The seed

Buffett took over Berkshire Hathaway in 1965 with a book value of $22 million. Half a century later he looked back and described the entire mechanism in one plain phrase:

“By retaining all earnings for a very long time, and allowing compound interest to work its magic, we have amassed funds…”

Retain all earnings, give it long enough, let compound interest do its magic. That’s Buffett’s own summary of his life’s work. Not IQ, not luck, not insider information. Time × rate × no interruption.

This isn’t a secret about investing. It’s a secret about a single equation: Value(t) = Principal × (1 + rate)^time. The math is two centuries old. What I want to write about is what Naval did with it:

“All returns in life, whether in wealth, relationships, or knowledge, come from compound interest.”

I want to unfold that one sentence across four domains that don’t look related: the stock market, learning, mental models, and raising a child. By the end I think you’ll see they’re four strokes of the same brush.

I. The stock market: the oldest compound

Buffett has another counterintuitive line that gets quoted a lot and rarely understood:

“Time is the friend of the wonderful business, the enemy of the mediocre.”

Most people read this as “time + investing = more money.” That isn’t what he’s saying. Time only amplifies quality that’s already there. If the business is mediocre (low or negative r), time only makes it more mediocre. Time is a friend only when the principal is good and r is sustainable.

So the first secret of stock-market compounding isn’t the rate. It’s the quality of principal. Buffett’s word for this is “circle of competence”:

“You only have to be able to evaluate companies within your circle of competence. The size of that circle is not very important; knowing its boundaries, however, is vital.”

The size of the circle doesn’t matter. Knowing where it ends does. High returns outside your circle are an illusion. They’re things you don’t understand, pretending to grow.

This is why WeWork went from a $47 billion valuation to bankruptcy. Why Terra/Luna evaporated $45 billion in a week. They weren’t failing to compound. Their underlying r was negative. FOMO and liquidity pushed prices up for a while; eventually the math caught up. Stepping outside your circle is adding leverage on top of a negative rate.

The hardest part of investing isn’t picking winners. It’s having the humility to say: “this is outside what I actually understand, so the apparent return is noise, not signal.” Buffett’s teacher Graham had a one-liner for it: “Price is what you pay; value is what you get.”

The market’s compound lesson isn’t a magic trick. It’s a discipline. Make sure r doesn’t go to zero before you start worrying about t.

II. Learning: knowledge compounds, too

Naval put learning right next to wealth in that one sentence. His follow-up essay on specific knowledge explains why:

“In leveraged domains, someone 90% correct vs 80% correct will literally get paid hundreds of times more by the market because of the leverage and because of the compounding factors.”

Why such a violent gap from a 10-point difference? Because learning’s compound isn’t linear. The deeper you understand something, the cheaper the next layer becomes. “I read about it once” and “I’ve watched it run on my own machine” are different categories. Every working example you’ve actually built lowers the activation energy for the next one. That’s the compounding mechanism.

But learning has a famous anti-pattern worth pulling apart: the 10,000-hour rule.

It sounds compound. Practice 10,000 hours and you become an expert. Anders Ericsson — the researcher whose original paper got summarized into that rule — pushed back on it directly:

“Unfortunately, this rule — which is the only thing that many people today know about the effects of practice — is wrong in several ways.”

His point: 10,000 hours was the average of his top violinists by age 20, not a threshold for any field. And it was hours of deliberate practice with feedback, goals, correction, not “time spent.” Hours don’t compound. Deliberate hours do.

This distinction is the whole game. Different skills have wildly different half-lives. Framework APIs decay in 1–2 years. Language syntax in 3–5. System design lasts 10–15. Deep domain expertise 15–20. Meta-skills lifetime. Learn for the same number of years, and the person stacking framework knowledge is compounding an asset that resets every two years. The person stacking domain expertise is compounding something that holds for two decades.

So the rate in your learning compound isn’t “how hard you tried.” It’s what you chose to compound on.

III. Mental models: the compound on top of the compound

Both previous domains required something to filter what’s worth compounding on. That something has a name. Charlie Munger:

“You’ve got to have models in your head. And you’ve got to array your experience — both vicarious and direct — on this latticework of models.”

And the follow-up, which matters more:

“Fortunately, it isn’t that tough — because 80 or 90 important models will carry about 90% of the freight in making you a worldly-wise person.”

80 to 90 models cover 90% of the work. Not “more is better.” The right 80 to 90.

A single mental model nudges your r in one domain. Two or three intersecting models can produce a sudden non-linear jump. Munger called this the Lollapalooza Effect: multiple forces aligning the same direction at the same time produce extreme outcomes.

The opposite of a mental model isn’t a worse model. It’s using a model once and forgetting it. The equivalent of selling a great company too early — you converted a compounding asset into a one-shot cash payment.

So in the compound equation for mental models, “time” doesn’t mean “how long I live.” It means how many years a model has been running in my head. Munger at 80 used the same lattice better than at 60. Same 80 models, 60 more years of operation.

IV. Raising a child: the longest compound of all

I’m both the experimenter and a variable in this one. Because I’m, myself, a product of early compounding.

I learned to read earlier than my classmates. Looks small. The downstream effects aren’t:

  • Earlier access to children’s books → read more → vocabulary expanded → could read harder books
  • Read more than peers → understood class faster → teachers gave positive feedback → liked learning
  • Liked learning → more independent exploration → learned more → confidence itself was compounding

This isn’t something I noticed at the time. There’s a famous psychology finding for it: the Matthew effect in reading. Kids who get the basics down before grade three keep accelerating in vocabulary, knowledge, and comprehension for life. Kids who don’t fall behind at an accelerating rate. The gap doesn’t stay constant. It widens.

The name comes from Matthew 25:29, which Gladwell put on the opening page of Chapter 1 of Outliers:

“FOR UNTO EVERYONE THAT HATH SHALL BE GIVEN, AND HE SHALL HAVE ABUNDANCE. BUT FROM HIM THAT HATH NOT SHALL BE TAKEN AWAY EVEN THAT WHICH HE HATH.”

There’s a more visceral version of the same effect. Canadian junior hockey selection cuts off on September 1. Pull the stats and you find elite professional players are massively over-represented in January, February, and March birthdays. Why?

  • Within a single grade, a January-born child is almost a year older than a December-born one.
  • At ages 8, 10, 12, a one-year size and coordination gap is enormous.
  • Older kids get picked for “good teams” → more training → gap widens → picked again at the next level → eventually professional.
  • Two children born the same year, separated by a few months on the calendar, end up on exponentially diverging tracks.

It’s uncomfortable because it shows that a child’s compounding starting point is largely set by parents and environment, not by the child.

But it’s also where the hope lives. If small early advantages get amplified into life-changing differences, the little bit parents can control matters disproportionately. The point isn’t “cram more knowledge into the kid.” That’s the worst reading of this. It’s also the conclusion Tsinghua’s Professor Liu Jia goes after directly: knowledge-based education is dead because all knowledge is in the language models now. The actual job is to grow the child — their interests, their AI-native instincts, their first-principles thinking.

What parents can actually compound is closer to three things:

  1. A positive-feedback starting point. Pick one domain, any domain, where the kid gets to feel “I’m a little better at this than the people around me.” Reading is the canonical one because early literacy radiates into every other subject.
  2. Confidence as a compoundable asset. Confidence isn’t built by praise. Carol Dweck wrote a great book about this — the punchline is: “Praising children’s intelligence harms motivation and it harms performance.” Praise intelligence and kids avoid challenge. Praise effort and they seek it. Confidence compounds from feedback on process, not feedback on outcomes.
  3. The time window itself. A child has 20+ years of unbroken compounding ahead. The longest single window any human gets. The famous version of this: 1% better every day for a year leaves you 37 times better. The math is elementary. The point is that “every day” lasts twenty years if you don’t break it.

One honest caveat. That 1.01^365 = 37.78 number is just exponential math, not anyone’s invention, and it doesn’t auto-execute on children. The “myelin → skill” story from Coyle is popularization. The actual neuroscience of myelination is fuzzier than the metaphor. The compound framework points a direction. It doesn’t promise the outcome will arrive on its own. The parent’s actual job isn’t to make 37.78 happen. It’s to not break the exponent. Don’t sever the kid’s intrinsic motivation. Don’t strip-mine their interests for your résumé. Don’t let the rate go negative.

Closing

Back to Buffett’s $22M-to-$349B. It is a story about patience and compounding. But that’s not why I wrote this.

The real secret in Value(t) = Principal × (1 + rate)^time isn’t that it makes money grow. It’s that life doesn’t have separate domains. Principal, rate, time, leverage. Those four dimensions are the same in stocks, in learning, in mental models, in raising a child. Discipline you build in one becomes leverage in the other three.

Stock-market discipline, learning selection, model assembly, parenting. They look like four unrelated things. They aren’t. The seam between them is where this post lives.

So the question I’d actually leave you with: which dimension did you cash out too early today? Did you pick something with a high r whose t will go to zero? Did you have good principal sitting idle? Did you have leverage you didn’t use? Were you compounding on a negative r without noticing?

The equation has four variables. Pick one, and move.

缘起

Buffett 1965 年接管 Berkshire Hathaway 的时候，公司账面净值是 2200 万美元。半个多世纪后回头看：股东权益长到了 3490 亿美元。他自己用一句很朴素的话描述了这件事的机制：

“By retaining all earnings for a very long time, and allowing compound interest to work its magic, we have amassed funds…”

留住所有利润，给它足够长的时间，让复利施展它的魔法。这是 Buffett 对自己一生事业的总结。不是智商，不是运气，不是独家信息。是时间 × 利率 × 不中断。

这不是投资的秘密。是同一条公式的秘密：Value(t) = Principal × (1 + rate)^time。这数学两百年前就有人写过了。真正值得写一篇文章的是 Naval 那句话：

“All returns in life, whether in wealth, relationships, or knowledge, come from compound interest.”

人生所有的回报——无论是财富、关系还是知识——都来自复利。

我想把这句话拆到四个看起来不相关的领域：股市、学习、心智模型、孩子。写完你会发现，是同一支笔画出来的四笔。

一、股市：最老的复利

Buffett 还有一句被引用很多、被理解很少的话：

“Time is the friend of the wonderful business, the enemy of the mediocre.”

时间是好生意的朋友，是平庸生意的敌人。

这话的含义跟大部分人想的不一样。大部分人以为时间长了 = 收益大。Buffett 说的是：时间只放大已经确定的质量。生意要是平庸的（r 很低或负），时间只会让它更平庸。本金好（r 可持续）的时候，时间才是朋友。

——讲到这儿其实就是金庸笔下”内功 vs 招式”那一套。乔峰会的招数不算特别多，降龙十八掌翻来覆去就那十八招。可他打太祖长拳都能打得江湖闻风丧胆。为什么？内功深。换成游坦之——冰蚕毒掌、易筋经、神功盖世，招式比乔峰还炫，结局呢？走火入魔、瞎了眼、自杀。招式是 r，内功才是本金。本金不对，招式炫得越多，崩得越快。

所以股市复利的第一个秘密不是利率，是本金质量。Buffett 自己用的词叫”能力圈”：

“You only have to be able to evaluate companies within your circle of competence. The size of that circle is not very important; knowing its boundaries, however, is vital.”

圈大小不重要，重要的是你知道边界在哪。 圈外的高利率是幻觉，是你不懂的东西在装作会涨。

这就是为什么 WeWork 会从 470 亿美元估值跌到申请破产。为什么 Terra/Luna 一周蒸发 450 亿美元。不是它们不复利。是它们本金的”利率”本来就是负的。FOMO 和流动性把价格暂时推上去而已。出圈，就是在负利率上加杠杆。

投资里最难的不是挑赢家。是承认”这件事我其实不懂，所以表面的回报是噪音，不是信号”。Buffett 的老师 Graham 有一句话能扎死人：“Price is what you pay; value is what you get.” 价格是你付的，价值是你得到的。

股市的复利不教什么魔法。它教一条纪律。先保证 r 不归零，再谈 t。

二、学习与成长：知识也在结息

Naval 把”学习”和”财富”并排放在复利那句话里。他另一篇关于 specific knowledge 的文章解释了原因：

“In leveraged domains, someone 90% correct vs 80% correct will literally get paid hundreds of times more by the market because of the leverage and because of the compounding factors.”

在有杠杆的领域，90% 正确和 80% 正确的回报，能差出几百倍。

为什么 10 个百分点的差距能撕出几百倍的回报？因为学习的复利不是线性的。理解越深，下一次理解的成本越低。”听说过”和”在自己机器上跑过”完全是两回事。每个亲手跑过的 working example 都降低下一个项目的启动能量。这就是知识的复利机制。

但学习有一个挺值得拆的反模式：10,000 小时规则。

听起来挺复利的：练 10,000 小时就是专家。但 Ericsson——也就是原论文作者——后来自己反驳了这条总结：

“Unfortunately, this rule — which is the only thing that many people today know about the effects of practice — is wrong in several ways.”

他的重点是：10,000 小时是他研究里前几名小提琴手到 20 岁的平均时间，不是任何领域的门槛。而且这是深度练习（有反馈、有目标、有纠正）的小时数，不是”做了多久”。小时数本身不复利。深度练习的小时数才复利。

这个区别就是命门。技能的半衰期差异巨大。框架 API 1–2 年，语言语法 3–5 年，系统设计 10–15 年，领域专长 15–20 年，元能力终身。学同样长的时间，选框架 API 的人在复利一个每两年归零的资产；选领域专长的人在复利一个二十年都不贬值的资产。

所以学习的复利，利率不是”你多认真”，是”你选了什么在复利”。

——还是金庸的话好懂。少林七十二绝艺一辈子学不完，可绝艺归绝艺，学十几门就够还。但《九阳真经》《易筋经》这种内功心法，越老越值钱。张三丰一百岁的时候武功还在涨。鸠摩智花二十年学招式，结果四十多岁就开始走火入魔。学什么，决定了你的复利能跑多长。

三、心智模型：复利之上的复利

前两个领域（股市、学习）都需要某个东西去筛选什么值得复利。这东西有个名字。Charlie Munger 这么说的：

“You’ve got to have models in your head. And you’ve got to array your experience—both vicarious and direct—on this latticework of models.”

他后面那句更关键：

“Fortunately, it isn’t that tough—because 80 or 90 important models will carry about 90% of the freight in making you a worldly-wise person.”

80–90 个模型能扛 90% 的重量。不是越多越好。是选对的那 80–90 个。

一个心智模型只能轻轻推一下某个领域的 r。几个心智模型交叉在一起，回报会非线性地跳。Munger 把这叫 Lollapalooza Effect：多个因素同向叠加产生极端结果。

心智模型的反面是什么？不是模型不够好。是用过一次就忘了。等于股市里把一家好公司过早卖了，把复利资产变成了一次性现金。

所以心智模型的复利方程里，“时间”不是”我活多久”，是”同一个模型在我脑子里跑了多少年”。Munger 80 岁比 60 岁用得更熟练——同一套 80–90 个模型，跑了 60 多年。

四、培养孩子：最长的一次复利

前三个领域我自己是实验主体。这一章我既是实验者又是变量。因为我自己就是一个”早期复利”的产物。

我小时候识字比同龄人早。看起来不起眼。后果其实是正反馈的：

  • 更早能读童书 → 读得多 → 词汇量扩展 → 能读更难的书
  • 读得比同学多 → 课堂上理解得快 → 老师反馈是正面的 → 更爱学习
  • 爱学习 → 更愿意独立探索 → 学到更多 → 自信心本身在复利

这不是我发现的模式。心理学里有个挺有名的发现：阅读上的马太效应。三四年级之前掌握了阅读基础的孩子，终生的词汇、知识、理解能力都在加速拉开差距。没掌握的孩子，差距在加速扩大。

名字来自《马太福音》25:29。Gladwell 把它放在 Outliers 第一章的扉页上：

“FOR UNTO EVERYONE THAT HATH SHALL BE GIVEN, AND HE SHALL HAVE ABUNDANCE. BUT FROM HIM THAT HATH NOT SHALL BE TAKEN AWAY EVEN THAT WHICH HE HATH.”

凡有的，还要加倍给他，使他多余。没有的，连他所有的也要夺去。

同一章里还有一个更直观的例子。加拿大青少年冰球队的选拔分年级，9 月 1 日为分界线。统计下来，顶级职业球员里 1、2、3 月出生的比例远高于 10、11、12 月。为什么？

  • 同一学年里，1 月出生的孩子比 12 月出生的大接近一岁。
  • 8 岁、10 岁这些关键年龄，一岁的体型和协调性差距是巨大的。
  • 大的孩子被选进”好队”→ 接受更多训练 → 差距进一步拉大 → 更高级别选拔时又被选上 → 最终进职业。
  • 同一年生的孩子，只因为月份差了几个月，走上了两条指数发散的轨道。

这个例子让人不舒服。它说明孩子的复利起点的很大一部分是父母和环境决定的，不是孩子能选的。

但这也是希望住的地方。如果起点的差距会被复利放大，那父母能控的那一点点早期优势，会被放大成人生的巨大差异。关键不是”给孩子灌输更多知识”。这是从清华刘嘉教授那里学到的一条最重要的：应试教育已经彻底失效，基于记忆和知识的教育毫无价值，所有知识都在大模型里。真正要做的是回到”孩子”身上：培养他自己的兴趣、AI 原生思维、第一性原理思维。

父母能复利的东西，不是”更多英语单词”，是这三件事：

  1. 一个正反馈的起点。 让孩子在某一个领域里能体验”我比周围人做得好一点点”。语言是最典型的，因为早识字会辐射到所有其他学科。
  2. 自信这个复利资产。 自信不是被赞美出来的。Carol Dweck 写过一本很好的书讲这件事，重点一句话：“Praising children’s intelligence harms motivation and it harms performance.” 夸聪明让孩子回避挑战，夸努力让孩子追求挑战。自信的复利来自对过程的反馈，不来自对结果的夸奖。
  3. 时间窗口本身。 孩子的 20+ 年是人生最长的一段不可逆复利。最有名的版本：每天 1% 的改进，一年后是 37 倍。数学很简单。重点是”每天”这个频率，在孩子身上有 20 年的连乘效果。

但得老实加一条警告。1.01^365 = 37.78 是标准指数数学，不是谁的原创，也不是孩子会自己自动实现的事。Coyle 关于”髓鞘 → 技能”的故事也是通俗化的比喻，严格神经科学里髓鞘和技能的因果关系比他说的弱得多。复利框架告诉你方向，不告诉你结果会自己来。父母的工作不是让 37.78 发生。是不让指数归零。 不切断孩子的兴趣。不剥夺内在动机。不把孩子当成自己简历的一部分。

收束

回到 Buffett 那个 2200 万长到 3490 亿的数字。它确实是一个关于耐心和复利的故事。但我写这篇文章不是为了讲 Buffett。

复利方程 Value(t) = Principal × (1 + rate)^time 的真正秘密，不是它能让钱变多。是它告诉你一件事：生活里其实没有”单独的”领域。 本金、利率、时间、杠杆这四个维度，在股市、学习、心智模型、孩子身上是同一套。你在任何一个领域学到的复利纪律，都是其他三个领域的杠杆。

股市的纪律、学习的选择、心智模型的组合、孩子的成长。表面看是四层不相关的东西。其实不是。这篇文章就住在它们之间的缝隙里——同一支笔画出的四笔。

那留给你一个问题：今天的某个决定，你在哪个维度上过早兑现了？ 是选了 r 很高但 t 会归零的东西？是有好本金却任它闲置？是有杠杆却没敢加？还是在复利一个负的 r？

方程给你四个变量。你今天要动哪一个？

也就这么回事。

Sources / 来源