But if you have choices, actually many choices you weigh with various factors. For example, considering your wife’s length. Then I think I’d reject the invite now; it’s not a simple choice. Only if I directly find it extremely good, I’d definitely go. Because many people’s decisions actually have many factors combined together.
A: Salaries in Guangzhou are about 10% higher than in Xi’an.
B: I don’t know; I don’t inquire.
B: Actually, a lot of things vary by day, like parking in Xi’an costs 3 yuan, but here in Guangzhou it might be 18 yuan, right? We have to understand these city differences. For example, in this aspect, you’ll compare yourself—like yesterday to today, today to yesterday. Has some new thing or viewpoint suddenly impacted you? I think that’s really great too.
AIPlayer, Agents, and Value Creation
A: I feel like you’re not very snobbish, not that utilitarian type who keeps comparing everything, thinking what’s the point of going to a big city. And you’re serving so many people, but it feels like you give the same answers and support to everyone—just help where you can. What about the future? In the next few years, where do you plan to focus? I heard you’re working on some agents stuff, and that MCP, and I saw those company things of yours. In the future, do you want to stay at the international bank? And for the next two or three years, and long-term like ten years, what direction do you plan to take?
B: What I’m doing now is just stuff I’m interested in. I just do it as it comes, like building a company-level intelligent agent platform. In the future, if other colleagues at the company want to use intelligent agents, they can start from our platform. But agents are all vertical and customer-facing, and solving a lot of problems takes a long time, like 20 minutes. Actually, in this, I think a lot of people are falling into a trap these days. Like with this AI, it’s fundamentally a tool. You only create business value if you truly help users create value. For example, if I say I have a super flashy AI now, it might still not sell to anyone. But if my AI can help you write emails every day or grab English emails, then it can. That’s a scenario where it solves a problem for you, and that’s what generates more value.
For me, I don’t need everything to be universal. I’m inside the international bank. Actually, the core is combining business knowledge with finance. So I think if you’re my business user and you solve problems using my platform, then I’ve created value. Our product is for the business.
A: This thing of yours is easy to promote to DBS Bank or various other banks, because their needs are same from ours. A lot of stuff is already being used with your tools and frameworks. How many people use AIPlayer to solve problems?
And your AIPlayer, for example, has two company WeChat groups. And then you’re about to start a third group. The output is fast, the momentum is already fierce—there’s already a third group! Is it like exponential growth, getting faster toward the end? Or gradual in the middle? Or in waves, and does external news also affect the growth curve?
B: I don’t pay attention to that, because I’m not going the fan traffic route. I do this AIPlayer to help me learn. It’s not for retaining fans. By doing this, I get to understand MCP and those things better, because I share charts with everyone, and I have to see some stuff to exchange for good ones. By sharing with everyone, at least it forces me to learn. Second, there’s no monetization in my fanbase at all—I don’t care about fan money. A lot of my fans are company colleagues; I just want to give them some stuff. First, you can use it for free, and I can support you in learning. Second, if you think I’m reliable, that’s enough. I don’t pursue the fan model.
A: What about your leader? Why did he promote you? Does he really understand all this work of yours?
B: I think first of all, for any boss, he won’t care how much external stuff you’ve done; he only cares if you can solve problems. If the boss has a tough issue, OK, I’ll help solve it, showing my ability and expertise. I’ll definitely leverage connections, technology, to help solve the boss’s problems. The boss doesn’t care how many fans you have, as long as you can solve problems.
Critical Thinking, Yin Wang’s Influence, and Worldvie
A: OK, and another point: last time I wanted to share Yin Wang’s blog. I’ve researched it for many years—his programming languages and articles, I look at them every few days, studied for years. From it, I gained a unique worldview. For example, reversing nearsightedness—I’ve already helped around 20 people. I’m like Ilya Sutskever; I’m not the inventor of backpropagation, but I was among the early ones helping people reverse nearsightedness, like reducing 50 or 100 degrees per year. By categorizing scenarios: one pair of glasses for driving, another for phone and computer usually, handling things with fine categorization.
Then I want to tell you the core conclusions. For example: nuclear weapons are fake. That’s what I read in the Yin Wang’s substack; I can’t teach it, but I can explain it to you. He also says hydrogen peroxide can be used for foot soaks, and Germany is decommissioning nuclear power plants. Current AI doesn’t use any quantum mechanics. Einstein’s theory is wrong because the premise of constant speed of light is incorrect; there were experimental dust interferences leading to misunderstandings. Yin Wang explains it in detail.
And he says things like learning cello, painting, various studies—he’s the best learner I’ve ever seen. For example, you look at OpenAI’s GPT; at first it seems amazing, but actually it’s nothing. Just buy a good graphics card and you can train something pretty similar.
B: No, very impressive. You know, training a GPT model—the parameters are all pretty shallow. Like, a couple million dollars can get you DeekSeek-level capability. GPT-3.5 cost over $56 million. The base model is one layer, but more importantly, it’s the applications. We make an application that can really be used.
A: Ah, what you’re talking about is this concept—it’s about CUDA, Transformers, and attention computation, plus what OpenAI says about the underlying stuff. There are some big shots there—one at OpenAI working on CUDA optimization. And now there’s reinforcement learning—how it reduces the need for heavy supervision in reasoning, and how it boosts math capability. Then there are models focused on code, since code instructions are different from natural language. Recently there’s the Grok Code Fast model—Elon Musk put a team together for that. Although Musk hasn’t really done much hands-on AI in the past couple of years—it’s mostly others building the tech, while he secures funding and does PR. But still, he has the vision.
B: Yeah yeah, impressive. Yeah yeah.
A: What I want to say is, I question these things. I think few ordinary people dare to question everything the way Yin Wang does. He says the Turing Award isn’t a big deal—that computers were invented collectively by many people. Just like at the international bank—others might idolize you today, but having walked the path yourself, you realize it’s nothing special. Um. Like Turing is just like Andrew Yao or Lou Tiancheng from Pony AI. Yeah. Yin Wang said long ago that AI hadn’t really progressed in the image field: because of lack of large-scale data and the fake nature of the top-5 accuracy test.
B: Yeah.
A: Back when I was hustling startups in Beijing, I added lots of big shots on WeChat: the co-founder of SenseTime, the co-founder of Megvii, the cofounder of quantum bit—I even visited some of their companies and knew them personally. In my early 20s, I ran a 30,000-member programmer community, with annual income of 3 million, leading a team doing small software projects. But later, because of housing prices, the real estate bubble, and free speech issues, I decided to give up domestic resources and focus on going abroad.
With autonomous driving—Yin Wang pointed out problems years ago, like cars misidentifying water reflections in the rain, or Tesla’s 2016 white truck crash, or even Xiaomi’s recent incidents. He even wrote blog posts educating Musk about such issues 8–10 years ago. Yin Wang questions everything.
Yeah yeah. Uh, I also had contacts from Machines with Heart and QuantumBit back then. Oh.
During the pandemic, I got the vaccine—two shots—then later found out AstraZeneca really had issues. Yin Wang had figured it out earlier, because he stayed up researching tons of overseas material at night. The inventor of PCR testing, Kary Mullis, himself said PCR can’t diagnose diseases; it only amplifies what’s there.
Yin Wang’s learning method impressed me a lot. He also says the atomic bomb is fake, humans didn’t land on the moon, and Musk’s rockets are mostly propaganda. Even Tesla’s “autonomous driving”—many people think it’s fully automatic, but in reality it’s still conditional. True unmanned driving is still far away—technically, legally, ethically—it’s not ready yet.
I also saw Yin Wang write that many AI applications are hype. He’ll dissect papers in detail, pointing out the experimental setups, data sources, and logical flaws. I never realized someone could question so deeply. Later, I understood—that’s the real way to learn. So my worldview shifted: I watched so much of Yin Wang that now I look at everything critically.
Like space photos: the cabin interior shows details, but outside there are no stars—just a black background. Look it up—whether NASA, SpaceX, or Chinese space agencies, it’s always the same. I’m convinced humans haven’t left Earth yet.
Historically, I really admire people like Galileo and Newton, but I think Einstein’s stuff is fake. Yeah. And SpaceX too.
B: Um, first off, I think questioning these issues is good—definitely very good. But are you enjoying the process?
A: Uh, yeah, because nothing feels that impressive anymore. One day, if I work hard, I can be like him too. Right. All those big shots—Andrew Yao, Lou Tiancheng—they’re not that impressive either, because Yin Wang breaks everything down. He’s got three PhDs, now at F5 (was at Intel as staff engineer before), super successful at 40, making probably millions in Japan. So this process? I’ve really learned a lot.
Doubting everything is hard—like starting AIPlayer, or challenging myself to be number one at the international bank. You don’t expect visible progress in the first few years.
B: Don’t chase huge goals. Do what you like, take each step seriously. Where’s your real interest? Are you willing to stick with it? Like AI—others might stay up gaming all night, but you keep grinding. You should enjoy the process; success doesn’t matter—you’ve already won by engaging. Our time and knowledge are limited; we can’t dive into everything. But if you’re interested, exploration itself is joyful—it makes you feel like a kid again. Human thinking itself is meaningful happiness.
For a career, if AI suits you, go for it. If not, just treat it as entertainment. Time is limited—invest in what makes you happy. AI made me rethink time management: save an hour today, build more tomorrow.
A: Speaking of creativity—remember you recommended Codex? I tried it—really good. I even bought the membership. I switched to Codex for integrated agents and design patterns. In some cases, it’s better than searching for custom models elsewhere.
B: Oh oh oh, it’s 12 AM already. Do you have any more questions?
A: Nothing, goodbye.
B: Bye.