The first phase of the AI era has ended. When the second phase arrives, no one knows.
I recently read the viral article "The Model Generation" and felt a deep resonance. I am currently writing my own observations from Silicon Valley, so this is the perfect opportunity to offer an overseas response to that piece based on what I have seen here.
- Bubble Wealth
"The Model Generation" spends significant time describing the income gaps between classmates who took different paths: tenfold differences, hundredfold differences, and even those earning over $100 million. These stories of immense wealth primarily happen in Silicon Valley, where I am now located in Northern California.
Every day, there are indeed many wealth myths circulating here: a 17-year-old raising millions, valuations skyrocketing by 100x, companies being acquired and moving from stairwells to mansions, recent graduates buying luxury homes... The list goes on. So, although people here are generally wealthier than those across the ocean, the anxiety is just as intense.
If domestic pressure is like "involution," crushing everyone into flat paper figures, then here it is a massive bubble. The bubble makes everyone feel light-headed, like a drug-induced high, yet also lost in fog or mirages, causing constant fear of loss. I think neither state is healthy.
Harvard Business School researchers surveyed over 4,000 millionaires. They found that people with net worths between $3 million and $7.9 million were not significantly happier than those with $1.5 million to $2.9 million. Happiness only jumped noticeably above $8 million (or $10 million in another sample), and even then, only slightly.
This means if you already have $1 million in net worth and decide to work harder to reach $3-4 million, your life won't change much: you might still have a house in South Bay, walk dogs on weekends, eat French or Japanese food occasionally, drive to San Francisco or Half Moon Bay, or fly to Yellowstone. Beyond that, nothing changes.
For example, Alex (pseudonym), a Big Tech employee, felt stuck in the $1.5M-$7M range and wanted a qualitative leap to over $8M. His solution? Quit his job and start a company. Then the anxiety began. Starting up is risky; before things settle, you live in endless pain: convincing investors, co-founders, and users; showing revenue and beautiful ARR numbers; worrying about your narrative being eaten by LLM capability updates; and constantly marketing in SF to stay visible to investors.
Take Corgi, an insurance startup that opened a 24-hour cafe downstairs from its SF headquarters. When the cafe became popular, they launched free shuttles around YC (the hottest incubator) to pick up programmers working overnight. This generated traffic and raised significant funds.
So, much of the money here is tied to traffic. Once you are attached to traffic, you can never escape; you spend your life chasing highs and lows, which is painful. Happiness is truly hard to find in San Francisco. If you love comparison, you will run endlessly in the gears of competition.
2. Same LLM, Huge Gap
Both graduated from Stanford. Law school students entering Biglaw (top US firms) earn about $220k-$230k in their first year. But Chinese students rarely get into Biglaw anymore; after a year of searching, they land at small firms earning ~$100k, with stress levels comparable to Big Tech engineers. Meanwhile, senior Big Tech programmers earn hundreds of thousands to over a million dollars annually, and Research Lab scientists make at least $1M, while regular programmers make over $200k.
The most absurd part: Both involve "LLM." Those working on Large Language Models start at $1M. Those studying for a Master of Laws (LL.M.) earn $100k. And an LL.M. at Stanford costs nearly 1 million RMB in tuition and living expenses. The ROI is staggering.
Income disparity here is huge, with wealth concentrated in tech. Other industries, especially white-collar jobs, barely make money. Of course, I know many friends at Stanford don't rely on their own earnings and study for experience, but that's too biased to discuss.
Back to AI. AI has indeed made many people rich overnight. For instance, if you joined Anthropic early last year when its valuation was $61.5 billion, and this week it lists on Nasdaq with a market cap of ~$1.4 trillion (potentially $1.5 trillion), the company's value multiplied by 20+ times in a year. If you negotiated good options, you became instantly wealthy.
However, working at Anthropic or OpenAI is exhausting, with reports of poor culture: internal discussions often revolve around "AI might destroy humanity." Anthropic's seven co-founders publicly pledged to donate 80% of their wealth, and the company has an employee stock donation matching program. Although voluntary, reportedly half the employees participate.
Additionally, the overflow of wealth and bubbles bring "support services": matchmaking events, companionship services, etc. You suddenly feel the world's doors are open, filled with hedonism. But this tests human nature and isn't conducive to a long-term stable, healthy life.
3. Bubble Connections
Regarding "meeting big shots": In SF, it's incredibly easy. I've met Thariq (Claude Code creator), Garry Tan (YC), and Pika's founder. You can chat with them casually. A founder having coffee with you today might raise $10M tomorrow. Before experiencing this, I envied it; after two months here, I realized it's commonplace and not that special.
I encourage anyone with a filter on this to visit SF; you'll realize it's okay. As "The Model Generation" said, chatting with investors, Zuckerberg, or Lei Jun seems impressive, but ultimately, people judge your actual work and projects.
Your best connections here are peers at similar stages, such as other early-stage founders. There are many hacker houses here. I visited a Taiwanese hacker house with a great atmosphere: everyone discussed products daily, testing each other's apps. When someone raised funds, they announced it in the house, and everyone had dinner together. New ideas led to new companies. This ecosystem is strong in SF, nurturing many successful YC companies and founders with pure intentions.
I believe a benefit of SF is finding founders with pure intentions, whereas some domestic founders may just be out to grab cash. The environment here allows purity; domestically, some start pure but turn cynical when survival becomes difficult.
There are tricks to raising money here. Consider two PhD graduates, Edward and Ashley. Edward continued teaching abroad. Ashley joined Tesla, then OpenAI. After starting her company, both giants gave her small grants ($10k-$50k). To investors, this is negligible, but backed by these brands, she easily raised millions in USD.
Ashley is smart; she aims for acquisition, not IPO, because IPOs are tiring. Acquisition targets allow for 5x-10x cash-outs. If you seek rapid wealth growth, starting up here works well if you find the right niche: autonomous driving data, world model data, robot data, or plugins/infra/data interfaces for Personal Agents.
4. Bubble Emotions
I believe the biggest opportunity in the AI era is emotional issues. Material abundance today far exceeds any previous era: medieval Europe had the Black Death; decades ago, we couldn't afford meat. Yet, people are not happier.
Mary (pseudonym), a T2 undergrad from China, Ivy League master's grad, elite woman commuting between Palo Alto and SF, feels miserable. She thinks everyone around her is richer and luckier. She regrets choosing finance over AI. But AI media doesn't pay (TechCrunch doesn't either); you must be a researcher. She tried dating researchers, only to find they use companionship services. It's like trying to take off the "long gown" (intellectual pretension), only to find the lane is crowded and distorted.
Identity issues: H-1B visa holders face a 60-day grace period after unemployment. The DHS proposed canceling this in September. Though not finalized, it hangs over heads. If laid off, she might have to leave immediately.
Doris (pseudonym) comes from a wealthy family. Her husband left Big Tech to start a profitable company. She is famous and invests heavily in children. Median home prices in Palo Alto are $3-4M, often old houses. Private elementary schools cost $40k-$50k/year. To reminisce about domestic life, she shops at Stanford Shopping Center for luxury brands. Eventually, she realizes the money passes through her; she is just a transient holder. Children need equestrian lessons and film projects for Ivy League applications.
Joe (pseudonym), a Stanford student. Others see him as brilliant: top CS major, junior engineer salary of hundreds of thousands. But he is anxious. He sees countless startup events but lacks ideas. He doesn't know how to use storytelling to attract investors. He wants a co-founder but struggles with marketing; his Twitter followers are double digits, and Xiaohongshu posts get throttled. People have their own ideas; no one wants to join him because his background is common at Stanford. He doesn't know what to build.
He doesn't want to do homework (AI does it). He tries open-source projects, but few care. He finds SF investors incomprehensible. He feels opportunities are right there, but he can't grasp them, paralyzed by noise.
Consequently, spiritual practices are abundant in SF. Many Zen temples exist (Chinese, Japanese, Indian styles). The demand for inner peace is huge because chaotic wealth iteration leaves people disoriented, seeking meditation in temples.
5. End of AI Phase One
First, cognitively, China and Silicon Valley aren't vastly different; China tracks SV closely. Phase One had a grand prologue from late 2025 to mid-2026: rapid model iterations, Claude Code opening the Agent era, OpenClaw leaving a mark in open source, Jensen Huang's July X post signing the "Open Weights & US AI Leadership" letter, representing a sense of mission. The ending was dramatic: On election night in November, Sanders called for a national pause on new data centers; Democratic lawmakers followed; a week later, Huang was "taken down." It felt cinematic. This phase is over.
Regarding startups: Hundreds of weekly events in SF, 70% talking about Agents. Of those, 70% focus on work/team scenarios. Imagination is lacking. The remaining 30% doing Personal Agents either automate calls/video or integrate IMs (already done by Big Tech) or add a layer of skin with anime avatars. How is this different from attending lectures in Haidian's AI Origin Community or Shanghai's Model Speed Space? Not much.
A paradox emerges: AI should stimulate consumption, like steam or electricity did. Steam increased cloth/coal production; electricity enabled nightlife; information revolution enabled global e-commerce/social.
But what needs does AI stimulate now? It hasn't stimulated even 1% of internet-era demand. Who pays? Only model developers. Why sell data? To earn investor money: Investors give money to model makers, who buy data. The core issue remains: The killer app point hasn't been found. It might appear in Phase Two, but no one knows when. Maybe someone somewhere invents it. It's fate.
Two types of AI-anxious people:
- Those who don't understand AI but hear influencers say "learn AI or die."
- Those in the center of the universe, under a bubble, hearing everyone makes millions, feeling cheated earning "only" hundreds of thousands, anxious to make more.
SF splits into two groups: AI workers and non-AI workers. Law/business students often still use chatboxes, ignoring Codex/Claude Code. At AI events, they are invisible. Same Stanford, same LLM acronym, huge gap in familiarity between Law and CS departments.
The other group is extremely hype/FOMO. At any event, you meet people obsessed with latest trends, GitHub projects, open-source stars.
These are extremes. The "outside the bubble" represents 95-99% of the world. Even in SF, AI hasn't penetrated everyone.
Why? Working agents aren't needed by most. Coding is the main use case, hence high penetration among programmers. Programmers feel the impact deeply: 2023 models hallucinated code; 2025 they write it; 2026 they submit/merge PRs. Other industries feel less impact. This distinguishes inside vs. outside the bubble. Inside, besides programmers, are investors wondering why projects get acquired so quickly.
Last year, I encouraged embracing AI. When anti-distillation skills went viral, many anxious friends asked me. I admitted AI liberated me; it handles editing/layout for my self-media, which I hate. It's powerful.
But my faith shattered seeing Dario and others lack empathy, enjoying playing god. Like "The Model Generation" says, they believe their work dictates the future.
Many "Jiahao" (arrogant egoists) in AI. Historically, Jiahaos drive progress. Musk is a Jiahao. Trump is a Gilded Age Jiahao. Their confidence infects others. 📷
Anyway, Jiahaos lack empathy. With empathy, they wouldn't let SF become so ugly, ignoring the bottom tier. 📷 SF street tents.
What is there to learn about AI? You don't know what it will become. Learning Dify in 2024 was useless when Claude Code arrived. With OpenClaw personal agents in 2026, you won't even open your computer. If learning out of fear, stop. Wait for Olympiad medalists to build robots for you. Brainwork is becoming industrialized/scaled. In this "genius nation in data centers," a global brain does all mental labor. What is there to learn? Unless, like me, video creation is a hobby, and AI handles chores—that's good. But if learning out of fear, it's unnecessary; the more you learn, the more afraid you become.
6. The Era of Active Mental Exercise
In eras of scarcity, people spent time finding food, avoiding obesity. Now, abundance causes obesity, requiring deliberate exercise to burn calories, creating fitness demands. AI is similar. Previously, intellectual scarcity required borrowing books, traveling for education (like Song Lian's story). Now, intellect is abundant, readily available via phone, pre-digested. This leads to mental laziness/obesity, rigid thinking due to lack of exercise.
To keep thinking fit, you must consciously exercise your mind, like going to the gym. Future "mental gyms" with cognitive tools may emerge.
I think AI learning bloggers like Zhang Zhala and Karpathy are suited to design these tools.
I cannot say "AI makes the world better" yet. Maybe Phase Two/Three will. Phase One ended dramatically. Old orders crumble; new ones haven't formed.
Before Phase Two, likely years remain. No global hot wars expected. Just live. Exercise body and mind to avoid physical/mental obesity.
Good methods? Read "The Model Generation" and write a resonant essay. Or do self-media, studying human inertia/attention. Understanding human nature is timeless power.
Or start a business, starting with end goals. Don't think about AI first; think about real human needs. I met interesting entrepreneurs who use AI as a tool, not for AI's sake. E.g., a matcha robot shop in downtown Palo Alto. They didn't open it for robotics; they used robots to cut costs/stabilize quality. One person runs the large store. I worked as manager for a day. It solves problems: expensive/unstable labor, 12-hour shifts challenging humans but fine for robots, hygienic.
Another does roof repair using AI to scan forums/local communities for leads, auto-messaging/contacting, sending cases. Roof repairs cost $10k+, very lucrative.
Emotional issues are huge. Why are Xu Fei/Li Na popular this winter? Pure emotion is scarce/supplied poorly. Modern people are cold. Seeing genuine emotion attracts crowds.
If my articles provide emotional value, they are excellent. I hope to find healthy ways to soothe emotions without relying on dubious spiritual practices. My self-media is the channel for now.
Wish everyone well. Bubbles burst. During the burst, realizing economic downturns are natural, merely delayed by AI, might help us find solutions by reading Japan's post-bubble history.





