Think Independently，并由此发展出高度泛化的能力需要大量竞争性的想法，巴菲特对他自己日常工作的总结就两件事：收集事实+思考事实【Appendix.8】。

Think Independently在两个工作中都起到重要的作用：

Think Independently是对“美感”或“优雅”（Elegance）的隐性偏好，本质是更有效率的解决问题。

巴菲特在2007年的股东大会有一个非常清晰的解释，他的逻辑很简单：（1）正确的思维框架；（2）大量竞争性的想法；（3）亲身体验增加想法的置信度，置信度=敢用程度【Appendix.4】。

可以类比AI 的 memorization + generalization：阅读是在扩大假设空间、提供候选解释模型；实践是在现实里拿到反馈信号（对/错、强/弱、快/慢），不断淘汰无效模型，最后把世界压缩成更短、更有效的知识结构。

所谓能力圈，就是反馈链条清晰、噪音较低、误差可控的子空间——在那里模型更容易产生“更短、更有效”的结构，更利于长期深耕，最终收获时间上的复利。

“aha” moment是一种高度泛化的现象，人工智能领域的科学家田渊栋专门为顿悟（grokking）做了数学上的解释【Appendix.12】。为了讲清楚“aha” moment，巴菲特在 2022 年股东大会用 ambiguous illusions（错觉）和 apperceptive mass（统觉团）解释过这种“突然看见”的现象【Appendix.10】。

“aha” moment不是凭空发生的，一个人的脑子里装满了各种观点：对A的看法、对B的看法，对C、D、E的看法，所有观点的背后有没有统一的、简洁优雅的知识结构？

独立思考（Think Independently）决定了注意力的投向，而正确思考（Think Correctly，即结构化的思维框架）决定“同样的注意力”能否更高效地产生压缩与泛化。Think Correctly对于Basic Trust的作用很容易理解，但对于Basic Mistrust的作用需要更深一步的思考。

Basic Mistrust的第一反应是“自我保护”和“处处防御”，它带来的麻烦不在情绪本身，其内心有一个高耗能后台程序——“威胁探测与自我防御系统”：必须时刻避免暴露弱点、必须自保、必须掩饰失误，分配给“实际问题”的算力被严重挤压，无法向更高维度的“泛化”演进，只能停留在应对具体情境“死记硬背（Memorization）”的层面，参考：【Appendix.5】。

Basic Mistrust会把人推向两种不同的防御模式：一种是低能量的“缩”；另一种是高能量的“抢”。两种自动化的防御机制很难根治，我小时候养过猪，猪会不会从猪圈里跑出去？关键变量有两个，一是猪的大小，Basic Mistrust如果是一种意图，是不是很难识别？我自己的理解，天赋（天生敏感）+专业训练，缺一不可；二是围栏的高度，Basic Mistrust是绝大多数，商业模式和企业文化能不能提供护栏很重要，但是，能不能形成有效的护栏要量力而行。

Think Correctly同样很难从根本上改变高能量的Basic Mistrust，相比于低能量的Basic Mistrust，高能量的模式更加难以改变，这里可能有必要区分心理层面的影响和实际行为上的影响。

高能量的Basic Mistrust有可能出现的情况是心理层面没任何的影响，甚至是负面的影响，但是行为上出现了虚假的改善，人工智能领域有一种说法：欺骗性对齐（deceptive alignment）【Appendix.13】。

Think Correctly只是被用来合理化，而不是真正发挥了效用，这种可能性远高于低能量的Basic Mistrust。

高能量的Basic Mistrust也会有真正的改善，并且行为上的改善更快、更显著，因为高能量的Basic Mistrust更加依赖心理上的支点，如果Think Correctly有一些足够简洁、足够有效、标签式的知识结构，这些标签又能同时绑定高能量Basic Mistrust所需要的心理上的高位，这样的支点能在行为上产生非常好的改善。

WARREN BUFFETT: Both the questions relate to futures of one sort, calls, or whatever they may be, and — I think that investors should stick to buying ownership in businesses. It’s not that you can’t come up with a theoretical argument for buying, say, a — I mean, if you think Coca-Cola’s attractive, you can say, well, I’d rather buy a five-year option on Coke than buy the stock directly because it introduces leverage without the risk of going broke. But I think that that’s a dangerous path to start down, because it — If it works well, it’s so — it’s dynamite to start playing with things that can expire and become worthless, or can be bought with very low margin, as the OEX options you were talking about. Borrowed money usually — or frequently — leads to trouble. And it’s not necessary.

WARREN BUFFETT：你的两个问题都涉及某种形式的期货、看涨期权（calls）或类似工具——而我认为，投资者应该坚持购买企业所有权（直接买股票）。并不是说你无法提出某种“理论上合理”的论证来支持买期权。比如，如果你觉得 Coca-Cola 很有吸引力，你可能会说：我宁愿买一张 5 年期的 Coke 期权，而不是直接买股票，因为这相当于引入杠杆，但又不像借钱那样有“破产”的风险。但我认为，一旦走上这条路，是危险的。因为——如果它奏效，那种诱惑太强了——你开始玩那些会到期、可能归零的东西，或者像你说的 OEX 期权那样，用很低的保证金就能买到的东西，这简直像在玩炸药。借来的钱通常——或者说经常——会带来麻烦，而且这并非必要。

I mean, if you had some compelling reason — if you’re going — if you had to double your money by the end of the year or be shot, you know, then, I would head for the futures market because, you know, you need to do it. I mean, you have to introduce borrowed money. But you really ought to figure out how you can be happy with the present amount of money you’ve got and, then, figure that everything else is, you know, all to the good as you go along, and — I don’t think people — once they start focusing on short-term price behavior, which is the nature of buying calls, or LEAPS, or speculating in index futures, once you start concentrating on that, I think you’re very likely to take your eye off the main ball, which is just valuing businesses. I don’t recommend it. Charlie?

我的意思是，如果你有某种“非做不可”的理由——比如你必须在年底前把钱翻倍，否则就要被枪毙（笑）——那我会建议你去期货市场，因为你确实“需要”那么做。也就是说，你必须引入借来的钱（杠杆）。但正常情况下，你真正应该做的是：先想清楚如何对自己现在拥有的钱“感到满足”，然后把之后的增量收益都视为“锦上添花”。我不认为人们一旦开始盯着短期价格波动——这正是买 calls、买 LEAPS、或投机指数期货的本质——当你开始集中精力在这上面时，你很可能就会把眼睛从“主球”上移开；而主球其实很简单：给企业估值。我不推荐这么做。Charlie？

In physics, theories are judged by their predictive power. That is why the theory of relativity is revered even today. It took an existing theory (Newton’s laws) and enhanced the predictive accuracy of an already successful equation. It is also the reason why string theory gets such a bad rap: it has no predictive power.  

在物理学中，理论要以其预测能力来衡量。这就是为什么相对论至今仍受人推崇。它在既有的牛顿定律基础上，提高了原本已相当成功的方程式的预测精度。而弦理论之所以饱受诟病，也正因为它缺乏预测能力。

Whenever, I hear of an idea in business or finance, I wonder about its predictive power. And I’m often surprised by how so many theories have zero to very little predictive power.  

每当我听到商业或金融领域的新观点时，我都会想，它到底有多少预测能力。而让我吃惊的是，许多所谓理论几乎没有或完全没有预测能力。

Let’s take an example, I will henceforth call it ‘the platform theory’. It’s an idea pervasive in the current business zeitgeist: i.e. that platforms represent the future of distribution’. McKinsey, BCG and Bain have done all done some fancy reports on this idea over the past few years. Now, there’s nothing wrong with the idea per se. I actually am fully bought into it: distribution will likely shift to platforms or ‘super-apps’ in the coming decade. But the theory doesn’t pass the ‘predictive smell-test’. If you’re an executive trying to use this statement to understand which companies will grow or how you might become a platform yourself, the idea hardly adds anything. Current literature on the topic is generic stuff around customer centricity and digitization as the way to respond. But these ideas predate the ‘platform era’. Before I offer a better version of this theory, let’s go down the annals of tech history. We won’t visit Silicon Valley though - our story starts in the neighboring city of Seattle.  

举个例子吧，我姑且称之为“平台理论”。当下商业界普遍流行这样一种观点：平台将代表未来的分销模式。麦肯锡、波士顿咨询和贝恩在过去几年里都为此写过华丽的报告。本身这个观点并无不妥，我自己也认同——未来十年分销渠道大概率会转向平台或“超级应用”。然而，这个理论却无法通过“预测嗅觉测试”。如果你是高管，想凭这句话判断哪些公司会成长，或者如何把自己变成平台，它几乎帮不上忙。现有文献谈的都是以客户为中心、数字化转型之类陈词滥调，而这些概念早在“平台时代”之前就提出了。在我给出更好的版本前，让我们回顾一下科技史。不过我们不去硅谷，而是从邻近的西雅图说起。

It is 2002. This is Amazon. But this isn’t the giant we know today. The stock trades for $15, a far-cry from the $1760 level it trades at today. There is no Prime, no 2-day shipping, no AWS, no Echo - so nobody really talks about the company then. Jeff Bezos, a maniacal micro-manager, sends an email to all his staff. The email is likely the single most important email in the history of the company. Heck, it might even be the most important email in the history of retail. Here’s how it reads:  

时间来到2002年。这是亚马逊，但还不是今天那个庞然大物。当时股价只有15美元，远不及如今的1760美元。那时没有Prime，没有两日达，没有AWS，也没有Echo——几乎没人谈论这家公司。狂热且事无巨细的杰夫·贝索斯给全体员工发了一封邮件。这封邮件可能是公司史上最重要的一封，甚至可能是零售史上最重要的一封。邮件内容如下：

Now if you’re somebody who has read a bunch on Amazon history, you likely already know about this email. But people often miss the key point of it. The email isn’t what created the ‘two-pizza teams’ or ushered in the Agile/DevOps era. Or it wasn’t something that made AWS an inevitability. Amazon already had product-teams even before Bezos pressed the send button then. And Bezos had been banging on about innovation and failure for a decade at that point already.  

如果你读过不少关于亚马逊的资料，这封邮件你也许早已耳熟能详。但人们常常忽略其核心所在。这封邮件并不是“两个披萨团队”的起点，也并未直接开启敏捷/DevOps时代，更没有让AWS成为命中注定。早在贝索斯按下发送键之前，亚马逊就已经存在产品团队；而在那之前整整十年，贝索斯也一直在强调创新与失败。

No, the email isn’t so much about what it said than what it did. It did two separate but related things. And neither of these were about the company’s IT or organization structure. Here is what these things were:  

不，这封邮件的重要性不在于它说了什么，而在于它产生了什么效果。它带来了两件既相互关联又彼此独立的改变，而且两者都与公司的 IT 或组织架构无关。具体来说，它做到的事情是：

Let’s start with the first bullet - Eating your own dogfood  

让我们先谈第一点——自家吃狗粮。

The idea predates Bezos’ email by at least three decades. There was even a popular TV ad in the 1970s. A dogfood company CEO feeds his own dogs the company’s product to show how good the food actually is. However, we don’t have to take things so literally. The insight here is that individual teams will optimize and continuously enhance their product or service only if they and their colleagues are constantly being exposed to the outcome. If that’s a mouthful, let me explain this with an analogy. Imagine you are a fruit shop owner and sell fruits for a living. Now let’s say, I tell you one day that you and your family will henceforth be eating fruits only for breakfast everyday. And the fruits will be randomly chosen from your shop. To up the stakes, all your friends and family will actually always be buying fruits from your shop. And much like you, they’ll get a random assortment from your store. So you have lots of ‘skin-in-the-game’ now. In this scenario, I am certain that most folks will start by ensuring that the quality of fruits they source are the absolute best. What else might you do? Well, over time you will try and enhance the ‘fruit-eating’ experience. You might squeeze a few fruits and offer customers the ability to buy bottled fresh juice. You might try and break the rules slightly and create some fancy pina-coladas and mix a few fruits. Basically, you constantly iterate and improve the product. And that is what the ‘eat your own dogfood’ mindset really brings.  

这一理念在贝索斯的邮件发出前至少已存在三十年。早在 1970 年代就有一则著名的电视广告：一家狗粮公司的 CEO 把公司的产品喂给自家的狗，以证明狗粮的品质。然而，我们不必把它理解得如此字面。真正的洞见在于，只有当团队自己以及同事持续直接接触到成果时，他们才会去优化并不断提升自己的产品或服务。如果这句话听起来拗口，那我用一个类比来说明。想象你是一位水果店老板，靠卖水果为生。现在我告诉你，从今天起你和家人每天早餐只能吃水果，而且水果将随机来自你店里的货架。为了加码，你所有的亲朋也都会固定在你店里买水果，并且同样随机拿取。这样一来，你就拥有大量的“利益攸关”。在这种情形下，我敢肯定，大多数人首先会确保进货水果的品质绝对上乘。你还能做些什么？随着时间推移，你会尝试提升“吃水果”的体验。你可能会把水果榨汁，给顾客提供瓶装鲜榨汁；或者稍微打破规则，调制些别致的菠萝可乐达鸡尾酒，混合几种水果。总之，你会不断迭代并改进产品。而这正是“自家吃狗粮”心态所带来的结果。

So when Bezos tells folks ‘I don’t care what technology you use, I want the service to be externalizable’, every team at Amazon suddenly become an end customer and an end supplier to others. And remember how we upped the quality of our fruit-product when we were the ones eating and selling it to our colleagues daily? Something similar happened at Amazon gradually. The quality of the Amazon’s IT architecture, its internal services and its products, all improved. And that’s really what made all the difference.  

Now, let’s tackle the second bullet - Offering an internal service at a price which can be benchmarked externally.  

现在，让我们讨论第二点——以可与外部基准比较的价格提供内部服务。

We will use another example to highlight what we mean here. Imagine, you and your friend want to start an insurance company. You likely start off as an MGA or broker. Over time, as you get a few customers you will likely backwards integrate across other parts of the value chain. Very soon you’ll have a bunch of departments - Finance, legal, pricing, compliance, customer service, claims etc. That is how most organizations are set up. In business parlance, you are a ‘vertically integrated’ company. But vertical integration and scale bring with it a related evil. Your departments now become monopolists. Companies set their prices by looking at market rates but also their internal cost structures. So while not explicitly doing so, every department becomes a mini-monopoly within the company. Legal tells the financial controller that it costs them $10M to run their department. And that’s how he accounts for it, treating as a ‘cost-of-doing-business’. Occasionally, there might be some negotiation around budget but that’s office politics. The legal department doesn’t get asked, ‘Hey btw competitor X’s legal department does the same job as you for half the price’. For one, such data is rarely available. But even when it’s available, it is usually masked under quips like ‘It’s not like-for-like etc’.  

我们再用一个例子来说明这一点。假设你和朋友想创立一家保险公司，一开始大概率会以 MGA 或经纪人的身份起步。随着客户增多，你们可能会逐步向价值链其他环节反向整合。不久之后，你们就会拥有一堆部门——财务、法务、定价、合规、客服、理赔等。这就是大多数组织的常见结构。商业语言里，你们是一家“纵向整合”的公司。但纵向整合和规模也带来相伴的弊端：各部门成了垄断者。企业在定价时既看市场价，也看内部成本结构。因此，即便并非明文规定，每个部门都在公司内部变成了一个小型垄断体。法务部门告诉财务主管，他们部门运营成本是 1000 万美元；财务主管就按这个数记账，把它当成“经营成本”。偶尔会就预算有所讨价还价，但那属于办公室政治。几乎没有人会问法务部：“顺便说一句，竞争对手 X 的法务部做同样的工作只要你们一半的花费。”一方面，这类数据很少有；就算有，通常也被一句“这俩不具可比性”之类的话糊弄过去。

And that is where the problem comes. Fat around costs comes in because internal departments in vertically integrated companies become price-setters. The brilliance of Bezos’ email really shone in this area. I am not sure if he could have fully realized it at that point. But it made Amazon into a machine that makes machines. By making every single service into something externalizable, Amazon could monitor these costs closely. And if the ̶d̶o̶g̶f̶o̶o̶d̶ service was good enough, they could offer it to the market over time. So if Amazon was the insurance company in our above example, it’s legal department would offer its legal services to any company that didn’t want to deal with the ordeal of insurance-related legal affairs. And over time, it would offer such services for every single part of the insurance value chain - distribution, call center, claims etc. This externalizable architecture is the genius that brought us AWS and Amazon marketplace - the two juggernauts on which Amazon’s dominance has been built. Over time, as Amazon signs up more customers for such services, it offers these services at even a lower cost (due to scale benefits). This further tightens its moat by enhancing the proposition for end customers. And the Amazon flywheel gets set in motion.  

这正是问题所在。在垂直整合公司的内部部门成为定价者后，成本周围就会滋生“肥肉”。贝索斯那封邮件的高明之处就在这一点上。我不确定他当时是否已经完全意识到这一点，但它把亚马逊变成了一台“制造机器的机器”。通过把每一项服务都做成可外部化的，亚马逊能够紧密监控这些成本。如果这项（狗粮）服务足够好，他们就可以逐步将其推向市场。因此，如果亚马逊在我们前面的例子中是一家保险公司，它的法务部门就会向任何不想处理保险法律事务麻烦的公司提供法律服务。随着时间推移，它还会为保险价值链上的每一个环节——渠道、呼叫中心、理赔等——提供此类服务。这种可外部化的架构孕育了 AWS 和 Amazon Marketplace 这两大支柱，正是它们奠定了亚马逊的霸主地位。随着亚马逊为此类服务签下越来越多的客户，由于规模效应，它们还能以更低的价格提供服务。这样一来，通过提升终端客户的价值主张，它的护城河被进一步加固，“亚马逊飞轮”便随之启动。

I don’t want to make this post to be just about Amazon so I will conclude some final thoughts around platforms before tying things up. There are many successful platform companies today (Facebook, Shopify, Apple, WeChat, Go-Jek etc.). But other than have great addictive products, they share a key feature. They enable other individuals and businesses to launch things faster and cheaper than before. Consider Facebook’s decision to open up its third-party developer api early on, way before focusing on mobile monetization. Zuckerberg realized the potential value of letting others build on top of Facebook. But it was a great deal for third-party developers too. They now had reach and distribution at a scale never seen before. Very soon after the API launch, hundreds of thousands of developers were building their own games and tools on top of Facebook. As a result, users began spending more and time on Facebook. And while a user might be spending 8 hours on Facebook, 5-6 of those hours were spent playing a game like Farmville. So Facebook succeeded where previous social network like Orkut and MySpace had failed - and it did so by becoming a platform.  

我并不想让这篇文章只谈亚马逊，因此在总结之前，先对平台再做一些收尾思考。当今有许多成功的平台公司（Facebook、Shopify、Apple、微信、Go-Jek 等），除了拥有令人上瘾的优秀产品之外，它们还有一个共同特征：让个人和企业以前所未有的速度和更低的成本推出新东西。想想 Facebook 很早就开放第三方开发者 API 的决定，当时它还未专注于移动端变现。扎克伯格意识到让别人基于 Facebook 构建产品的潜在价值；而对第三方开发者来说，这也是笔划算的买卖，他们一下子拥有了前所未有的用户触达和分发规模。API 发布后不久，数十万开发者就在 Facebook 上构建自己的游戏和工具。结果，用户在 Facebook 上花的时间越来越多；如果一个用户在 Facebook 上待 8 小时，其中有 5-6 小时可能都在玩《农场庄园》这类游戏。通过成为一个平台，Facebook 在 Orkut 和 MySpace 失败的地方取得了成功。

So, if the thesis is ‘Platforms will dominate distribution’ - What can most incumbents do? I believe there are a couple of options. One is to try and copy Amazon and ‘out-Amazon’ the Seattle giant within one’s own industry. This likely requires tonnes of capex, a re-investment of profits (to pay lower taxes ala Amazon), de-verticalization and a complete culture change. And while some executives might convince public markets, convincing employees may prove a lot more difficult. This is because de-verticalization requires setting up the organization such that every service is externally and internally consumable. This is quite antithetical to the concept of the modern corporation. So, I am skeptical that this will work as most workplaces have decades of cultural baggage around vertical integration.  

那么，如果“平台将主宰分发”这一论断成立，大多数既有企业能做什么？我认为有几个选项。一个是试图模仿亚马逊，在自己所在行业里“比亚马逊更亚马逊”。这可能需要巨额资本支出、将利润再投资（像亚马逊那样少缴税）、去垂直化以及彻底的文化变革。即便某些高管能说服公开市场，想说服员工却要困难得多，因为去垂直化意味着必须把组织架设成每一项服务都可供外部和内部消费的形态，而这与现代企业的概念相悖。因此，我对这条路径能否奏效持怀疑态度，毕竟大多数职场在垂直整合方面背负了数十年的文化包袱。

The second idea is simpler. It is likely the correct approach for most companies. Focus only on your core strength (or IP) and outsource everything else to the low-cost player that is focused on just one thing. Consider this: Uber, the company that has most captured the public imagination over the past decade, is actually a company largely built on top of four APIs: Stripe, AWS, Checkr, Twilio. Most people know Stripe and AWS. Very few have heard of Checkr and Twilio. Checkr conducts background checks on drivers via an API. Twilio allows companies to send and receive text messages and calls via a simple API. All four of these companies are multi-billion dollar companies in their own right: Stripe ($35B), AWS ($500B+), Twilio ($16B+) and Checkr ($2.1B+). Uber didn’t try and re-invent the wheel on any of these services. And despite building on top of these services, it has a market cap north of $50B. And many predict it to be the next platform company. (If recent news is something to go by, it definitely seems so)  

第二种思路更简单，且很可能是多数公司的正确做法。只专注于自己的核心优势（或知识产权），把其他一切都外包给专注于单一领域且成本最低的供应商。举个例子：过去十年最受公众关注的公司 Uber，其实主要构建在四个 API 之上：Stripe、AWS、Checkr、Twilio。大多数人知道 Stripe 和 AWS，很少有人听说过 Checkr 和 Twilio。Checkr 通过 API 为司机做背景调查；Twilio 通过简单的 API 让公司发送和接收短信及电话。这四家公司本身都是市值数十亿美元的巨头：Stripe（350 亿美元）、AWS（5,000 亿美元以上）、Twilio（160 亿美元以上）、Checkr（21 亿美元以上）。Uber 并没有在这些服务上重复造轮子；借助它们，Uber 的市值仍超过 500 亿美元。许多人预测它将成为下一家平台公司。（从最近的新闻看，也确实如此）

So, to conclude - here’s the better version of the ‘platform theory’.  

因此，总结一下——这就是“平台理论”的更佳版本。

Platforms will dominate distribution for most products and services in the Web 3.0 era. Incumbents may respond to this threat/opportunity by:  

在 Web 3.0 时代，平台将主宰大多数产品和服务的分发。现有企业可以通过以下方式来应对这一威胁/机遇：

1. Making their services externally surface-able via commercially available APIs. Ultimately, they may try to corner the market around a particular niche and be the ecosystem of choice here (This is both hard, costly and time-taking)   

通过商业化 API 将自己的服务向外部开放，最终可能努力在某一细分领域占据市场并成为首选生态系统（这既困难、昂贵又耗时）

2. Divesting most non-core functions and outsourcing these to the lowest-cost players available (This is relatively cheap and in-expensive but goes against individual bias for growth and larger revenues)   

剥离大部分非核心职能，并将其外包给成本最低的供应商（这相对便宜且成本低，但与个人追求增长和更高收入的偏好相冲突）

For startups, this mostly represents a huge opportunity. There are tonnes of verticals that have still not be attacked. Internet and smartphones have enabled distribution at a scale not seen before. Given that total addressable markets are 10x previous levels, the niches one need target to build something interesting is actually much smaller. Checkr has built a unicorn around something seemingly trivial like background checks. And yet, many such opportunities will arise to build businesses around ideas that might seen unappealing to the blind eye.  

对于初创公司而言，这主要意味着巨大的机会。仍有大量垂直领域尚未被触及。互联网和智能手机带来了前所未有的分发规模；鉴于可服务市场总量是过去的 10 倍，要建立有趣业务所需瞄准的细分市场其实更小。Checkr 就在看似琐碎的背景调查领域打造了独角兽。未来仍将出现许多类似机会，围绕乍看并不起眼的想法建立业务。