# Dear San Francisco, this is how the outside world sees you.
**作者**: Hardeep
**日期**: 2026-05-20T05:10:13.000Z
**来源**: [https://x.com/hardeep_gambhir/status/2056965696307986803](https://x.com/hardeep_gambhir/status/2056965696307986803)
---

When I first came to San Francisco, I lived in a hacker house. It was where Cursor's founders and Dwarkesh Patel emerged from. The house was my introduction to the Bay. It bloomed with techno-optimism: Compassion. Kindness. Curiosity. We had come from everywhere to find like-minded nerds and found them here.
Yet over the past year as I’ve flown in and out of SF, the vibe has changed. Each time I come back to SF, the chase looks more like one of attention.
Launch videos are a signal of prestige, not product. The seed round is being displayed left and right. The grift has stopped being something a few people did and started being something a generation aspires to.
And to those outside of SF, they’ve felt the vibe change as well too. The news that makes it out of tech twitter isn’t the best of news…
Copied Cluely playbooks from other Gen-Z companies. The Friend companion billboards in New York. The nineteen-year-old is raising eight million dollars for something that won't exist in two years. The messages of the top labs and what the potential impact of AI will happen.
It makes sense why the outside world views AI negatively.
## The pipeline
Those that come to San Francisco are prestigious and performative-maxxing.
College students that used to want a Goldman or McKinsey internship now want a seed round from a16z or a YCombinator acceptance email.
When the metric for achievement in the valley is securing Tier-1 capital, and the new wave of founders migrates to San Francisco chasing the clout of Cluely's Instagram reels instead of being inspired by the essays of Paul Graham.
Don’t get me wrong, I have talked to people who have lived here for twenty years. They tell me the grifters have always come. What's new is how many of them there are and how fast they get funded.
Charlie Munger said it best: show me the incentive and I will show you the outcome. This is what I call a reference class collapse. The cohort of people a young founder looks up to and competes against is being swapped out.
The replacement is shallow, unhinged, and prestige-focused. The word "startup" still means startup. But those who play in the arena of startups have now changed.I have known of female founders being subjected to unwelcome sexual advances by their own investors. “ABG” interns sleeping with the founders of companies. It's not the Silicon Valley of the nerds anymore.
## The grift
Cluely broke the dam. It was the first to turn fluency in ragebaiting into a startup playbook. They knew that attention was the only moat. Piss people off loud enough and the algorithm does your distribution for free.
That was a sign of just the beginning.
Consider Luel.AI, which raised $31 million despite being a verbatim copy of an existing startup, even down to the landing page. General Catalyst led the round, with others following.

Every other Gen-Z company now tries to follow a similar playbook with an emphasis on attention over product. The fear of missing the one company that wipes away all losses is what keeps the money moving.
## The trajectory
Delve and Luel.AI are the beginning of disasters in SF.
I don't say this to be pessimistic but realistic: The world isn’t ready as the worst is still ahead. The Cluely imitated reels and the Friend billboard will seem less crazy as the models get better. Those outside the bay are going to remember the loudest version of AI and what it made them feel.
Crypto never got a second chance to fix its narrative. We can’t let that happen to AI. All everyone hears about is the negative narrative, not the positive.
## What this technology is also doing
AI cracked a public health crisis by identifying a four-decade-old herbicide, Picloram, as a probable link to the spike in early-onset colorectal cancer—a clue no human could ever read from DNA data.
In Vermont, an AI scribe slashed primary care provider burnout by almost two-thirds in four months, giving doctors their evenings back to focus on patients, not paperwork.
Students in rural South Africa are now learning math in Xitsonga, Sepedi, and isiZulu, thanks to an AI tutor breaking down language barriers and delivering education where it was previously impossible.
After 50 years of struggle, AlphaFold solved protein folding in months, accomplishing work that would have demanded hundreds of millions of years of manual labor, and gifted the world a free map of nearly every known protein. Every drug discovered in the next twenty years will be discovered faster because of this. Every disease pathway mapped in the next twenty years will be mapped faster because of this. It is one of the largest acts of scientific generosity in modern memory.
The world has not heard about this. The world is only aware about Gen-Z startups.
## What's at stake

Garry Tan is not wrong. Tech has been genuinely good to a lot of us inside this industry, and his last line — we must make that into prosperity for everyone — is the actual obligation. The question is why, when tech is more capable than ever of doing exactly that, the public is increasingly convinced it is doing the opposite.
The valley that birthed the semiconductor and seeded the open web has become a place where people promote pissing off others in the name of distribution. The labs and companies actually changing the world have been so busy selling the time they free up that they have forgotten to make a case for what the time is for.
That is the problem. Not the grift itself. The grift is a symptom. The problem is that the public is forming its opinion of AI based on the loudest five percent of San Francisco — the part that is performing — while the people doing the substantive work are too busy doing it to push back on the story being told.
When the harms arrive and they will, because as these models get better the worst is still ahead, the public is not going to remember Anthropic's safety research. They are going to remember whatever the loudest version of AI made them feel: the Cluely reels, the Friend billboard, the billboards that say “Stop hiring humans.”
The story being told about a technology is becoming inseparable from the technology itself. Public opinion is what decides how AI gets regulated, who funds it, who is allowed to build it, and which version of it the next decade gets shaped by.
If we lose the public on AI the way we lost the public on crypto, we will not have lost a marketing battle. We will have lost the good coming out of AI.
I love San Francisco too much to see it go down like this.
-- Hardeep + edited by friends
## 相关链接
- [amy ᯓ★ reposted](https://x.com/notesbyames)
- [Hardeep](https://x.com/hardeep_gambhir)
- [@hardeep_gambhir](https://x.com/hardeep_gambhir)
- [95K](https://x.com/hardeep_gambhir/status/2056965696307986803/analytics)
- [being inspired by the essays of Paul Graham](https://paulgraham.com/)
- [Luel.AI](https://luel.ai/)
- [Luel.AI](http://luel.ai/)
- [Upgrade to Premium](https://x.com/i/premium_sign_up)
- [1:10 PM · May 20, 2026](https://x.com/hardeep_gambhir/status/2056965696307986803)
- [95.2K Views](https://x.com/hardeep_gambhir/status/2056965696307986803/analytics)
- [View quotes](https://x.com/hardeep_gambhir/status/2056965696307986803/quotes)
---
*导出时间: 2026/5/21 09:41:59*
---
## 中文翻译
# 亲爱的旧金山,这就是外界眼中的你。
**作者**: Hardeep
**日期**: 2026-05-20T05:10:13.000Z
**来源**: [https://x.com/hardeep_gambhir/status/2056965696307986803](https://x.com/hardeep_gambhir/status/2056965696307986803)
---

当我初到旧金山时,我住在一个“黑客之家”。那是 Cursor 的创始人和 Dwarkesh Patel 走红的地方。那所房子是我进入湾区的入口。它充满了技术乐观主义:同理心、善意、好奇心。我们从世界各地汇聚于此,寻找志同道合的极客,并在这里找到了他们。
然而在过去的一年里,随着我频繁往返于旧金山,这里的氛围变了。每次回到旧金山,这种追逐看起来都更像是一场对注意力的追逐。
发布视频成了声望的信号,而非产品的展示。种子轮融资被到处炫耀。投机取巧已不再是个别人的行为,而成了一代人的向往。
对于旧金山以外的人来说,他们也感受到了这种氛围的变化。那些传到科技推特圈之外的消息,往往都不是最好的消息……
抄袭其他 Z 世代公司的 Cluely 式操作手册。纽约的“Friend”伴侣广告牌。那个 19 岁的年轻人为一个两年后就不会存在的东西融资了 800 万美元。顶级实验室的言论,以及关于 AI 潜在影响的各种预言。
难怪外界对 AI 的看法如此负面。
## 人才输送链
那些来到旧金山的人,都是追求声望和极致表演的人。
过去想要高盛或麦肯锡实习机会的大学生,现在想要的是 a16z 的种子轮融资或 Y Combinator 的录取通知。
当衡量成就的标准变成了获得顶级资本(Tier-1),当新一波创业者为了追逐 Cluely 在 Instagram 上的名气而涌向旧金山,而不是受保罗·格拉汉姆文章的感召时。
别误会,我和在这里住了二十年的人聊过。他们告诉我,投机者一直都有。但新在于他们的数量之多,以及他们获得资金的速度之快。
查理·芒格说得最好:给我看看激励机制,我就给你看看结果。这就是我所说的“参照系崩塌”。年轻创业者仰望和竞争的群体正在被置换。
替代者是肤浅的、毫无底线且沉迷于声望的。“初创公司”这个词仍然是那个意思,但在这个竞技场里角逐的人已经变了。我听说有女性创始人遭到了自己投资人的性骚扰。“ABG”类型的实习生与公司创始人发生关系。这里不再是极客们的硅谷了。
## 投机取巧
Cluely 决堤了。它是第一个将“激怒人”的熟练度转化为初创公司操作手册的公司。它们知道,注意力是唯一的护城河。只要把人惹得够恼火,算法就会免费为你做分发。
这仅仅是个开始的迹象。
看看 Luel.AI,尽管它是对现有初创公司的逐字复制,甚至连着陆页都一模一样,却融资了 3100 万美元。General Catalyst 领投了这一轮,其他人紧随其后。

现在每隔一家 Z 世代公司都试图遵循类似的操作手册,强调注意力胜过产品。对错过那一家能抹平所有损失的公司的恐惧,正是资金不断流动的原因。
## 发展轨迹
Delve 和 Luel.AI 是旧金山灾难的开始。
我这么说不是为了悲观,而是为了现实:世界还没准备好,因为最糟糕的情况还在后面。随着模型变得更好,Cluely 模仿的短视频和 Friend 的广告牌看起来将不再那么疯狂。湾区以外的人会记住 AI 最喧闹的那个版本,以及它带给他们的感受。
加密货币从未有机会修复它的叙事。我们不能让 AI 也重蹈覆辙。所有人听到的都是负面的叙事,而不是正面的。
## 这项技术还在做什么
AI 破解了一场公共卫生危机,它识别出一种已有 40 年历史的除草剂——毒莠定,是早发性结直肠癌激增的可能推手——这是人类无法从 DNA 数据中读取的线索。
在佛蒙特州,一名 AI 抄写员在四个月内将初级保健提供者的职业倦怠减少了近三分之二,把晚上还给医生,让他们专注于病人,而不是文书工作。
在南非农村的学生现在正用聪加语、北索托语和祖鲁语学习数学,多亏了一位打破语言障碍的 AI 导师,它将教育带到了以前不可能到达的地方。
经过 50 年的挣扎,AlphaFold 在几个月内解决了蛋白质折叠问题,完成了本来需要数亿年手工劳动的工作,并赠送给世界一张几乎所有已知蛋白质的免费地图。未来二十年内发现的每一种药物都会因为这项技术而发现得更快。未来二十年内绘制的每一条疾病通路都会因为这项技术而绘制得更快。这是现代记忆中最大的科学善举之一。
世界并没有听说过这些。世界只知道那些 Z 世代的初创公司。
## 悬崖之上

Garry Tan 并没有错。科技对我们这一行里的许多人来说确实很好,而且他的最后一句话——我们必须将其转化为所有人的繁荣——才是真正的义务。问题是,为什么当科技比以往任何时候都更有能力做到这一点时,公众却越来越确信它在做相反的事情。
那个孕育了半导体、催生了开放网络的湾区,已经变成了一个人们以分发为名宣扬如何激怒他人的地方。那些真正在改变世界的实验室和公司一直忙于推销它们节省下来的时间,却忘记了去论证这些时间是为了什么。
这才是问题所在。不是投机取巧本身。投机取巧只是一个症状。问题在于,公众正在基于旧金山那最喧闹的 5% 的人——那些在表演的人——来形成对 AI 的看法,而那些做实质性工作的人则忙于工作,无暇反驳那个正在被讲述的故事。
当伤害到来时——它们一定会到来,因为随着这些模型变得更好,最糟糕的情况还在后面——公众不会记得 Anthropic 的安全研究。他们会记得 AI 最喧闹的那个版本带给他们的感受:Cluely 的短视频,Friend 的广告牌,那些写着“停止雇佣人类”的广告牌。
关于某个技术正在被讲述的故事正变得与技术本身不可分割。公众舆论决定了 AI 如何被监管、谁资助它、谁被允许构建它,以及未来十年将塑造哪个版本。
如果我们在 AI 上失去了公众,就像我们在加密货币上失去公众一样,我们将输掉的不是一场营销战。我们将失去 AI 所带来的好处。
我太爱旧金山了,不忍心看着它就这样走下坡路。
-- Hardeep + 朋友编辑整理
## 相关链接
- [amy ᯓ★ reposted](https://x.com/notesbyames)
- [Hardeep](https://x.com/hardeep_gambhir)
- [@hardeep_gambhir](https://x.com/hardeep_gambhir)
- [95K](https://x.com/hardeep_gambhir/status/2056965696307986803/analytics)
- [being inspired by the essays of Paul Graham](https://paulgraham.com/)
- [Luel.AI](https://luel.ai/)
- [Luel.AI](http://luel.ai/)
- [Upgrade to Premium](https://x.com/i/premium_sign_up)
- [1:10 PM · May 20, 2026](https://x.com/hardeep_gambhir/status/2056965696307986803)
- [95.2K Views](https://x.com/hardeep_gambhir/status/2056965696307986803/analytics)
- [View quotes](https://x.com/hardeep_gambhir/status/2056965696307986803/quotes)
---
*导出时间: 2026/5/21 09:41:59*