#226 Garry Tan: Billion-Dollar Misfits — Inside Y Combinator’s Startup Formula
2025/4/29
The Knowledge Project with Shane Parrish
G
Garry Tan
Garry Tan: 我认为Y Combinator的成功源于Paul Graham和Jessica Livingston的杰出个人特质，以及他们对帮助他人的意愿。Paul Graham的著作和才华吸引了志同道合的人，为Y Combinator的成功奠定了基础。Y Combinator不仅提供资金，更重要的是免费提供专业知识和指导，吸引那些希望创造改变世界软件的人。在筛选过程中，创始人所创造的产品和能力比简历更重要。尽管有软件辅助，但筛选仍高度依赖人工判断，10分钟的面试旨在快速评估创始人和市场潜力。我们寻找的是能够解决实际问题、拥有持久竞争优势的创业公司。我们持续从成功和失败案例中学习，以改进筛选机制。我们专注于早期投资，与其他风险投资公司形成差异化竞争。我们投资的许多创始人非常年轻，需要时间来成熟和学习。Y Combinator的环境能够帮助创始人获得同侪支持和专注于产品开发。许多创业公司关注的是融资和名声，而不是实际的产品和市场。创业者应该专注于解决问题，而不是追求名利。留在旧金山的团队更有可能成为独角兽公司。我们也会拒绝一些看似有潜力的项目，因为它们可能对社会有害。合伙人拥有独立的投资决策权，并注重价值观的一致性。“真诚”是Y Combinator最看重的创始人特质之一。Brian Armstrong是真诚的创业者的典范，而Sam Bankman-Fried则恰恰相反。真诚和坚持不懈是成功的关键。
AI领域最大的变化之一是，AI公司的快速扩张不再是成功的唯一途径。云计算降低了创业的门槛，使得小型团队也能取得成功。现在，越来越多的技术型人才担任CEO。创始人需要积极参与公司运营的各个方面。缺乏创始人领导力的公司容易出现权力斗争和低效运营。许多大型科技公司未能有效利用人才，导致技术进步减缓。AI监管应该避免过早干预，保持市场竞争。AI模型提供商之间的竞争对消费者和社会有利。AI监管应该避免扼杀创新。教育应该培养学生的自主性和创造力。AI监管需要考虑全球范围内的影响，并确保其益处能够惠及所有人。目前，AI模型还无法完全取代人类在模型开发中的作用。AI有潜力在生物技术、制造业等领域取得突破性进展，改善教育，并提高人们的生活水平。AI模型的数据来源和版权问题是一个灰色地带。AI生成的科学成果的知识产权问题有待解决。AI在生物恐怖主义方面的风险已经被现有的法律所涵盖。AI带来的风险和机遇并存，需要谨慎对待。AI领域不太可能出现一家独大的局面。Meta在AI领域的投入是积极的，但其产品策略有待改进。AI将改变人机交互方式。大型AI模型正在被压缩成更小的模型，以便在个人设备上运行。AI监管措施可能并不总是有效。AI安全领域的资金投入可能导致不必要的担忧和监管。AI监管需要在安全性和创新之间取得平衡，并考虑全球范围内的影响。AI监管需要快速响应，以适应技术快速发展的步伐。AI可能会加剧信息茧房效应，但同时也可能带来更多选择。AI监管应该鼓励开放系统和用户选择。科技公司应该持续创新，而不是依赖于暂时的竞争优势。许多人对AI的期望是不切实际的。AI可能会取代人类在某些领域的决策作用。AI与国家之间的互动将是一个复杂的问题。人机协同是目前最有效的AI应用方式。AI领域最前沿的研究人员正在关注测试时间计算和推理模型。对AI算力的需求将持续增长，NVIDIA等公司将从中受益。AI正在取代某些工作岗位，但同时也创造了新的机会。AI正在被应用于客户服务等领域，提高效率和一致性。AGI的定义尚不明确，但其在许多领域已经展现出强大的能力。未来几年是AI发展的黄金时期。有效的提示工程需要将复杂任务分解成更小的步骤。AI应用的持久竞争优势在于对特定领域数据和流程的深入理解。在AI时代，持续的竞争优势在于构建良好的用户体验和软件。
Shane Parish: (问题和引导性发言，未形成完整观点)
supporting_evidences
Garry Tan: 'And it will never get fixed because people don't care. No one's paying attention. That's just one symptom out of a great many that is, you know, the result of basically treating people like, you know, hoarded resources. The world is full of problems. Let's go solve those things.'
Garry Tan: 'I want to start with what makes Y Combinator so successful. I guess I can't talk about YC without talking about Paul Graham and Jessica Livingston. I mean, it started because they're remarkable people.'
Garry Tan: 'And so I think Paul Graham and his essays became a shelling point for people who this new thing that could really happen in the world.'
Garry Tan: 'And then more importantly, I think the know-how is we give it away for free.'
Garry Tan: 'But the most important thing to me is not necessarily the biography. It's actually... what have you built? What can you build?'
Garry Tan: 'And now it's, you know, let's just use AI for everything. But I think that the human element is still very important.'
Garry Tan: 'I guess the surprising thing that has worked over and over again, ultimately, is in those 10 minutes, either you learn a lot about both the founders and the market or you don't.'
Garry Tan: 'And I think what we're looking for is actual signal that there's a real problem to be solved. There are people on that end who are willing to pay.'
Garry Tan: 'And the remarkable thing is that that's why it only has to be 10 minutes.'
Garry Tan: 'Oh, definitely. All the time. I mean... I think that's the trickiest thing. I think the system itself will always produce both false positives and false negatives because it is only 10 minutes.'
Garry Tan: 'And you're still the most or at least top tier in terms of success. Yeah, I mean, what's great is I don't want to compete with Sequoia or Benchmark or Andreessen Horowitz or, you know, they're our friends, honestly.'
Garry Tan: 'And now we're around 5.5%. Some batches from... Maybe 2017, 2018 are pushing 8% to 10%. Oh, wow.'
Garry Tan: 'It takes three to five years to mature, to learn how to iterate on software, how to deliver really high quality software, how to manage people, how to manage people effectively, give feedback.'
Garry Tan: 'Oh, yeah. For the first time in my life. And they're super ambitious.'
Garry Tan: 'The default startup scenario out there is not about signal, it's about the noise.'
Garry Tan: 'I think I've seen Peter Thiel talk about this. He doesn't really want people who want to start startups.'
Garry Tan: 'the teams that come to San Francisco and then stay in San Francisco or the Bay Area they actually double their chance of becoming a unicorn.'
Garry Tan: 'I think what I want for San Francisco, for instance, is I think the rent should be lower.'
Garry Tan: 'I mean, if it's scary but might or probably will be good, I think we want to fund them.'
Garry Tan: 'I think we're pretty aligned there.'
Garry Tan: 'And the thing is, you know, you can't get that.'
Garry Tan: 'And the 10,000 hours of human training people get from their childhoods comes up in Paul Graham's essays.'
Garry Tan: 'They're not going to have as earnest people who become as formidable around you.'
Garry Tan: 'When Paul and Jessica created YC, I went through the program myself in 2008, and I came out transformed.'
Garry Tan: '25% of the people who reach some form of product market fit at YC do it in year five or later.'
Garry Tan: 'So I'm like, that's crazy. I'm really glad that YC exists now.'
Garry Tan: 'I think of the word earnest.'
Garry Tan: 'Like, you know, Brian Armstrong is an incredibly earnest founder who literally read the Satoshi Nakamoto white paper and said, this is going to be the future.'
Garry Tan: 'I'm hoping that you know young people listening to this right now take that to heart.'
Garry Tan: 'I think the biggest thing that is increasingly true, and we're seeing a lot of examples of it in the last year, is blitzscaling for AI might not be a thing.'
Garry Tan: 'That was even why Y Combinator in 2005 could exist.'
Garry Tan: 'And, you know, that's the other big shift.'
Garry Tan: 'And he spoke very eloquently and in a raw way about how your company ends up not quite being your own unless you are very explicit.'
Garry Tan: 'And then you have sort of fiefdoms that are fighting it out with one another.'
Garry Tan: 'They probably hired way too many people.'
Garry Tan: 'I feel pretty good because it feels like there are five, maybe six labs, all of whom are competing in a fair market.'
Garry Tan: 'Because you're going to have choice.'
Garry Tan: 'I guess there are a bunch of different models that I could see happening.'
Garry Tan: 'And people often ask like, what should we be teaching our kids?'
Garry Tan: 'Yeah, I think the biggest question around it is, of course, I mean, the existential fear is like, where are all the jobs going to go?'
Garry Tan: 'Yeah, we're not there yet.'
Garry Tan: 'So, you know, obviously these are like relatively early in toy examples.'
Garry Tan: 'The funniest thing, my main response to all of that around like provenance of the data itself is at some point like it feels like it actually is fair use though.'
Garry Tan: 'But if AI generates not only the science behind it, like we're at a point where, you know, maybe in the next couple of years, AI is doing more science than we've done.'
Garry Tan: 'The wild thing is basically possessing instruments of creating bioweapons is already illegal.'
Garry Tan: 'I mean, it's brand new, right?'
Garry Tan: 'It might still happen, right?'
Garry Tan: 'I think that, you know, what Zuck and Ahmad and the team over there are doing is frankly God's work.'
Garry Tan: 'Clearly, the way that we're going to interface with computers is totally going to change.'
Garry Tan: 'And so there's distillation happening from the frontier, very largest models with the most data and the most intelligence down into smarter and smarter tiny models.'
Garry Tan: 'And so, you know, some of it is like, why are we doing these measures that may not actually even matter?'
Garry Tan: 'And, you know, uh, someone was telling me recently, like if you looked at on LinkedIn for, um, some of the people in these sort of giant, uh, the giant NGO morass of think tanks,'
Garry Tan: 'But we also operate in a worldwide environment where other people might not think the same way about safety that we do.'
Garry Tan: 'It is actually a real concern that the space is moving so quickly that if it takes legislation two years to make it through, that might be too slow.'
Garry Tan: 'Well, my hope is that there's such a flowering of choice that... it's going to be your choice, actually.'
Garry Tan: 'And so, you know, the kind of regulation that I would hope for is actually open systems, right?'
Garry Tan: 'I don't want to give you a golden ticket.'
Garry Tan: 'It'd be super interesting to be able to have an advantage if you're big tech and you're a company and you come up with this, but have that advantage erode automatically over time.'
Garry Tan: 'I don't tend to get into politics here but so many people in the administration are already incredibly wealthy'
Garry Tan: 'It might be that point that I just gave you.'
Garry Tan: 'Will we have a company that just literally gives in to whatever the central entity says, that's what we're going to do.'
Garry Tan: 'I was thinking about this last night watching the football game actually.'
Garry Tan: 'There's sort of like a Straussian reading of them.'
Garry Tan: 'And then I think that's why, in general, we really hate politics inside companies because...'
Garry Tan: 'I guess now that we're sitting here thinking about it, it's sort of conceivable, but in sort of all of these cases, I would much rather there be a real human being.'
Garry Tan: 'One of the crazier ideas I've been talking to people about that I feel like would be a fun sci-fi book would be just speculation playing out on how this interacts with nation states.'
Garry Tan: 'And I think it was Tyler Cowen who said that, like he had a word for it, mixing the technologies.'
Garry Tan: 'And then in that case, like my argument would be like the most American version of it is that like, you know, you and I have our own ASI.'
Garry Tan: 'I mean, you know, the hard part is like I spend most of my time not with those people.'
Garry Tan: 'Well, the demand for intelligence is going to... Trillions of dollars of investments in AI.'
Garry Tan: 'And so increasingly what we're starting to see is like, especially the bleeding edge is probably customer support.'
Garry Tan: 'Yes. I mean, I funded a company in this very current batch that is called Leaping AI.'
Garry Tan: 'I guess the funniest thing is Microsoft, I think, is defining it when it gets its $100 billion back.'
Garry Tan: 'Probably the person I would most point people to is Jake Heller.'
Garry Tan: 'And somehow that worked.'
Garry Tan: 'And then that's how, you know, Jake Heller figured out he could create something that would, you know, basically do the work of hundreds of, you know, lawyers and paralegals.'
Garry Tan: 'And so I just only saw it this morning.'
Garry Tan: 'Yeah. I mean, I think, you know, ultimately the model itself is not the moat.'
Garry Tan: 'So if that's true, then that might be a durable moat for this year, but it might not be past...'
Garry Tan: 'But all the other advantages still apply.'
Garry Tan: 'And it's the second best because the first best will be a really great UI made by a really good product designer who's a great engineer, who's a prompt engineer, who actually creates software that doesn't require copy-paste.'
Garry Tan: 'Okay, I guess, you know, is it cheating to say I'd put even more money into the YC funds that I already run?'
Garry Tan: 'And maybe on the order of the Manhattan Project, and we just haven't really thought about it enough, right?'
Garry Tan: 'And then even going back to the robotics question, it's like if we end up making universal basic robotics, the limit will still actually be sort of the climate crisis.'
Garry Tan: 'And then why Meta?'
Garry Tan: 'Oh, yeah. He's great. I mean, he was very brusque with me.'
Garry Tan: 'Unfortunately, you know, unfortunately, and this is the thing, like, when you're trying to make smart content, it's actually kind of tricky because you don't want necessarily... more clicks, you want more clicks from people who are smart.'
Garry Tan: 'Usually, a person looking into the camera seems to help a lot.'
Garry Tan: 'But if you're going to post very regularly, that's pretty important, actually.'
Garry Tan: 'And then there is an interesting interaction.'
Garry Tan: 'And then the funniest thing is subs do very little, actually.'
Garry Tan: 'Because if you hit the bell icon and they have notifications on, that's the only thing that is almost as good as having their email address and emailing them.'
Garry Tan: 'I think the number one thing that is very hard, but is so... I mean, you can see it and read it in his essays. It's to be plain spoken.'
Garry Tan: 'If we're not being authentic to ourselves or if we're just trying to follow trends or do things that get clicks, it's like that's not helpful to them either.'
Garry Tan: 'Like, you know, I sometimes am in danger of like caring too much about like the number of followers I have and things like that.'
Garry Tan: 'And then also that is actually thinking, like writing is thinking.'
Garry Tan: 'And you need to be able to breathe that belief into other people.'
Garry Tan: 'And so, and even then, I mean, I guess it's kind of fractal.'
Garry Tan: 'And so that's what a founder really is.'
Garry Tan: 'he was fooling the world because you know right like you you know it's hard to fool somebody who knows versus somebody who doesn't know and he wasn't trying to appeal to you he was trying to appeal to you yeah other people who didn't know'
Garry Tan: 'And then they do tend to get defensive or sort of angry with you for not understanding what they're saying'
Garry Tan: 'if you can't get it clear in like two sentences, you might miss an opportunity.'
Garry Tan: 'I mean, in full transparency, I mean, probably, you know, the median YC startup still fails, right?'
Garry Tan: 'Like, it's just unbelievable that...'
Garry Tan: 'And it was really, really hard to walk away from that.'
Garry Tan: 'And he's like, Gary, do you understand what this means?'
Garry Tan: 'And then the rest sort of takes care of itself.'
Garry Tan: 'There's always going to be something that stands in the way of any given person.'
Garry Tan: 'I'm a huge fan in equal opportunity, but unequal outcome.'
Garry Tan: 'And I just think of Steve saying, we wanna put a dent in the universe.'
Garry Tan: 'And I sat down with Paul right after we shook hands and he's like, Gary, do you understand what this means?'
Garry Tan: 'We, you know, kind of like I think what Sam did with OpenAI with, you know, pulling forward large language models and AI and bringing about AGI sooner.'
Garry Tan: 'And it's not more complicated than how do we get in front of optimistic, smart people who, you know, have been benevolent, you know, sort of goals for themselves and the people around them.'
Garry Tan: 'And then the rest sort of takes care of itself.'
Garry Tan: 'Like, you know, this thing has never existed before like this.'
Garry Tan: 'There's always going to be something that stands in the way of any given person.'
Garry Tan: 'I'm a huge fan in equal opportunity, but unequal outcome.'
Deep Dive
Shownotes Transcript