At Valeri’s invitation, I’m writing about what I have seen and thought about during my nine months in Silicon Valley. I graduated from Peking University’s School of Physics in 2025 and am now pursuing an MS in Electrical Engineering at Stanford, working primarily in AI.
Being at Stanford feels like entering a vast tropical rainforest. I look up in wonder at the intricate ecosystem that sustains innovation, then look back at myself and wonder how to find a place in it. This journal offers a few glimpses of that ecosystem and compares my experiences at Stanford and Peking University. I hope they give others something to think about, too.
I have been at Stanford and in the Bay Area for less than a year. My view is inevitably incomplete, and many of these judgments will need to be tested and revised through time and conversations with more people.
1. Coupa Café
Cafés and outdoor gathering spaces are everywhere on Stanford’s campus. At Peking University, they were relatively uncommon outside a few café areas. Perhaps California’s mild weather helps: students like to study and talk outdoors. Coffee chats are a familiar part of life at Stanford and in Silicon Valley—a way to connect over academic interests or career plans. These conversations often happen in public spaces. Coupa Café, a well-known outdoor coffee spot on campus, has been the natural choice for my own coffee chats.
People in North America talk about connections all the time. I think this has a lot to do with the broader social culture. At Peking University, the Chinese word for personal connections, renmai, sometimes carried a mixture of envy and irony. Cultivating relationships to find opportunities was not always regarded with approval; people were more likely to mention it half-jokingly. At Stanford, building connections is widely accepted as important, and the phrase usually carries a positive meaning.
Learning about a research group through someone you know, getting a referral, or sharing resources seems entirely ordinary here: people stay in touch and help one another. Without those connections, many things—including finding a job or an internship—feel considerably harder.
These networks often grow out of families and schools. People seem to place considerable value on being alumni. Stanford has a directory of alumni, many of whom indicate that they are willing to help. If you have an academic or career question, an email and a Zoom call can be a starting point. University affiliation is openly treated as a form of social capital. In the campus culture I experienced at Peking University, there was more reluctance to announce that affiliation outside the university. It felt like something to be modest about, something that should remain understated.
Here, a Stanford affiliation can lower the barrier to trust. People deliberately use it to bring groups together and share opportunities. If a company already has many Stanford alumni, and a senior manager is also one, more Stanford graduates may be recruited in turn, extending the alumni network.
2. Slides from a Project-Based Course
Stanford’s CS courses broadly fall into two groups: exam-based and project-based, with some combining the two. Exam-based courses are familiar from my education in China: assignments, quizzes, midterms, and finals. I have mostly chosen project-based courses. We form our own teams, usually with fewer than four people, and select a project to explore over a ten-week quarter. There is a proposal, a milestone, and a final report or presentation, sometimes followed by a demo day, poster session, or peer review.
There is considerable freedom in what we build. Professors and teaching assistants rarely impose a narrow direction. In CS 230: Deep Learning, my project was a small research project. CS 224G, a course on building applications with LLMs, was more oriented toward users; a Bay Area founder also met with us regularly as our project advisor. Much of the grade depended on the final report or presentation. For CS 224G’s demo day, the two instructors invited numerous founders and venture capitalists to hear our pitches. For the final presentations in CS 337U: Understanding the User, a vice president of design at Google came to offer feedback. My MSEE program gives me considerable flexibility, and most CS courses count toward the degree.
From conversations with friends, I have learned that these industry guests are not just there to make a ceremonial appearance. Some friends were contacted by VCs after demo day and went on to discuss investment. Others made an impression on founders and found work opportunities, or applied for venture funding—of which Stanford offers quite a few sources—to keep their projects going. Such outcomes are surely rare; the best resources remain concentrated among a small number of people. Still, project-based courses offer a path from the classroom into the market. A good idea can keep growing beyond the assignment.
My experience at Peking University was different. I took four CS courses there, nearly all of them exam-based. Even when there was a project, the topic and requirements were usually fairly fixed, with freedom left only for the final portion. I hardly encountered courses assessed entirely through an independent project. Nor did I often see final presentations that brought in many people from industry or outside the university to evaluate and discuss the work. The educational environment felt relatively enclosed.
The difference extends beyond the projects. Many Stanford CS courses feel remarkably current. There are still courses in traditional theory, and the degree still has its requirements, but much of the material is close to the frontier. On my first visit to CS 230, Andrew Ng explained that we could study the theory through his Coursera videos, quizzes, and assignments; classroom time would be used for discussion.
One day he asked: suppose you are the CTO of a startup building a facial recognition system. How would you obtain the data, and how long would it take? That was new to me. At Peking University, I had mostly learned theory and was rarely asked to reason so practically about a specific situation, explicit constraints, and a clear need. Another CS 230 instructor gave us a different scenario: the model has broken, the VP has called you to fix it—what do you check first? These questions placed us inside a real situation and asked us to think as decision-makers. How valuable that training is remains open to debate, but I wonder whether it helps explain why some Stanford students I know are so determined to start a company and make their own decisions rather than join a large one.
Many CS courses are also taught by people from industry—tech leads and founders serving as adjunct professors or lecturers—rather than exclusively by tenure-track faculty. These are formal teaching appointments for people who teach regularly; someone visiting for a single session is usually a guest lecturer. Their day jobs may be running companies, with a few hours each week set aside for teaching. I find that fascinating. They bring current problems they are facing directly into the classroom. Their perspective includes the instincts of someone working in the field, beyond an academic research frame.
The two CS 224G instructors, for example, are Bay Area founders, while the CS 337U instructor is a Google director. This is not unusual at Stanford, but I rarely encountered it at Peking University. I know comparable courses exist there, though they seemed to make up a smaller share; transmitting knowledge remained the prevailing model. I also took courses with younger Peking University professors returning from overseas who were introducing more current material, new approaches, and exploratory assignments. Those changes felt like encouraging signs.
Some details of the coursework were striking, too. In CS 329X: Human-Centered NLP, homework 2 surprised me. Thinking Machines Lab released Tinker on October 1, 2025; two weeks later, the October 14 assignment already asked us to use it. This much-discussed new tool had entered the classroom almost immediately. We learned its framework and used it for supervised fine-tuning and direct preference optimization on an open-source model. I also used Tinker for my course project, with credits provided to support the work.
GPT-5 had only recently appeared, and Tinker was so new that even an agent struggled with its syntax and usage. Tool use in frontier models was less mature than it is today. I still had to work through the Tinker Cookbook and write the code myself. Here was a new tool that even GPT did not yet know how to use well. Looking back, that already feels like another era. AI is moving extraordinarily fast.
We could also submit completed projects to the Tinker community. I suspect Thinking Machines Lab’s support came through research connections with our professor’s lab, which then extended into the classroom. These experiences made the relationship between industry, teaching, and research at Stanford feel very tangible. Geographic proximity seems to help sustain those connections.
Classroom culture deserves a mention as well. Interrupting an instructor with a question feels normal here, though it took some adjustment for me. During orientation, someone from the international center told us that when a professor asks for questions and the room stays silent, they may wonder whether they have explained the material clearly enough. In my previous experience, silence was more likely to be taken as a sign that students understood.
Our CS 329X professor would often pause and say the classroom seemed too quiet, encouraging us to ask anything. The course had roughly a hundred registered students, with about fifty attending a typical lecture. It also had a technology ethics dimension, exploring the purpose of NLP and its human implications. Almost every class included a ten-minute hot-take debate. Three volunteers on each side would debate a topic related to the lecture for course credit. The audience voted before and after the debate to see whether anyone had changed their mind. The winning team did not have to write a report; the losing team did.
My spoken English was not especially strong, so I honestly did not enjoy the prospect of debating native speakers. But I forgot to submit the alternative arrangement for earning credit and ended up taking part in a debate about whether AI companions would make people happier. At Peking University, I had attended physics seminars with frequent, lively discussion, but those were small classes of around twenty people. That atmosphere seemed much less common in medium or large physics and CS courses. Instructors rarely paused for questions, and even when they did, students generally stayed quiet. I felt that asking a question would interrupt the class and inconvenience others, so I preferred to wait until afterward. Here, I find myself raising my hand and interrupting quite naturally. Nobody takes offense. Classroom culture makes a real difference.
3. An Unread Stanford Inbox
My Stanford inbox is a small reflection of my life here. During the quarter, I receive more than ten emails a day on average. Many announce events: VC dinners, demo days, fireside chats, guest lectures, studios, workshops, and dissertation defenses. I spend a little time each day scanning them and adding interesting events to my calendar. I will not attend most of them, but an inbox that keeps filling up makes me feel connected to the world outside my immediate routine.
At Peking University, I usually received fewer than five emails a day, mostly administrative messages rather than event announcements. Events may have circulated more through WeChat accounts and group chats. Still, the density and quality of academic and career events I experienced there felt far below Stanford’s.
Visits from prominent people in technology are not unusual at Stanford. I enrolled in CS 153 during the spring quarter, and that one course brought Jensen Huang, Sam Altman, Andrej Karpathy, and others into the classroom. Their talks are easy to find on YouTube, and much of what they said had been said elsewhere. The ideas themselves were not necessarily surprising.
I increasingly feel that knowledge and opinions are abundant and inexpensive to access. Andrew Ng’s Coursera and Stanford’s online lectures were created in part to make knowledge more widely available. Almost every course I have taken at Stanford has recorded lectures available through Stanford Online. The knowledge barrier once associated with a university degree seems much less substantial now. Access depends more on an individual’s willingness to learn.
What feels more valuable about these visits is the possibility they make imaginable. When Jensen Huang speaks for five minutes about US–China relations and the world around us, with extraordinary command of the room, and the audience breaks into applause; when Andrej Karpathy produces a stream of reasoning so quickly that he seems almost like a thinking agent—you realize that these people are real, like you and me. Ten or twenty years ago, some of them were Stanford students, cycling between palm trees and red-tiled buildings, listening to the previous generation of CEOs, and wondering about their own futures. It is natural to ask: why couldn’t I be someone like that tomorrow?
I think this openness also reflects a cultural habit: building in public. CEOs and ordinary engineers alike are often willing to share what they are thinking. To me, Twitter, now X, has become a meeting place for academia and industry. Product launches, tips for using agents, and new features appear there constantly. Reading it can feel like an exchange of information: finding out what others are building, sharing views, and trying to persuade people so that an idea earns enough support and confidence to become something real.
“Fake it till you make it” is a familiar phrase in Silicon Valley. CEOs appear on podcasts, write blogs, and advocate for their ideas. You can learn how they approach problems and what their latest workflows look like. Much of that information is public and accessible even to people outside the Valley. The culture influences startups, too: presenting a product and its ideas on Twitter has become almost an essential calling card. It has also produced a controversial type of founder whose main goal seems to be gaining followers and growing an audience rather than building a product.
I saw much less of this culture at Peking University. I rarely saw emerging CEOs or executives visit simply to share ideas. When they did come, the talk often served a specific purpose, such as fall recruiting. People in Silicon Valley also discuss their companies, but recruitment usually feels less central; the occasion is closer to an exchange of ideas. I did attend some excellent talks at Peking University that were simply about sharing, but the broader conditions for building in public felt less developed.
4. AI Ads at the Bus Stop
As a newcomer to Stanford and Silicon Valley, I felt the presence of AI immediately. Bus shelters, the outside of Caltrain carriages, and highway billboards carried the names and slogans of AI products, including those of major technology companies. It was hard to imagine the same scene in China: public transport serves a broad population, not just a particular professional group. Here, technology workers seemed to make up so much of the surrounding population that the ads themselves conveyed how concentrated that audience was. I once saw a Tencent recruiting billboard at the bus stop outside Peking University’s southwest gate, but it looked more like an effort aimed specifically at students than advertising a product to the public.
After getting off the plane, I could strike up a conversation with someone nearby and naturally talk about AI, vibe coding, company news, or stocks. In the Bay Area, these feel like dependable conversation starters. Meeting someone outside technology—someone working in the arts, for example—can feel almost “out of distribution.” More often, I hear that someone is an engineer or product manager, an SDE, MLE, or strategist at a large company, or works in hardware or robotics. Beijing feels more varied to me. It is a vast city, with districts that differ in function and culture and overlap in complicated ways. Almost no profession would surprise me there.
At Stanford, this atmosphere takes more concrete forms. The open space in front of the bookstore reminds me of the gathering area near Peking University’s Centennial Hall. At Peking University, I usually saw student union events, department activity weeks, or recruiting booths organized through the university. Outside Stanford’s bookstore, I sometimes see students promoting products they have built and inviting passersby to download them. Religious outreach booths are a regular presence, too. Occasionally, students even march around campus carrying signs promoting a product. Small ads for AI startups also appear on utility poles and dorm entrances, though they tend to be removed quickly.
The atmosphere reaches people outside technology as well. In the spring quarter, I took an academic writing class taught by someone from Stanford’s language center. In the first session, he discussed prompt engineering and how to use GPT to make writing more academic. He also talked with us about agents. I found that amusing and told him afterward that I had not expected a writing instructor to know those terms. His answer was simple: in the Bay Area, it is hard not to know about them.
Stanford takes particular pride in entrepreneurship. I might still misspell the word, but I can say it without hesitation because I hear it so often. It has become a description of the atmosphere itself. From roadside billboards to everyday conversations, the word keeps returning. Stay here long enough, and you gradually become part of that atmosphere, too.