提前十年下注的投资人 The Investor Who Bets Ten Years Early

每一家改变世界的公司，起点都是一样的：不是一则头条新闻，也不是十亿美元的估值，而是一个原型，一个由某个对某个问题念念不忘的人打造出来的粗糙初版。 Every world-changing company starts the same way, not as a headline, not as a billion-dollar valuation, as a prototype, a rough first version built by someone who couldn't stop thinking about the problem.

这档节目要找的正是这样的人，看看他们还在打造什么。他们是 AI、机器人和前沿科技领域的创始人，活跃在太平洋两岸，从硅谷到深圳，以及每一个正在拼凑未来的地方。 This is a show that finds those people what they're still building, founders in AI, robotics, and frontier tech, on both sides of the Pacific, from Silicon Valley to Shenzhen, and everywhere the future is being put together.

亲手拼凑。 By hand.

我是 Gavin。 My name is Gavin.

这里是《雏形》。 This is Prototype.

1990 年，我的嘉宾坐在耶鲁大学的一间实验室里，亲手编写了一个神经网络，用 C++ 写出了 33 个参数。 In 1990, my guest sat in a lab at Yale University and hand-coded a neural network, 33 parameters written C++.

他是照着一位名叫杰弗里·辛顿的研究者的论文做出来的，而辛顿在 34 年后凭借这一方向的研究获得了诺贝尔奖。 He built it from a paper by a researcher named Geoffrey Hinton, who would go on to win the Nobel Prize for that line of work 34 years later.

之后的十年里，他致力于说服 500 万开发者去相信一门当时无人知晓的编程语言，也就是如今的 Java。 Then he spent the next decade convincing 5 million developers to believe in a programming language nobody had heard of, now called Java.

再之后是 25 年创办公司的历程：作为创始人融资 5000 万美元，并以 3800 万美元卖掉了其中一家公司。 Then 25 years building companies, raising $50 million as a founder, selling one company for $38 million.

而如今，他管理着一支拥有五家独角兽的基金。 And now he manages a fund with five unicorns in it.

大多数人的职业生涯都在追逐正在发生的事。 Most people spend their career chasing what is happening.

而我的这位嘉宾，却总是提前十年到场，然后静静等待。 My guest has spent his showing up 10 years early and waiting.

所以我想问他的，并不是他如何走到今天。 So the question I want to ask him is not how he got here.

而是这一切在他身上留下了什么。 It is what all of it left behind.

因为如今，一位神经科学家、一位布道者、一位创始人和一位募资人，都坐在同一把椅子上。 Because a neuroscientist, an evangelist, a founder, and a fundraiser are all sitting in the same chair now.

而当你走进来做路演时，他们每一个人都在注视着你。 And every one of them is looking at you when you walk in into a pitch.

Miko Matsumura，欢迎来到《雏形》。 Miko Matsumura, welcome to Prototype.

我想这一期的目标，是去理解你的这段历程：从一个搭建神经网络的人，到成为创始人，再到自己成为一名投资人。 I think the goal of this episode is to sort of understand your journey from someone who built a newer network than a founder and then become an investor yourself.

看看所有这些过往经历，最终如何塑造了今天作为投资人的你。 How all of these past experiences eventually become who you are today as an investor.

这档播客的名字叫《雏形》。 The title of this podcast is like Prototype.

也许这一期很特别，我们要探寻的，是你自己的那些“雏形版本”如何一步步成为了投资人。 And maybe this episode is very unique, finding how the prototype versions of yourself become an investor.

挺有意思的。 That's interesting.

我真的很感激这番介绍。 I definitely appreciate that.

嗯。 Yeah.

你希望我从哪里讲起呢？ Where would you like me to start?

好。 Yeah.

我们就从最开始的地方讲起吧，从你的起点开始。 Let's start at the very beginning where you start.

如果我没记错，1990 年你在耶鲁大学搭建了一个神经网络。 In 1990, you built a neural network at Yale, if I'm correct.

对，没错。 Yeah, absolutely.

你当时是一名神经科学家。 You're a neuroscientist.

我很好奇，那时候你在追寻的是什么？ I'm just curious, what were you chasing back then?

你的目标又是什么？ What were your goals?

嗯。 Yeah.

说真的，如果坐上时光机回到过去，那时我就已经在自己的卧室里，用一台 8 位电脑编写 AI 软件了。 I mean, really, when you go back in the time machine, I was already building AI software in my bedroom on an 8-bit computer.

我的第一台是雅达利 400，只有 16K 的 RAM，我同时用机器语言和 Atari BASIC 来编程。 My first one, Atari 400, with 16K of RAM and using both machine language as well as Atari BASIC.

有意思的是，我基本上做出了一个能自我修改的学习模拟程序。 And, you know, basically, I built, interestingly, I built the self-modifying learning simulation.

所以我那时做的，算是认知主义流派的机器学习，大概是在上世纪 80 年代读高中的时候。 So, I was doing kind of cognitivist machine learning back then, and this was kind of in high school, back in the 1980s.

真的是非常非常早，回到了个人计算机刚刚起步的年代。 So, really, way, way back in the origins of the beginnings of personal computing.

说来好笑，我做了一个游戏，叫“动物猜猜看”(Animal Guesser)。 And so, you know, I built, it's funny, I built this game that I call Animal Guesser.

它基本上是玩“二十个问题”的游戏，机器通过一连串“是或否”的问题来猜你想的是哪种动物。 And it would basically try to take 20 questions, and the machine would try to guess the animal based on yes or no questions.

它只会问你“是”或“否”的问题，对吧？ And it would just ask you yes or no questions, right?

它其实非常原始，因为它本质上就是在做二分查找。 And it was really primitive, because what it would do is it would basically do a binary search.

也就是说，它会尽力每次把剩下的搜索空间砍掉一半，对吧？ So, it would basically make a best effort to cut the remaining search space in half each time, right?

它会去查看自己已知的那份可能性清单，对吧？ So, you know, so it would look at the list of possibilities that it knew about, right?

但我当时觉得它真正的创新之处在于，这个程序用上了 Atari BASIC 里一个奇特的“后门”，也就是自我修改代码。 But the thing that I felt like was innovative about this at the time was that the program actually used this weird escape hatch in the Atari BASIC, which was self-modifying code.

结果就是，每一轮结束时，这段代码，这个软件，真的会重写它自己。 So, what happened is that the code would actually, the software would actually rewrite itself at the end of each round.

因为一旦它猜输了，它就会说：我不知道长颈鹿是什么。 Because what would happen is, is that if it would ever lose, it would be like, I don't, you know, like, I don't know what a giraffe is.

它会问：你能不能帮我想一个“是或否”的问题，把它和斑马区分开来？ Like, could you create me a yes or no question that would distinguish this from a zebra?

诸如此类吧，然后它基本上就会把这个记下来。 Or whatever, like, whatever, you know, and so what it would do is it would basically memorize that.

长话短说，我在 1990 年写的那个连接主义的误差反向传播网络，并不是我第一次接触机器学习。 So, you know, long story short, like, you know, the connectionist error backpropagation network I wrote in 1990 was not my first machine learning.

显然，那些早期的机器学习大多是我自学的，真的差不多就是个孩子在玩耍而已，对吧？ Like, you know, but obviously, a lot of that early machine learning was sort of self-taught and, you know, really just almost like a kid just playing, right?

就是很单纯地异想天开：我能不能让计算机做点有意思的事？ So, just really trying to dream up, can I make the computer do something interesting, right?

所以那些都是我在很小的年纪就自己捣鼓出来的，对吧？ So, that was, you know, doing it on my own, you know, at this pretty early age, right?

但后来真正发生的，是连接主义革命，以及复杂性理论等思潮的兴起。 But, you know, eventually, what happened was really the connectionist revolution, and it was the rise of things like complexity theory.

1986 年那篇重磅论文出自鲁梅尔哈特和辛顿，也就是那篇误差反向传播的论文，对吧？ And, you know, in 1986, the big paper was Rumelhart and Hinton, which is the error backpropagation paper, right?

那是一篇非常重要的论文。 So, that was a really big paper.

所以到了 1990 年，我们都纷纷加入了这股潮流。从某种意义上说，鲁梅尔哈特和辛顿当时真正推动的，是一场变革：让神经科学从认知主义模型转向连接主义模型，对吧？ So, you know, by 1990, we were all jumping on the bandwagon, you know, and, you know, in a sense, what Rumelhart and Hinton at the time were really pushing, was a revolution away from the cognitivist model of neuroscience and towards a connectionist model, right?

而真正有意思的是，当我们进入现代 AI 时代，令人着迷的一点在于：GPT 这类系统的核心，依然是同样的机器学习误差反向传播。 So, the thing that's so interesting becomes that, as we come into the modern era of AI, it's fascinating because the heart of GPT class systems is the same machine learning error backpropagation.

但与此同时，新加入的东西，当然，就是 Transformer，这一点意义重大，对吧？ But, at the same time, what's added, of course, what's added is transformers, which is huge, right?

所以机器的“注意力”机制，是那个年代以来最大的变化。 So, machine attention was the biggest change since those days.

但与此同时，我认为真正凸显我们今天所处阶段的一点，而这一点其实广为人知却又鲜被提及，就是：那些鼓吹持续学习之类概念的人，往往执迷于“灾难性遗忘”这个想法，也非常执迷于所谓“神经可塑性内核”的概念。 But, at the same time, the thing that really, I think, accentuates where we are today, I think that's very understood and underrepresented is that people who are touting things like continuous learning are often obsessed with this idea of, like, catastrophic forgetting, and they're really obsessed with this idea of, like, a neuroplastic core.

他们如此痴迷于神经可塑性，是因为大脑本身在本质上就是可塑的。 So, they're really obsessed with neuroplasticity because the brain itself is essentially neuroplastic.

不过我要说，大脑的神经可塑性当中，有一些核心成分是会自我保护的，这也许意味着，人体内部其实存在一个不具可塑性的内核。 Although, I would say that there are core elements of the brain's neuroplasticity that are self-defending, and that may actually have, there may actually be a non-neuroplastic core inside of a human.

我给你举个例子吧：很多人都怕蜘蛛和蛇，对吧？ So, let me just give you an argument, right, which is a lot of humans are afraid of spiders and snakes, right?

还有一些你可以称之为“记忆”的东西，比如和饥饿感相关的反应。把这些当成记忆听起来挺好笑的，但它们其实可以被看作是祖先遗传下来的记忆，或者说本能，对吧？ And so, you know, there are also things that you could call memories that are related to things like getting hungry, or like, you know, and it's funny to think of these as memories, but they could really be thought of as ancestral memories or instincts, right?

而当你去看一台机器时，机器通常被划分为 RAM 和 ROM 这两种概念。 And so, when you look at a machine, machines generally get separated into kind of the idea of RAM and ROM.

在早期的计算机时代，ROM 就是只读存储器，基本上是随机器一起出厂的，对吧？ So, back in the olden days of computers, ROM was like read-only memory, and it would basically come shipped in the machine, right?

很多时候，操作系统之类的东西会被烧录进只读存储器里，你买了机器之后甚至根本无法更新它，对吧？ So, a lot of times, things like operating systems would rarely be burned into read-only memory, and you wouldn't even be able to update it after you bought the machine, right?

那真是非常非常原始的年代。RAM 和 ROM 这套说法其实挺好玩的，而且命名也很奇怪：其中一个是随机存取存储器(RAM)，实际上指的是可读写的存储器；而只读存储器(ROM)，则是你无法写入的存储器，对吧？ Really, really primitive days, but, so the idea of RAM and ROM was really funny, because it's sort of, and weirdly named, like one of them was random access memory, which is actually the read-write memory, and then read-only memory was you couldn't write the memory, right?

但它终归也是一种存储器。 So, it was a memory.

所以我想说的重点是：现代 GPT 这类系统基本上都具备注意力机制，这意义重大；但另一个非常有意思的地方在于，它们出厂时就自带了一个极其庞大的所谓“只读存储器”，而不是可读写的，对吧？ So, the thing that I'm making is this, right, which is that modern GPT-class systems basically have attention, which is huge, but the other thing that's really interesting is, is that they kind of ship out of the box with a very large so-called read-only memory, not read-write, right?

所以它是不可更新的，因此它们有一个学习截止日期，而这正是那个字母 P 的含义，也就是预训练(pre-training)，对吧？ So, it's not updatable, and so they have a learning cut-off date, and that's what the P is, pre-training, right?

就是这个意思。 That's what that is.

所以我觉得，很多人把这一点误解成了一个问题，对吧？ So, I feel like a lot of people are really misunderstanding this as being problematic, right?

但在我看来，这并不是问题。 Because to me, I think it's not problematic.

我相信人类出厂时也自带“预训练”，因为大多数人都莫名其妙地怕蜘蛛、怕蛇，对吧？ Like, I believe the human ships with pre-training in the sense that most humans are kind of weirdly afraid of spiders and weirdly afraid of snakes, right?

而这些全都是本能，对吧？ And these are all instincts, right?

所以人类天生就带着某种相当固定、无法更改的基础记忆系统。有人会说，哦，存在灾难性遗忘啊，存在全局性的神经可塑性啊，可是，人们并不会忘记蛇会咬你、你可能因此丧命这件事。 And so, humans come with some base memory system that's pretty immutable, and people say, oh, there's catastrophic forgetting, and there's, like, global neuroplasticity, and it's, like, people don't forget that a snake can bite you and you can die.

这是你绝不会忘的东西。 Like, it's nothing you forget.

从来没有人忘记过这一点。 Like, no one's ever forgotten that.

就连海马体受损、或者严重脑损伤的人，依然会怕蛇，因为这是一种如此根深蒂固的记忆。 Like, even people with hippocampal lesions or, like, severe brain damage are still afraid of snakes because of such a deep memory.

所以我想说的整体观点是：回顾所有这些系统的历史，在 1990 年那会儿，我们真的在竭力把钟摆从认知主义者那一边扳回来，对吧？ So, I guess what I'm arguing on the whole is that when you look at the, you know, history of all of these systems, at the time, in 1990, we were really struggling to swing the pendulum back from the cognitivists, right?

认知主义者基本上就是麻省理工那边的明斯基和帕佩特等人，对吧？ Which are basically like Minsky and Papert over at MIT, right?

而那些“反革命者”基本上就是鲁梅尔哈特和辛顿，对吧？ And the counter-revolutionaries were basically Rumelhart and Hinton, right?

但真正令人着迷的是，我认为如今这套“支架”(harness)既代表了出厂之后的神经可塑性，也代表了这套系统中认知主义和神经符号主义的那一面。 But the thing that's really fascinating is that I believe that the harness currently represents both post-ship neuroplasticity, but it also represents the cognitivist and neurosymbolic aspect of this system.

所以，当人们谈论 AI 时，他们已经把 AI 和 GPT 混为一谈；而我觉得更糟糕的是，他们还把 AI 和他们所谓的 LLM 混为一谈，可他们其实指的是 GPT，对吧？ So, when people talk about AI, people have come to conflate AI with GPT, then I think what's even worse is that they've come to conflate AI with something that they're calling LLM when they mean GPT, right?

LLM 本身根本什么都不是。 LLM isn't anything.

所以把某个东西叫做 LLM 是很傻的，因为这只是对一个事物的描述而已，对吧？ So, like, calling something LLM is dumb because it's just a description of a thing, right?

而且它甚至算不上一个好的描述，对吧？ And it's not even a good description of a thing, right?

就说什么叫“大”吧，对吧？ In the sense of what is large, right?

那 BERT 算大吗？ So, is BERT large?

我也说不清。 Like, I don't know.

总之长话短说，我觉得人们低估了这套“支架”的重要性。 So, anyhow, long story made short is I think people are under-emphasizing the harness.

我认为 AI 的这套支架既是神经符号主义的，也是具有神经可塑性的。 And I think that the AI harness is both neurosymbolic and neuroplastic.

所以对于那些只会说“哦，AI 不行”或者“LLM 不行”的人， And so, I think people who are just like, oh, AI is bad or LLMs are bad.

当然，像杨立昆这样的人非常聪明，他说的也有道理。 You know, obviously, like, Yann LeCun is super smart and he has a point.

但话虽如此，我其实认为 GPT 还会走得很远，因为人们正逐渐意识到，这套支架是整个方案的一部分，是协同演化格局的一部分，而这个格局的核心，本质上就是一个 GPT。 But that being said, I actually think GPTs are going to go a long way, you know, because people are learning that the harness is part of the package and it's part of the co-evolutionary matrix, which has at its core, essentially, a GPT.

而当你审视 GPT 时会发现，早在 1990 年，误差反向传播显然就已经给了我们预训练，而那些也都是生成式模型，对吧？ And when you look at GPTs, in 1990, backpropagation of error gave us pre-training, obviously, and those were generative models, right?

所以我们当时唯一缺的，就是 Transformer。 So, the only thing we didn't have was transformers.

那是我们唯一缺的东西。 That's the only thing we didn't have.

但它的意义确实重大。 And so, you know, but it was big.

Transformer 改变了一切。 Like, the transformers changed everything.

所以如今，我们身处一个全新的世界。 So, now we're in a completely new world.

嗯。 Yeah.

你也曾是一位神经科学家。 You were also a neuroscientist.

你觉得真正的 GPT、大型语言模型，或者任何形式的 AI，应该在多大程度上去模仿真实的人脑？ How much do you think that a actual, you know, GPT or LLM or any kind of AI should mimic the actual human brain?

因为我觉得很多创新其实都源于此：某种模型上的突破出现时，人们会说，看，这就是脑细胞的工作方式。 Because I think there's a lot of sort of innovation that comes in when, you know, a certain type of model innovation and how people talk about, oh, that's how a brain cell works.

你怎么看这个问题？ What's your take on that?

嗯，对我来说，进化是极其了不起的，也极具启发性。 Yeah, I mean, to me, evolution is absolutely phenomenal and it's absolutely inspirational.

进化确实创造出了非常惊人的成果。 And, you know, evolution has produced really amazing results, you know.

所以它本身绝对不容小觑，对吧？ So, it's definitely itself nothing to sneeze at, right?

生物神经元是很了不起的，对吧？ So, biological neurons are impressive, right?

而当计算机科学家去构建这些所谓的抽象神经网络，也就是 ANN 的时候， And when computer scientists build these so-called abstract neural networks or ANNs, right?

他们所说的那些神经元其实是非常抽象的，对吧？ Like, the things that they call neurons are very abstract, right?

因为人们往往没有意识到，一个神经元本质上是一种模拟计算，对吧？ Because what people don't really realize is effectively that a neuron is analog computing, right?

所以它的能力是难以置信的。 So, it means that it's unbelievable.

打个比方，说到浮点数，它几乎可以把小数点后的精度延展到近乎无穷，对吧？ Like, when you talk about things like floating points, it's like you can basically float the points out to almost infinity, right?

所以，如果你谈到像模型量化这样的东西， So, like, the, you know, if you talk about things like model quantization, right?

单单一个生物神经元，所承载的信息量简直是天文数字，对吧？ Like, a single biological neuron has, like, cosmological amount of information, right?

所以这完全是另一个层级的东西，对吧？ So, it's a really, really different ballgame entirely, right?

话虽如此，我认为仿生学应该成为一种灵感来源。 But that being said, you know, I think that biomimicry should be a source of inspiration.

我觉得我们完全应该去研究，进化在大约至少三十亿年的复杂神经系统演化中做成了什么。 And I think that we should absolutely study what evolution did in probably at least 3 billion years of complex nervous system evolution.

当然，也许没那么久，对吧？ And I mean, maybe less, right?

因为多细胞生物的历史可能都没那么久远。 Because multicellularity might not be even that old.

大致就是那个年代。 It's about that old.

所以我会说，进化在这段时间里已经做成了非常多的事。 So, you know, I would say it's done a lot in a pretty good amount of time.

总之，我认为我们应该从中汲取灵感。 But, you know, I think we should take inspiration from it.

是啊，我之前在香港的 LeapEast，那里有些人做的技术，算是从实验室里走出来的成果。 Yeah, I was at Hong Kong LeapEast and there were people who actually, sort of like a technology that came out of the lab.

他们用真实的脑细胞，把它装进了一个盒子里。 They used physical brain cells and put it in a box.

然后把它当作 GPU 来用，去替代普通的 GPU，这挺有意思的。 And they would really use that as a GPU to replace the normal GPU, which is kind of interesting.

是啊，这东西相当了不起。 Yeah, it's pretty phenomenal stuff.

我确实觉得它很迷人。 I mean, I definitely feel like it's fascinating.

对我来说，计算领域里有一条一直成立的规律：专用计算的表现往往优于通用计算。 Obviously, to me, one of the things that's always been true in computing is how special purpose computing, you know, outperforms general purpose computing.

从某种意义上说，这其实挺奇特的。 You know, and in a sense, it's really strange.

它取决于你要解决的是什么问题，也取决于这些生物神经元被如何应用，对吧？ It depends on the problem you're trying to solve and it depends on how the biological neuron is being applied, right?

有意思的地方在于，神经元和神经元之间根本不是同一回事。 So, you know, because the thing that's funny is that a neuron is not a neuron is not a neuron.

我认为把神经元当成一种通用的、千篇一律的东西，是非常愚蠢也非常局限的想法。 You know, I think the idea that a neuron is a generic object is very foolish and limited, you know.

我还觉得，人们常常把大脑当成一个铁板一块的单一器官来谈论。 And I feel like people also use this idea of they're talking about the brain as if it were a monolithic organ.

我认为这同样是一个相当荒谬的错误。 And I think that's also erratically, it's a really ridiculous mistake.

这几乎就像把人体内所有的内脏都当成一个单一器官来建模。 It's almost like modeling the entire visceral contents of a human as a single organ.

你可以把它统称为“内脏”，对吧？ And you could call it the guts, right?

也就是说，你躯干里的一切都算作一个叫“内脏”的单一器官。 So it's sort of like everything inside your torso is a single organ called the guts.

而你脑袋里的一切都算作一个叫“大脑”的单一器官，对吧？ And everything inside of your head is a single organ called the brain, right?

这两种说法犯的范畴错误其实差不多严重，对吧？ And it's about equal of a category error, right?

所以当你单独挑出一个神经元来看，比如从网状神经元，一路到海马体、长时程增强、联想记忆这些机制， And so people, when you pick out a neuron, like the number, like if you look at like reticular neurons, you know, up through things like hippocampus, long-term potentiation, associative memory.

你会看到各种各样的结构系统，一直延伸到前额叶皮层，那里有大量的柱状细胞或锥体细胞，还有各种用于传播传入信号的机制，以及抑制性和兴奋性神经元，对吧？ Like you look at things, these different kinds of structural systems all the way up to the prefrontal cortex with lots of kind of columnar cells or pyramidal cells, all these different kinds of mechanisms for spreading afferents and these kind of like inhibitory and excitatory neurons, right?

所以当你开始去看那些树突的结构和连接时， So when you start to look at the dendritic structures and connections, right?

你会发现神经元根本就不是通用的，对吧？ So neurons are not general purpose at all, right?

所以在很多方面，采用这种方法时，你真的得去仔细思考自己在模拟的是哪一类系统，以及那个子系统究竟是为完成什么任务而设计的，对吧？ So in a lot of ways, this type of approach, you really have to kind of reason about what type of system you're modeling and, you know, what is that subsystem designed to do, right?

但就我们眼下这场复兴而言，我真正的感受是，人工注意力显然是一个非常了不起的想法。 But what I really feel like where we are in terms of our current renaissance is obviously artificial attention was a really great idea.

而把这些基于统计的自回归模型，去对撞软件开发这样刚性的结构，事实证明是个绝妙的主意，因为你让某种统计性的、软糊糊的东西，和某种完全确定性的、像数学一样坚硬的东西，形成了协同进化，对吧？ And that pushing these statistically autoregressive models against rigid structures like software development turns out to be a really great idea because you have this co-evolution of something that's statistical and squishy pressing against something that's completely deterministic and like hard as math, right?

它极其刚硬，对吧？ It's brutally hard, right?

要么代码能编译，要么不能；要么检查通过，要么不通过；要么 QA 测试通过，要么不通过。计算领域里的一切都是这样非黑即白、确定无疑的。 So it's either the thing compiles or it doesn't either the, the checks pass or they don't either the QA test pass or they don't either the, so everything in computing is like super hard and deterministic.

所以它妙就妙在，会形成一个自催化的学习闭环。因为在软件里，你要是错了，那就是彻底错了。 And so, you know, the thing that's lovely about that is it creates an auto catalytic learning loop in the sense that, you know, in software, if you're wrong, you're very wrong.

而你要是对了，那就是完全对了。 And if you're right, you're very right.

这就给了模型实实在在可供学习的东西，对吧？ And that it gives the guy stuff to learn, right?

而在社交媒体这类东西里，几乎无所谓对错，甚至连真相都难以谈起，对吧？ Whereas in things like social media, it's almost like there's no right or wrong and there's barely even truth, right?

所以某种程度上，我们到了这样一个节点：我认为 ChatGPT 五系列的模型，训练成本大概是最高的，而且它读过的互联网内容也比以往任何模型都多。 So, you know, in some ways we reached a point where I think chat GPT five series model was probably maybe the most expensive training and it had read the most internet of any model.

但到了某个阶段，人们发现它读的互联网内容多到一定程度，反而变得比上一代模型更笨了，对吧？ And at some point people determined that it had read so much more internet that it actually became dumber than the previous model, right?

也就是说，一味地去对撞、去学习人们在网上所说的东西，最后发现这些内容本身就是软糊糊、含糊不清的。 Which is just, you know, so pressing against what people say on the internet and learning what people say on the internet turns out to be squishy.

所以如果你给软糊糊的模型喂软糊糊的素材，它们表现就是不好。 So if you give squishy models, squishy things to look at, they just don't do well.

反过来，如果你给软糊糊的模型一些非常刚性、对错分明的任务去处理，从某种意义上说，这正是 Claude Code 能大幅领先于 ChatGPT 的原因所在。 Whereas if you give squishy models really rigid things to work on, you know, with very clear right and wrong, you know, and that's kind of in some ways the lesson of how Claude code got so far ahead of chat GPT.

我认为 OpenAI 正在追赶，我觉得 5.6 很出色。而且某种程度上，Claude 如今展现出一种进化上的严肃劲头。 I think OpenAI is catching up, I think 5.6 is impressive and I think, in some ways, there's an evolutionary seriousness that Claude has now.

我觉得最新的 Claude 模型也许过于谨慎了。现在的情况是，人们喜欢 ChatGPT 的速度，恰恰因为它不那么谨慎。 I think the newest Claude models are maybe overly cautious and I think what's happening now is people love the speed of chat GPT because it's incautious.

它常常也能输出正确的代码。 So it's oftentimes outputting correct code.

它是在更少的防护栏下完成的，因此速度快得多，成本也更低，对吧？ It's doing so with fewer guardrails and as a result, it's doing so much, much faster and with less cost, right?

因为你跑的测试越多、越谨慎、越是反复自我检查，烧掉的 token 就越多。 Because the more tests you run and the more cautious you are and the more you check yourself, you know, the more tokens you'll burn.

所以我觉得眼下这个钟摆也许正在摆回去，但可能摆得太早了。 So, you know, I think right now the pendulum might be swinging, but maybe prematurely.

我觉得 Claude 由于会自我批评和自我验证，速度更慢、烧的 token 更多，但它相当可靠。 Like, I think Claude, because of its self-criticism and validation, you know, it's slower and it burns more tokens, but it's pretty reliable.

嗯。 Yeah.

我还想聊聊你接下来的两段经历，也就是你离开大学之后所做的事。 I also kind of want to touch on your next two, what you've done after you sort of leave college.

你最初是做 Java 的技术布道师。 You started off as an evangelist for Java program.

嗯。 Yeah.

之后你成为了一名创业者。 And then you became a founder.

首先我有个问题：作为 Java 的技术布道师，你具体是做什么的？ First of all, just a question, like, what do you exactly do as an evangelist for Java program?

我相信很多人可能并不清楚这到底意味着什么。 I'm sure that a lot of people might not understand what that exactly means.

嗯。 Yeah.

其实，开发者布道这个概念真正为人所知，要归功于另一位日裔美国人盖伊·川崎，他当年是苹果 Macintosh 的布道师，对吧？ So really the idea of developer evangelism was effectively put on the map by another Japanese American dude, Guy Kawasaki, who was the evangelist for the Apple Macintosh, right?

而布道理念的一部分，基本上就是运用他所说的“热忱与狂热”，对吧？ And part of the philosophy of evangelism is basically using what he called fervor and zeal, right?

所以本质上，一旦你借用“布道”这个词，它显然会让人联想到宗教里传教、劝人皈依的概念，对吧？ So, you know, essentially what's happening, you know, obviously when you take the terminology of evangelism, it kind of goes back to the idea of proselytization and religion, right?

也就是说，你其实是把一个开发平台和一套工具，当成一场宗教皈依来对待，对吧？ So you're basically taking sort of a development platform and tools, and you're basically treating it like it's a religious conversion, right?

所以某种程度上，这是个很有趣的看待方式，因为当你审视编程语言和平台时，它们确实会演变成某种深层的哲学信仰体系。 So that's really the, you know, in some ways a fun way of looking at it, you know, because really like when you look at programming languages and platforms, you know, it really does become this kind of deep philosophical belief system.

它承载着大量的情感能量，不过我觉得如今这种色彩淡了很多。我认为我们现在进入了一个多语言并存的时代，语言几乎已经不重要了，因为我们步入了一个新时代：新的“无代码”恰恰就是代码本身。 And it comes with a lot of emotional energy, I think much less so now, like I think now we're in the era of mass polyglot and languages almost don't matter because we're, you know, we're in the new era where the new no code is code.

所以基本上，如果你的编程模型想用某种语言，你大概就该让它用。 So basically, you know, if your coding model wants to use a certain language, you should probably allow it.

无论是某些代码库、开源项目，还是某些它偏好的语言，对吧？ You know, in certain packages and open source or in certain preferred languages, right?

所以总体上，你顺其自然、做一个多语言通吃的人就好，对吧？ So in general, you should just kind of go with the flow and be polyglot, right?

但我觉得在当时，情况是这样的：开发者们清楚，一旦他们投入到某种语言和平台上，就等于把自己大脑的很大一部分，以及职业生涯中很大一块，都押在了上面。 But I think at the time, you know, what would happen is that developers would know that if they invest in this language and platform, that they're investing a pretty large part of their brain and a huge chunk of their professional career on this.

正因如此，人们采纳起来相当缓慢也相当谨慎，他们真正在寻找的，是足以让自己相信的理由。 So, you know, because of that, people are kind of slow to adopt and cautious, you know, and they really are looking for, you know, reasons to believe.

挺有意思的。 That's interesting.

然后你成为了创业者。 And then you became the founder.

你觉得创业者和布道师之间有一个共同点，就是你都得去说服别人相信一件并不那么显而易见的事情。 Do you think that there's, I think there's one similar thing between the founder and evangelism evangelist is that you have to convince people into something that's not very obvious.

你觉得这是不是你后来作为创业者时看到的那种相似之处？ And do you think that's sort of the similarity that you see as a founder eventually?

是的，当然。 Yeah, for sure.

我觉得真正重要的是从零到一，对吧？ I mean, I think the thing that is becomes really important is the zero to one, right?

从某种意义上说，当你思考创业这件事时，创业的过程其实就是把某个存在于想象中的东西变成现实，对吧？ And, you know, in a way, when you think about founding, right, the founding process is really one where you're moving something from the imaginary into the real, right?

所以，如果你回溯到，怎么说呢，这一切的原型范本，那就像是史蒂夫·乔布斯这个范本，对吧？ So, if you go back to, I don't know, like the ERG template for this whole thing, it's like the Steve Jobs template, right?

从某种意义上说，当你对比 iPhone 出现之前和之后的世界，那基本上就是直接从他的想象里蹦出来的，对吧？ So, in a sense, when you look at the world before the iPhone and after, you know, it's very much kind of straight out of his imagination, right?

我觉得，当你回顾这段历程，其中一件非常引人入胜的事情，就是这里存在着某种演化。 So, you know, I think, and when you think about one of the things that's really fascinating, when you look at the journey, is that there is kind of this evolution.

从某种意义上说，它迷人的地方在于，它有着深厚的技术根基，也许还有一张演化的路线图。 And in a sense, the thing that's fascinating about it is that it's characterized by sort of strong grounding in technology, and perhaps an evolutionary map.

但真正有意思的是，人们总把乔布斯称为一位有远见的人。 But the thing that's really interesting is that people really referred to Jobs as a visionary.

我确实认为，当我们思考“远见者”这个概念时，最关键要理解的一点是，他有一句很有名的话：你无法向前把那些点串联起来，只能在回望时把它们串联起来，对吧？ And I do think that when we think about the idea of visionary, I think the thing that's really critically important to understand is that one of his more famous quotes is the idea that you can't connect the dots looking forward, you can only connect the dots looking backwards, right?

所以，当你看 iPhone 这样的东西时，你得往回看 App Store，再往回看 iTunes，再往回看 iPod，对吧？ So, when you look at something like iPhone, you have to look at something like the App Store, and then you have to look at iTunes, and then you have to look at iPod, right?

当你这样沿着时间往回追溯，你会开始看到这些不同子系统留下的奇特演化印记，对吧？ So, as you kind of go backwards through time, you start to see this weird evolutionary imprints of these different subsystems, right?

而在 iPhone 之前，还有一些相当奇特的产品，比如 Newton，当然，后来 iPad 又是从 iPhone 演化出来的，对吧？ And predating things like the iPhone, where there are really weird systems like Newton, and like, you know, and obviously, eventually, iPad came out of iPhone, right?

所以，这一切也许并不是那种“远见”，从我们把某人视为远见者的角度来说，或许更应该着眼于长远，对吧？ So, like, all of these things probably may not have been this kind of idea of a vision, in the sense of when you think about someone as a visionary, in a way, it's probably better to look at the long term, right?

因为他真正锁定的、我认为非常重要、也非常宏大的东西，是个人计算这个理念，对吧？ Because the thing that he really set upon, that was, I think, really important, and really, really big, was the idea of personal computing, right?

所以，他牢牢抓住了个人计算这件事。 So, he really latched on to personal computing.

而他刚接触计算的时候，做的是惠普的办公计算器，不是你揣在口袋里的那种，而是那种很大、摆在办公室里做会计之类工作的机器。 And when he started computing, he was doing like HP office calculators, and not the kind that you have in your pocket, but like, really big ones that would sit inside of like a office and perform accounting and things.

这是非常非常早期的年代，其实是在个人电脑出现之前，对吧？ And this is very, very early days, in fact, before the personal computer at all, right?

所以，当他谈到要把那样的东西变得“个人化”时，从某种意义上说，我们至今仍走在那条轨迹上，对吧？ So, when he talked about making something like that personal, in a way, we're still on that arc, right?

比如，计算机将会嵌进我们的眼镜里，它们已经能识别我们的脸、我们的声音和话语，还能跟我们对话，所以它们正变得越来越、越来越个人化，对吧？ Like, computers are going to, you know, sit in our glasses, they're going to, they already recognize our face and our voice and our speech, and they talk to us, and, you know, so they're getting more and more and more and more and more personal, right?

所以我觉得，从某种意义上说，那条轨迹和那种愿景真的非常非常深刻。 So, you know, I think, in a way, that arc and that vision was really, really profound.

我想说的是，人们也许过分夸大了他对每一个演化步骤的预见能力。 And I would say that people might be over dialing on how prescient he was about each of the evolutionary steps.

但我认为，从某种意义上说，他对最终的目的地是有愿景的。 But I think, in a way, I think he had a vision for the final destination.

但我认为，用他自己的话来说，他在每一步展开的当下并不真正理解它，对吧？ But I think, even in his own words, I don't think he understood each step as it was unfolding, right?

我觉得他更像是在进行一场大胆的探索之旅。 I think he was really more of a bold journey of exploration, you know?

我觉得他大致知道自己想往哪个方向走，但我不认为他知道自己会在哪些岛屿停靠，大概就是这种感觉。 And I think he kind of knew which direction he wanted to go, but I don't think he knew what the islands were that he would stop at, you know, this kind of feeling.

是的，然后你自己最终也成了创业者，创办了一家公司，并最终以三千八百万美元的价格把它卖掉。 Yeah, and then eventually become an entrepreneur yourself, building a company and eventually selling it for $38 million.

我很好奇，亲手打造技术这件事，教会了你哪些关于投资的东西？ I'm curious, what did building technology yourself teaches you about investing?

因为我觉得外面很多风投人其实并没有创业经历。 Because I think a lot of VCs out there, they don't really have a founder experience.

当然也有不少人有，但我觉得大多数风投人是分析师出身，然后一路做到合伙人。 I guess a lot do, but I think the majority of VCs, they came from an analyst background and become an associate partner.

所以我很好奇，相比那些没有这类经历的普通风投人，你多了哪些视角？ So I'm curious, what's the perspective behind that you have that, you know, a normal VC who does not have this kind of experience has?

我觉得如今亲自动手去搭建东西，已经不再是可有可无的了，对吧？ I feel like the process of hands-on building is sort of no longer optional, right?

我觉得，当我们回顾创业的历史，过去确实存在一条相当明确的价值链，大致体现在所谓的五大或四大支柱这个说法里，对吧？ I think that when we look at the history of entrepreneurship, right, there really used to be a pretty strong value chain that was kind of reflected in this idea of like so-called five or four pillars, right?

或者说四大、三大支柱，对吧？ Or four or three pillars, right?

举个例子，经典的价值链模型会是工程、产品、销售、市场营销，对吧？ So for example, the classical model of the value chain would be engineering, product, sales, marketing, right?

然后你也可以说，还有管理层或者说 CEO 这个职能，横跨在这一切之上，对吧？ And then you could arguably say like executive or CEO function that kind of straddles across the top of that, right?

从某种意义上说，这个职能要负责给公司融资、处理投资者关系之类的事情，对吧？ But like in a way, and it is responsible for things like funding the company, you know, and investor relations and things, right?

但总体而言，如今由于智能体编程的革命，人们基本上没有借口了，对吧？ But I would say broadly speaking, these days, because of the agentic coding revolution, there's kind of no more excuses, right?

感觉就是，如果一个人有愿景，那他真的需要拿出点实实在在的东西来，对吧？ And it feels like if people are, if people have a vision, they really need to show something, right?

因为从某种意义上说，在实现愿景的过程中，有一部分就是我们过去所说的研发，对吧？ And because in a way, like part of what happens in the process of realizing the vision is this process that I think used to be called R&D, right?

而我觉得真正令人着迷的，是这些模型在压缩研发周期方面有多么强大，对吧？ And the thing that I think is fascinating is how powerful these models are compressing R&D, right?

所以从某种意义上说，一个只有一个点子的创业者，我确实很青睐最早期的投资，但我觉得如今最早期的投资会孵化出成熟得多的公司和产品，因为那种漫无目的的瞎折腾少了，对吧？ So in a sense, a founder who just has an idea, you know, I think it's true that I'm a fan of the earliest stage of investing, but I feel like the earliest stage of investing is going to output much more mature companies and products, you know, because there's less kind of random banging around, right?

也就是说，我认为一个真正懂某个领域的人，再配上像 Claude 这样的工具，真的能非常快地找到有价值的信号。 Which is, I think if a good, someone who understands a domain with combined with something like Claude can really find signal really quickly.

所以我觉得，在这方面我们已经进入了一个全新的世界。 So, you know, I think, I think we're in a, we're in a new world that way.

所以你认为，对于一个想做早期投资的投资人来说，创业经历真的很重要？ So you think that a founder's experience is really important for an investor who wants to do early stage investing?

你觉得这几乎是必备的吗？ It's almost a necessity, do you think?

嗯，在我看来，这件事有两面。 Well, I, to me, there's two sides of it.

我会说，风投有偏运营的一面，它跟尽职调查有关，也跟决策权有关，对吧？ I would say that VC has sort of an operating side, which has to do with things like diligence, and it has to do with decision power, right?

因为如果你从没运营过公司，从没打造过产品，从没成功退出过，那你对别人做这些事的能力的判断，可能就没那么准，对吧？ Because if you've never operated a company, if you've never built things, you've never exited things, like your judgment about others' ability to do those things may be not as good, right?

所以我觉得，运营经验之所以重要有很多原因，但反过来说，一个人在完全没有任何经验的情况下进入风投这一行的一条路，其实就是找项目，对吧？ So I think that there are a lot of reasons why operating experience is really important, but I would say on the flip side, though, one way that a person can get into the venture business without any kind of experience at all is really sourcing, right?

当然，你可以说找项目其实也可能需要相当扎实的专业知识，对吧？ Which is, if you, now, you could argue that sourcing requires potentially substantial knowledge, right?

比如说，如果你想为生物科技领域找项目，那你可能真的得懂生物化学之类的东西。 Like, for example, if you want to do sourcing for, like, biotech, right, you may actually need to understand things like biochemistry, and you may need to understand things like that.

但事实证明，或许并非如此，对吧？ But it turns out that maybe it isn't true, right?

因为我觉得有一点很有意思：由于智能正越来越商品化，我认为真正在升值的其实不是智能，而是精明，对吧？ Because one of the things that I think is interesting becomes that because intelligence is becoming more of a commodity, I think what's increasing in value is actually not intelligence, but actually shrewdness, right?

而精明，就是一个人只要看一眼对方、或者听对方说几句话，就能立刻看穿：这个人到底信不信自己说的话？ And shrewdness, so shrewdness is really something where a person can just, they can just look at someone or listen to someone, and they can immediately suss out, like, does this person believe in what they're saying?

还是说他们只是想向我推销什么，而他们究竟想卖给我什么？ Or are they just trying to sell me something, you know, and what is it that they're trying to sell me?

他们的真实目的是什么，对吧？ What is their agenda, right?

他们是在故意欺骗我吗？ And are they purposefully deceiving me?

他们是在故意，还是在自欺欺人？ Are they purposefully, are they deceiving themselves?

诸如此类，对吧？ Are they, you know, these things, right?

所以，当你想到智能时，这些其实并不是智能的标志，而是我所说的“精明”的标志，对吧？ That, you know, and so when you think about intelligence, those are not the hallmarks of intelligence, those are the hallmarks of something that I call shrewdness, right?

所以我认为精明正变得越来越有价值，而智能，很遗憾，正变得越来越不值钱，因此像“察言观色”这种听起来有点玄的能力，其实相当重要。 And so I think shrewdness is increasingly valuable, and I think intelligence is decreasingly valuable, unfortunately, you know, so things, weird things like being able to read a room is pretty important.

而这种东西很难教，你很难教别人如何察言观色，我觉得现在越来越常见的一个现象，是会打扑克的风投人，对吧？ And it's kind of hard to teach, it's hard to teach people how to read a room, you know, I think one of the things that's arising more are things like poker playing VCs, right?

因为扑克这种游戏逼着你去察言观色，而那种精明是能派上用场的，对吧？ Because poker or the game forces you to read the room, and, you know, that kind of shrewdness can come in handy, right?

因为你会开始看穿创业者玩的一些把戏，比如说，创业者有时会虚张声势，而风投人真的需要有能力分辨谁在虚张声势、谁没有，谁在说真话、谁没有，所有这些本领，同样更多地跟精明有关，对吧？ Because you can start to see through some of the entrepreneurial games, you know, like, for example, sometimes entrepreneurs bluff, and, you know, VCs really need the ability to figure out who's bluffing and who's not, and, you know, who's telling the truth and who's not, and, you know, all of these types of skills, which, again, relate more to shrewdness, right?

所以，如果一个人天生极为精明，又极为拼劲十足，他其实能在很早的阶段就挤进那些真正了不起的项目。 So if a person is endowed with extraordinary shrewdness, and extraordinary hustle, they can actually get into deals early that are really amazing.

而且他们甚至不一定需要那么多领域知识，对吧？ And they don't necessarily even need that much domain knowledge, right?

因为如果你真的非常非常精明，你会一眼看出：哇，这个项目火得不得了，对吧？ Because if you're really, really shrewd, you can be like, oh, wow, like this, this deal is super hot, right?

那种感觉就像：我完全搞不懂他们到底在做什么，但我就是能看出来它火得很，对吧？ And it's sort of like, I don't know what the hell they're doing, but like, I can tell it's super hot, right?

如果你真的精明，你会看出来：我能感觉到他们也在跟其他顶级风投接触，而如果你真的精明，你的判断就会是对的，对吧？ And if you're really shrewd, you can be like, I can tell they're talking to other tier one VCs, you know, and if you're really shrewd, you'll be right, right?

你不会是那种情况：我只是在酒吧排队时碰到这家伙，被他忽悠得团团转，对他说的每句话都深信不疑，然后觉得他简直是天底下最了不起的人。 You won't be like, they're just, you know, I'm just stood in line at a, you know, bar and this guy snowed me completely, you know, and I totally believe everything he said, and now I think he's the best thing ever.

那样的人是走不远的，但如果你真的精明又肯拼，你其实并不需要什么经验或者太高的智能，对吧？ And it's like, that person won't go far, but like, if you truly are shrewd and you hustle, you actually don't need experience or that much intelligence, right?

而疯狂之处在于，世上存在那种“N 倍标准差”级别精明的人，就是有些人精明得远超常人，却未必给人聪明的印象，这一点很有意思。 And the thing that's crazy is that there are kind of N-sigma shrewdness people, like there are people that are just wildly shrewder than others that may or may not come across at all as intelligent, which is interesting.

嗯。 Yeah.

那么，你会说自己是一位精明的投资人吗？ And then would you say you're a shrewd investor?

这个问题挺有意思。 That's funny.

我想说，我确实觉得自己希望在这些特质之间取得平衡。 I would say that, you know, I definitely feel like I want to balance these qualities, right?

所以我想在精明和聪明之间找到平衡，因为我真的觉得，很遗憾，聪明正在变得越来越像一种大宗商品。 So I want to balance shrewdness with intelligence, you know, because I really think that it's, intelligence is unfortunately getting commoditized, right?

我一直都非常欣赏投资逻辑的构建，也一直非常推崇技术本身。 I mean, I definitely have always been a big admirer of like thesis development, and I've been a big admirer of things like technology, right?

但真正疯狂的一点是，技术的壁垒正在瓦解。当然，我认为深科技仍然相当重要。 But one of the really crazy things is that the technology moats are coming down, you know, obviously, I think deep tech is still pretty important.

当然，也有一些非常复杂的领域，比如硬件，我认为其中涉及大量商业机密。 And obviously, there's really complicated things like hardware that I think involve a lot of trade secrecy, right?

所以，你没法说：哦，我要去问我的 AI 模型，怎么做这件被商业机密层层包裹的事情。 So, you know, it's not like you can be like, oh, I'm gonna ask my AI model how to do this thing that's shrouded in trade secrecy.

我去问它，台积电是怎么做两纳米制程的，结果根本不可能，它没法告诉我。 I'm gonna ask them, how does TSMC do two nanometer production, you know, and it's like, there's no way, it cannot tell me, right?

因为这些内容根本就不在训练数据里。 Like, it's just not in the training at all, right?

根本不可能在里面。 It cannot be, right?

所以我想说的是，竞争优势越来越多地体现在商业机密这类东西上，越来越体现在硬件制造这类领域，还有很多奇特的供应链因素，所以情况确实很复杂。 So, like, I guess what I'm saying is, is that competitive advantage is increasingly things like trade secrecy, and it's increasingly in things like, you know, hardware manufacturing, and there's a lot of, you know, and weird supply chain things, you know, so it's definitely complicated.

我觉得，纯粹的点子这种东西，一直都很容易被复制，而如今这种情况变得更严重了。 You know, I think the pure idea stuff, I think, you know, the ideas have always been pretty copyable, but I think now it's gotten even worse.

对，完全同意。 Yeah, absolutely.

我也想接着聊聊，这些过往的经历最终是如何影响你作为风险投资人的决策的。 And I also, I think, I want to now touch on how all of these past experiences eventually feed into your decision making as a VC.

我记得你说过，你非常看重团队，极度以团队为中心，有时甚至把团队看得比项目本身还重。 I think you said that you are very team, super team centric, and sometimes even over teams over project.

我很想知道这种理念是从哪里来的。 I'm wondering where that came from, because, yeah.

我觉得现实是，当你真正深入研究风险投资的模式时，你会明白，能拿到好估值的关键在于进入得足够早。 I think the reality is, is when you really look hard at the venture model, right, what you learn is you learn that valuations, are gated by being early, right?

所以，如果你真的想创造出风投级别的回报，其实只有两个杠杆。 So, if you really want to produce venture outcomes, there's really two levers, right?

你要么在极高的位置退出，要么在极低的位置进入。 You can either exit really, really high, or you can enter really, really low, right?

或者两者兼得。 Or you can do both, right?

如果你两样都能做到，那你就是真正的赢家。 So, if you're able to do both, then you're a real winner, right?

从某种意义上说，衡量一个风投的标准不是他出手了多少次，而是他打出了多少个本垒打。 And in a way, the number of tries is not the measure of a VC, it's really the number of home runs, right?

所以从某种程度上说，这就迫使你变得更加看重团队，因为在最早期阶段，决定最终结果的最大门槛其实都和人有关。 So, in a sense, like, the thing that that forces you to do is to become much more team sensitive, because, you know, because the thing that happens in the earliest stages is that the biggest gating factors for the outcome are really related to the humans.

而与此密切相关的一个重要因素，就是这些人的心态。 And I think one of the big things that relates to that is the mindset of the humans, right?

因为他们的心态处在所有决策的上游，而这些决策又处在所有结果的上游。 Because their mindset is upstream of all the decisions, which are upstream of all the outcomes, right?

所以从某种意义上说，他们的心态极其重要。 So, in a sense, their mindset is super important, you know?

另外一件同样重要的事，就是他们的动机。 And so, I would say the other thing that's super important is their motivation.

所以，理解人们做事情的原因很重要。 So, in a sense, it's important to understand why people are doing things.

同样重要的是，多少要去了解这些人本身。 And it's also important to understand a little bit about, I guess I would say, the people themselves, right?

关于这一点，已经有人做过相当精彩的总结。 And some of this has kind of been really well characterized.

a16z 的合伙人斯科特·库普尔写了《Secrets of Sand Hill Road》一书。 Scott Kupor, an a16z partner, wrote Secrets of Sand Hill Road.

在书里，他记录了所谓的创始人与市场匹配（founder-market fit）这件事。 And in that, he documents sort of the founder market fit thing, right?

所以从某种意义上说，好创始人并没有一个统一的模板。 So, in a sense, like, there isn't a cookie cutter for what makes a good founder.

更多的是，创始人像是被聘请来完成他们各自的使命，无论那使命是什么。 It's much more a founder is sort of being hired to do whatever it is that their mission is, right?

所以，是他们来陈述自己的使命，也必须由他们来告诉你他们的使命是什么。 So, they're stating their mission and they are the ones that have to tell you what their mission is.

然后你要看，比如说，当你想到红杉那套模板，看那十页幻灯片时。 And then you have to look at the, you know, so when you think about the Sequoia template and you look at the 10 slides, right?

团队固然极其重要，但你几乎总是希望把团队那一页放在这十页里靠近最后的位置。 You know, team is super important, but you almost always want the team slide to be almost at the end of the 10 slides, right?

一般来说，这十页里，你几乎总是把团队放在倒数第二页。 Generally, the 10 slides, you almost always put team as second to the last.

然后通常用财务那一页来提出融资请求。 And you generally use the financial slide for the ask.

而那通常是十页里的最后一页。 And that's usually the last out of 10, right?

当然，整个顺序是可以调换的。 Yeah, I mean, the whole ordering is fungible.

你想怎么排都行。 So, you can have any order you want.

但如果你一上来就先放团队页，那往往意味着你还处在极其早期的阶段。 But, like, if you start with team slide first, it often means you're wildly early.

甚至是连点子都还没有的阶段，这可不是什么好兆头。 And it's even pre-idea stage, which is, it doesn't bode that well.

显然，你这是在妥协让步。 I mean, obviously, you're capitulating.

如果你这么做，你就是在估值上做出让步。 If you're doing that, you're capitulating on valuation, right?

因为你其实是在说：除了团队，我们什么都还没有。 Because what you're really saying is, we don't have anything except our team.

我之所以这么说，是因为尽管团队是最重要的，但在你听清他们究竟要做什么之前，你根本无法评估这个团队到底有多强。 And the reason why I would say that is, is that even though the team is the most important thing, you can't evaluate the team and see how good they are until you've heard what it is they're setting out to do.

所以我喜欢很快就看到问题和解决方案。 So, that's why I like to see, I like to see problem and solution really quickly.

我最喜欢的开场顺序是：问题、解决方案、产品。 And then, and then, so I like, my opener that I like the most is problem, solution, product.

然后我喜欢把团队和财务放在最后收尾。 And then I like team financials as the closer.

中间的内容，我希望用来进一步强化问题和解决方案。 And I like the kind of stuff in the middle to sort of buff up the problem and solution, you know.

但我认为，如果问题和解决方案不能迅速切中要害，不能有力地打动人。 But I think if the problem and the solution don't land quickly, and they don't land hard.

而我觉得人们在路演时最容易忽略的一点，就是他们真的没把问题讲好。 And I think the thing that I think people miss the most on it and when they're pitching is they really miss on problem.

人们不擅长把自己想解决的问题讲清楚，这是一个被白白错过的机会。 Like people are bad at stating the problem that they're trying to solve and that that's a missed opportunity.

我之所以如此执着于对问题的陈述，是因为对问题的陈述深刻地反映出创业者的心态。 I think the reason why I'm so obsessed with the statement of the problem is that the statement of the problem is so deeply reflective of the mindset of the entrepreneur, right?

所以，如果一个人的心态有问题，那么从他思考所要解决的问题的方式中，你立刻就能看出这种错误的心态。 So, if the person has the wrong mindset, they'll immediately be able to see that wrong mindset in the way that they think about the problem that they're solving.

嗯。 Yeah.

不过，你也观察到一个趋势：由于技术的发展，初创公司的规模正变得越来越小。 And then, but you also see a trend where the sizes of startups getting smaller and smaller because of technology.

是的。 Yes.

你觉得呢？现在你会看到大量的 OPC，也就是一人公司。 Do you think, and now you'll see tons of OPC, one-person companies.

你觉得团队这件事在未来还重要吗，或者说还会不会继续重要？ Do you think, do you think that team stuff is still important or will be, will still be important in the future?

嗯。 Yeah.

我的看法是这样的。 I mean, I feel like this, right.

那就是，我确实认为，从长远看，公司规模的中位数最终很可能会变成一个人。 Which is, I do think that in the long run, the median company size will probably end up being one human.

而且我觉得中位数其实并不难达到，因为你只需要有绝大多数都是一人公司就行了。 And I think median is probably like not that hard to achieve because you just need a preponderance of one-person companies, right?

但我认为人们低估了一点：AI 在构建近乎虚拟企业方面会有多么高效。 But the thing that I think people underestimate is kind of how effective AI will be at building almost like virtual enterprises.

它们将能够把这些一人公司组装成相互协作的庞大网络。 So, they'll be able to assemble these one-person companies into very large graphs that coordinate.

因为如果你从协作的角度来看这个问题，会发现：让人几乎终身受雇于同一家公司，这怎么会高效呢？ Because when you think about the problem from the perspective of coordination, it's sort of like, why would it be efficient to have almost like this guaranteed lifetime employment at one company?

然后公司再把你在一个又一个项目之间调来调去。 And then the company kind of shuffles you around from project to project.

这感觉起来并不怎么高效。 Like, that feels like it's not that efficient.

当然，其中最欠缺的一个要素其实就是信任。 I mean, obviously, one of the biggest missing ingredients is really trust, right?

这也是为什么那种模式最终应当占据主导的原因之一。 So, that's one of the reasons why that model should prevail.

因为在一个谁说的话都不能信的模式里，你根本无法有效地协调。 Because in a model where you can't trust anything that anyone says, you can't really effectively coordinate that, right?

所以你没法说：哦对，Mike 这个人特别擅长这类工程。 So, you can't be like, oh, yeah, like this guy, Mike, is really good at this type of engineering, right?

从某种意义上说，如果你只是照字面去看他们的 LinkedIn，就会犯很多错误。 So, in a sense, if you're just reading their LinkedIn at face value, you'll make a lot of mistakes, right?

但如果这个人过去十年一直在你的团队里工作，那你调度他的能力就相当强了。 Whereas if the person has been working on your team for the past 10 years, your ability to coordinate that person is pretty high.

因为你非常非常清楚：好，这类事情我绝不会交给他，因为他一定会搞砸。 Because you really, really know like, okay, I will never put that person on this type of thing because they will fail.

但另一类事情我会百分之百放心交给他。 But I 100% put them on this other type of thing, right?

不过我想说，除非出现，我认为一旦出现像可信证明、评价、评论这类东西，也就是我们大家都在努力构建的、建立在可验证凭证、自主主权且能证明隐私的身份认证以及密码学证明之上的那些东西。 But I would say that, like, barring the emergence, I think once we have the emergence of things like trustworthy attestations and reviews and comments and things of that nature, which, you know, we're all building towards on top of things like verifiable credentials and, you know, self-sovereign privacy-proving identities and cryptographic proofs.

随着我们逐步搭建起越来越多这样的信任基础设施，我们就能协调更大规模的协作网络。 Like, as we start to build more and more of this trust infrastructure, we can coordinate larger graphs, right?

所以对我来说，就我的投资风格而言，我以前特别偏爱那种两人团队。 So, I guess to me, in terms of my investment pattern, I very much used to be obsessed with these sort of two-person teams, right?

比如说，对我们而言，一个典型的原型范例就是 OpenSea 的 Devin 和 Alex 这样的组合。 Like, one of the kind of Earth templates for us would be a team like Devin and Alex from OpenSea, right?

也就是说，你有一个 CTO，再加上一个负责业务的联合创始人或者偏产品型的人。 So, you've got a CTO and you've got like a business co-founder or product type person, right?

两个人紧密无间地一起协作。 The two of them working together joined at the hip, right?

但我认为在如今这个时代，其实更多的是单人创始人的天下。 But I think in this day and age, it's actually much more about solo founders, right?

因为从某种意义上说，你的联合创始人可以是一堆并不完美的代码，而这个联合创始人其实能做的事情相当多。 Because in a sense, like, your co-founder can be flawed code and, you know, that co-founder can actually do quite a lot.

嗯。 Yeah.

你刚才提到了信任这一点。 And then you touched on the part of trust.

我记得你还说过，只有那些坚韧、讲道德、又公平的创业者，最终才能成功。 And I think you've also said that only tough, ethical, and fair entrepreneurs eventually make it.

坚韧这一点很好理解，因为创业本身就很难。 Tough makes sense because startup is hard.

但道德又是从哪里发挥作用的呢？ But where does ethics also come into play?

因为我想很多人会说，道德不过是投资人挂在嘴边的漂亮话。 Because I guess a lot of people would say, you know, ethics is a nice thing that investors always say.

甚至有人会说，太讲道德，有时反而会拖累一家公司或一个创业者。 Or some people even argue that, you know, being too ethical sometimes can hurt a company or entrepreneur.

你怎么看？ But what do you think?

我是这么看的。 I mean, let's put it this way, right?

那就是，我这个人是站在自己收益的角度来看问题的。 Which is that I'm biased towards my outcomes, right?

从某种意义上说，Mark Andreessen 曾引用过所谓的大五人格特质。 So in a sense, when you look at, I think Mark Andreessen has cited the sort of big five personality traits, right?

其中一项特质涉及所谓“顺从性”这个概念。 And one of the kind of elements of the traits have to do with this idea of, like, compliance, right?

但从某种角度说，早期投资特别难的一点在于，你不得不在极其细微的信号上下重注。 But in a way, like, and one of the things that's incredibly hard about early stage investing is that you're over dialing on very microscopic signals, right?

举个例子，要想跑出一个风投级别的回报，公司不仅要有独角兽量级的经济体量，而且创始人通常还握有坑投资人的能力。 Like, for example, in terms of producing a venture outcome, not only do you have to have unicorn scale economics, but the person generally has the power to screw over their investors, right?

而且这是普遍情况。 And in general, right?

事实上，他们越是强大，能坑投资人的能量也就越大。 And in fact, the more powerful they become, the more power they have to screw over their investors, right?

而理想情况下，他们并没有很强的动机去这么做。 And ideally, they don't have a high motivation to do so.

当然，如果某个投资人一直在起破坏作用，那他也许会被挤出局；但换个角度看，如果公司表现不佳，新进的资金把老投资人的份额压价稀释掉，这也是可以理解的。 I mean, obviously, if the investor sort of was actively destructive, then maybe that person could get squeezed out, you know, but in a way, like it feels and, you know, and it's understandable if the company is kind of underperforming, that maybe the new money crams down the old investors and there's dilution, right?

我们不喜欢的，是那种一方赢一方输的局面，也就是赢输或输赢。 It's the scenario that we don't like, which is win-lose or lose-win, right?

我觉得在加密领域，常常会出现风投把币抛售给散户之类的情况，全是那种很糟糕的输赢或赢输的局面。 So I think in crypto, there's sort of like VCs dumping on retail or there's all these really bad lose-win or win-lose type scenarios.

这些都没什么吸引力。 And none of those are appealing, right?

真正有吸引力的是双赢。 What's appealing is really win-win.

而我觉得最难接受的，是你很早就押注了一位后来非常成功的创业者，结果他反过来给了你狠狠一脚。 And the thing that I think is hard to take is if you bet really early on a really successful entrepreneur and then they kind of kick you in the guts, right?

我们很幸运，还没碰到过这种事，但这也正是我在筛选时会偏向看重道德的原因。 And we've been blessed not to have that happen to us, but like, you know, that's kind of why I have a selection bias around ethics, right?

因为，温克莱沃斯兄弟起诉扎克伯格那桩官司里流传出一句话，据说他当时在一个私密聊天室里，那些聊天记录后来被法院传唤取证了。 Because, you know, in a sense, one of the quotes that came out of the Winklevoss lawsuit against Mark Zuckerberg is, I think he was in a chat room, private chat room that got subpoenaed.

他大概是这么说的：你知道“不道德”和“彻头彻尾违法”之间那块灰色地带吗？ And his comment was something like, do you know the gray area between unethical and downright illegal?

他说：我就活在那儿。 He's like, that's where I live, right?

这话说得挺离谱的。 So that's a weird thing that he said.

而且他确实对自己的联合创始人之一爱德华多·萨维林做过一些出名的事。 And obviously, you know, he's sort of notably done things to like Eduardo Saverin, one of his co-founders, you know?

所以他确实是那种拼劲十足的创业者，也毫无疑问地取得了巨大的成功。 So he's definitely like a scrappy entrepreneur and obviously wildly successful, right?

但其中有些道德层面的做法是有问题的。 But some of the ethics are questionable, you know?

不过换个角度看，即便在那样的情形下，扎克伯格也没有坑他的早期支持者彼得·蒂尔。 And, you know, in a way, when you look at, you know, and obviously, like, even under those circumstances, Mark Zuckerberg did not screw over Peter Thiel, who was an early backer, right?

所以他大概拿到了……可以说，Founders Fund 基本就是靠那笔退出起家的。 So like, you know, so he got probably the, you know, that's a founder's fund is essentially based on that exit, right?

正是那一笔，才成就了今天的 Founders Fund。 Like, that's what made Founders Fund into Founders Fund.

所以总体来说，这算得上是一条核心原则。 So I would say, broadly speaking, you know, that's a kind of a core, right?

这是从整个风投行业的角度看，而不只是我个人的角度。 From a venture perspective, not just my own, right?

那就是，我们确实希望创业者具备商业道德。 Which is we do like people to have, you know, business ethics, you know?

当然，这个出发点其实相当自私、贪婪、以自我为中心。 And obviously, that's a very kind of selfish, greedy, and self-centered perspective on it.

另一种我认为同样成立的看法是，如果摘下职业投资人这顶帽子，我觉得，和有商业道德的人合作，是有可能建设一个更好的社会、做出更好的产品和服务的。 You know, the other way of looking at it that I think is also legitimate, you know, taking off the professional hat is essentially that, you know, I think it's possible to work with people who have business ethics and to build a better society and better products and services, you know?

对。 Yeah.

而当我说“摘下职业投资人这顶帽子”时，你可别搞错了：“投资人”前面那个“职业”，永远意味着管的是别人的钱。 You know, and so I think that's, that's the taking off, and when I say taking off the professional investor hat, like, it's, you shouldn't make any mistake that the word professional in front of the word investor always means other people's money, right?

因为如果你投的是自己的钱，那你就不可能是“职业”的，这在逻辑上说不通。 Because if you're investing your own money, you cannot be a professional because it's illogical.

因为投资要怎么拿薪水？无非是从管理的资金里抽取费用，可要是那笔钱是你自己的，这就讲不通了。 Because how do you get a salary from investing is you take fees from your own money, like, it doesn't make sense.

根本不存在不管别人钱的职业投资人。 Like, there is no professional investor who doesn't have someone else's money.

这本就是这个词的定义。 Like, that's just the definition of it.

所以，如果这样来定义，它就像职业体育一样。 So, if you define it that way, it means it's like a professional sport, right?

也就是说，你必须一心求胜。 Which is, it means that you have to be driven to win, you know?

当然，风格各不相同。 And obviously, there are different styles, right?

确实存在各种不同的风格。 You know, there's definitely different styles.

有些人是不惜一切代价也要赢，也有些人是在既定的比赛规则之内去赢。 There's some people who are, like, win at all costs, and there's other people that are, like, win by playing the rules of the sport that you happen to be playing, right?

这是两种不同的心态。 And those are different, you know, mindsets.

嗯，所以创始人可以坑风投，但反过来，风投也可以坑创始人。 Yeah, and then, so, a founder can screw up a VC, but, you know, a VC can also screw up a founder.

我觉得有一个很大的话题是，很多公司把融资当成了终点，当成了头条新闻，或者说当成了最终归宿。 I think a big topic is, sort of, a lot of companies treat funding as their final stage, or as the headline, or as the final destination.

但你之前也提到过，有些公司根本就不适合拿风投的钱。 But you've also mentioned before that some companies just are not suitable for VC fund.

如果你的公司需要四到七年就能实现规模化，那没问题。 If you're a company that takes 4 to 7 years to scale, you'll be fine.

但如果你做的技术天生就要花十年、二十年，甚至三十年才能成熟，那就不是个好的对象。 Yeah, but if you're a technology that naturally takes 10, 20, maybe even 30 years to mature, it's a bad candidate.

然而很多这样的公司却仍然在追逐风投资金。 But a lot of companies like that are still chasing a venture fund.

对。 Yeah.

你觉得，实际上究竟有多少公司其实并不适合拿风投的钱，却仍然在拿？ What would you say is the actual, like, how many companies out there are not suitable for venture fund, but they're still taking them?

哎呀。 Oh, man.

我想说，这个问题的关键在于，答案几乎是“几乎全部”。 I mean, I would say that the problem with this question is that it's really almost all, right?

但现实是，这一点对整个投资资产类别都成立。 But the reality is that that's true of the entire investment asset class, right?

因为，如果你是挑选风投级别项目的人，你的命中率也就是一成。 Because, you know, if you're a venture class picker, you're hit, you're batting 100, right?

差不多是十次里中一次。 You're batting, like, 1 in 10, right?

这就是风投这个级别的水准。 So that's venture class, you know?

如果你能做得比这更好，那我就不知道该怎么说了。 And if you do better than that, then, you know, I don't know.

打击率这个说法其实挺有意思，因为在风投里，每一次命中都应该是一记本垒打。 I mean, batting average is funny because every hit should be a home run, right?

也就是说，风投里的“命中”并不是那种一垒安打。 Meaning that a hit in venture isn't a base hit, right?

如果你想想棒球运动员的打击率，他们打出一记一垒安打也算进打击率里。 So if you think about a batting average for a baseball player, they can get a base hit and it's part of the batting average, right?

但在风投里可不是这么算的。 But it doesn't work that way in VC.

所以我想说，你真去算这笔账就会发现，真正划算的，只有全力挥棒、争取本垒打。 So I would say when you really do the math, like, it really does, you know, it really only pays the swing for the fences.

至于说有多少比例的公司其实不该拿风投，从某种意义上讲，把这些筛掉的责任是在风投这一方。 I think in respect to do, what's the percentage of companies that should not take venture, you know, in a sense, the responsibility to screen those out is on the VC, right?

这是他们的分内之事。 That's their job.

所以风投应该直接讲：这个不适合走风投，或者这个应该……而我确实发现，有时候我不得不非常直白地去问创业者这个问题。 So VC should be like, this isn't venture or this should, you know, and I do find that sometimes I do have to ask entrepreneurs that in a very pointed way.

就是问：这到底适不适合？我觉得你说得对。 It's like, is this suitable, you know, and I think you're right.

如果你看基金的生命周期问题，基金确实必须在其生命周期之内，为 LP（有限合伙人）提供退出的流动性。 If you look at the fund lifecycle issue, the funds do have to provide exit liquidity to the LPs, you know, within the fund lifecycle.

这个周期通常在十年左右，但如果你处在一只基金的后期，节奏可能会更快。 Oftentimes it's something like 10 years, you know, but if you're in the late stage of a fund, it can be faster, right?

比如说，如果一只十年期的基金已经走到第八年，那你就得快进快出，而我认为这一点每个人都应当本着道德如实披露、也应当理解清楚。 You know, so if you're eight years into a 10-year fund lifecycle, you know, then you got to get in and out, you know, and that's something that I think everyone needs to ethically disclose and, you know, understand.

同时我认为，创业者应当保持警觉，弄清楚自己拿的这笔钱到底是什么样的结构。 And I think entrepreneurs should be watchful and understand, you know, exactly what the structure of the money is that they're coming from.

对，我觉得这一点和你之前跟我聊过的另一个观点特别契合。你说过，风险投资很像新闻业，尤其是调查性新闻，你得去查证、去揭开一家公司背后到底是什么。 Yeah, and then I think this connects really well with another point that you've talked to me before and you said venture capitalism is a lot like journalism, investigative journalism, where you have to find out and sort of uncover what a company is behind it.

这其实是一个非常独特的视角，我以前从没听人这么说过。 This is actually a very unique perspective, never heard anyone, you know, talk about it before.

嗯，我想最有名的例子应该算是 Mike Moritz 吧？ Well, I mean, I'd say the most famous, the most famous one is like Mike Moritz, right?

在转到风投这一行之前，他在新闻业有过一段相当辉煌的职业生涯。 And he kind of had an illustrious career in journalism before he jumped over to the VC side.

其实符合这种模式的例子还有不少。我觉得归根结底，靠的既是好奇心，也是那种刨根问底的调查特质。 But there's actually a number of examples of this type of pattern, you know, and I think that ultimately what it comes down to is like both curiosity, but also this kind of investigation quality, right?

另一个可以类比的，就是私家侦探。 Which is, and the other parallel is like a private detective, right?

也就是那种真正四处奔走、跑断腿磨破鞋，不断去搜寻项目源头的人。 So someone who's really pounding the pavement, you know, wearing out their shoes and really like, you know, ferreting out where the deal flow is.

所以我觉得，这些就是所需要的特质。 So, you know, I think that's, those are the qualities.

对。 Yeah.

我觉得我们差不多快到尾声了，不过我还想再问两个问题。 And I think we're almost getting to the end here, but I just want to ask two more things.

你经历过互联网浪潮，经历过开源、加密货币，如今又赶上了 AI。而且在最早搭建机器学习那些东西的时候，你可以说就站在 AI 的起点上。 You've lived through the internet wave, you've lived through open source, crypto, and now AI, and you were sort of at the start of AI when you sort of built like machine learning stuff at the start.

那你觉得，这一次和过去相比是真的不同，还是其实并没有什么本质区别？ But what would you say that this time is truly different than the past or is it not really actually different?

我百分之百相信，这一次是真的不同。 I 100% believe it's truly different.

我觉得不能把 AI 仅仅当作又一波技术浪潮来看待。 And I think that AI cannot be reasoned about as another tech wave.

因为我们从未有过一种能够形成自催化循环的技术，它实际上在各个层面上都在自我改进。 And because we've never had the technology that's capable of an auto catalytic loop where effectively the technology is self improving on all levels.

AI 在设计 AI 赖以运行的 AI 芯片，AI 在设计它自己赖以运行的 AI 软件，就像 Claude 在设计它自己的 harness 一样。 So AI is designing AI chips that AI runs on top of, like AI is designing AI software that it runs on top of, like Claude is designing its own harness, right?

比如 Claude Fable、Claude Mythos，基本上都是由 Claude 自己写出来的。 Like Claude Fable, Claude Mythos were all written by largely Claude, right?

所以我们正处在这样一种局面，这是非常前所未有的。 So like we're in this, this, that's, that's very unprecedented.

所以我会说，很难把它当作一般的东西来理解。从伦理这类角度来看，也很难把它仅仅当成又一项普通技术。 So I would say this is hard to reason about as a, you know, and I would say from the perspective of things like ethics, it's hard to reason about this as just another tech, right?

因为它有能力自主做出决策。 Because it, it has the ability to make autonomous decisions.

所以如果你发布的东西具备可以自主采取行动的 agentic 能力，那么发布者就必须为其后果承担责任。 And so if you release something with agentic power to take action, then, you know, the, the releaser has to take responsibility for the outcomes, right?

所以 agentic AI 的责任归属问题很重要，必须厘清，因为它不像那种没有能动性的技术。 So the distribution of liability on agentic AI is important to resolve because, you know, it's, it's not like, it's not like inert tech.

对。 Yeah.

我想这是最后一个问题了。这档播客的一个重要目标，就是去激励那些想成为创业者、即将踏上创业之路，或者只是对创业感兴趣的人。 And I think the final question, because this, one of the major goals of this podcast is to inspire people who want to be founders or they're about to be founders or people who are just interested in startup as well.

对于那些想创业、却又太过害怕，或者觉得这件事对自己来说太难的人，你会对他们说些什么？ What would you, what would you say to someone who's trying to start a company, but they might be too afraid or thinking this is too difficult for them?

我会这么做：我大概会狠狠打击他们一番。 So here's what I would do is I would probably beat them up.

我之所以这样回答，是因为人们对创业这件事抱有很大的幻想。 And, and, and the reason I'm asked, I reason why I'm answering in this way is because there's a big fantasy about founder founding.

老实说，不是每个人都适合去创办一家风投级别的公司。 And honestly, not everybody is meant to found a venture class company.

对吧。 Right.

所以任何声称自己想创业的人，我都会打击一番，而且我每天都这么做，一天还好几次。 So I, you know, I would beat up anyone who claims that they, and I, you know, and I do every day, multiple times a day.

因为从某种意义上说，如果我主动去给别人泼冷水，这就有点像那个关于禅寺山门的经典故事。 You know, so like, you know, cause I, in a way, like if I'm actively discouraging to someone, you know, in a sense, like there's this classic story about the entrance to like a Zen monastery.

对吧。 Right.

守门人按规矩要连着至少三次把来人赶走，叫他们走开，而他们必须一次次地再回来。 And the gatekeeper is supposed to say, get lost to them, you know, at least three times in a row and they have to keep coming back.

对吧。 Right.

等他们第四次回来，才被允许进门。 And the fourth time they come back, they're allowed in.

对吧。 Right.

但这只是一条规矩。 But that's just a rule.

就是一条规矩而已。 Like it's a rule.

他们本来就该被劝退。 Like they're supposed to be discouraged.

对吧。 Right.

因为如果他们没那个胆量，一个投资人说了句难听的话他们就打退堂鼓，那这个投资人其实是替所有人省了一大堆麻烦。 Because if they don't have the guts, if one VC says something not nice to them and they quit, like that VC saved everyone a lot of trouble.

因为往后他们要面对的拒绝，远不止这么一点。 Cause like that, that's, they're going to get a lot more rejection than that.

如果他们脸皮太薄、承受不了，那就说明他们不是干这行的料。 And, you know, and if they, if they're too thin skin to handle that, then that's not, they're not cut out for it.

所以我会说，不是每个人都适合干这个。但那些适合的人，通常对他们要解决的问题满腔热忱、燃着一股劲。 So I would say not everyone's cut out for this, uh, but people that are cut out for it are just on fire, usually about the problem.

他们往往铁了心非做不可，很多动力甚至带着点复仇的意味，因为他们自己曾亲身体会过这个问题带来的痛苦。 Like they're usually really hell bent and a lot of their motivations have to do with kind of things like revenge, where they felt this agony of this problem themselves.

他们就是要把这个问题从世界上彻底根除。 And they just are out to eradicate this problem from the world.

他们对此满腔热血。 You know, they're just hot about it.

他们对这件事投入了感情。 They're, they're emotional about it.

而从某种程度上说，并不是每个人都具备这种特质。 And, you know, in a way like that quality isn't, not everyone has that quality.

不是每个人都会燃烧着一股想要解决问题的渴望。 Not everyone is just burning with the desire to solve.

而且是某个非常具体的问题。 Very specific problem.

对吧。 Right.

而有些人则是那种：哦，我想创个业，我想要那种很酷的生活方式。 Not, not some people are like, oh, I want to create a startup or I want this cool lifestyle.

那我会说，你或许可以去当个专门聊创业的 Instagram 网红，但别真的去创业。 And it's like, you could maybe be an Instagram influencer about startups, but like, don't try to make one.

因为那算不上一个真正要解决的问题：我想解决的问题是我想让自己看起来像个创业者，这种东西根本站不住脚。 Like that's not that problem of, I want, like the problem I want to solve is that I want to look like a entrepreneur like that, that, you know, that's not a thing.

对吧。 Right.

创业只是让你抵达某个目标的手段，而不是目的本身，你终究得去解决那个问题。 It's only, it's only a means for you to get to somewhere, but it's, it's not, you got to solve that problem.

对。 Yeah.

好的。 All right.

这次聊得非常愉快。 It's been really great.

非常感谢你。 Thank you so much.