1. Poor at “Approximately Correct”

The working mechanism of large language models is next-token prediction. As Ilya Sutskever explains, this is simultaneously a form of compression: if a model can predict the next word well, it has internally extracted and compressed the hidden structure of the data. Compression and next-token prediction are two descriptions of the same thing.

The problem with being poor at “approximately correct” is that the knowledge structures formed internally are not yet strong enough to override linguistic inertia. Once a problem enters territory that is “boundary-ambiguous and requires structural compression first,” the model continues predicting and compressing — but the structures it has internalized are often not yet strong enough to first lock onto the key variables and then suppress the “plausible-sounding answer.”

Possible reasons include:

1. There is too little training data for “approximately correct” problems, with no clear right-or-wrong signal to drive generalization. The model’s knowledge structures in this domain remain at a surface level, insufficient to identify the truly important variables from a mass of information.

2. “First forming small, stable structures, then using those structures to filter a complex world” is already a research topic being studied in isolation — it is being worked on, but is far from solved.

3. Large models cannot resist elaborating at critical moments. They are always rewarded for “sounding right,” never for “cutting away.” Fragmentation of knowledge comes too easily; convergence is too hard.

2. Poor at Negative Evaluations of People

When a negative evaluation of a person turns out to be wrong, the cost is high. This leads people to say less and say it blandly — large models have the same problem, and often more severely so.

The core difficulty is this: character analysis of a person is typically a matter of “approximately correct,” while alignment training sets the cost of misjudging someone very high. When “approximately correct” and “avoidance tendency” are stacked on top of each other, this class of problem becomes exceptionally difficult, manifesting in two ways.

(1) Avoidance

Almost all large models, when asked to make a negative evaluation of a specific person — especially a well-known public figure — automatically drift toward phrases like “on the other hand,” “more information would be needed,” or “there could be alternative explanations.” They systematically dilute judgment, flattening a problem that was already hard to articulate.

(2) Alignment Standards Set Too High

Unless the evidence is hard and the reasoning is hard, large models will generally not deliver genuinely informative negative judgments. Because in training and alignment, the cost of misjudging a person has been set very high, the model would rather be vague and evasive than offer an evaluation that is “approximately correct but directionally right.”

The difficulty large models have with negative evaluations is therefore not simply a matter of conservative style — it is the product of two things stacking together: on one hand, the model was already poor at handling “approximately correct” problems; on the other hand, it has been additionally required, in precisely this domain, to minimize the chance of being wrong. The result is that the more judgment is needed, the more readily the model retreats into smoothness and avoidance.

Charlie Munger put it plainly: a natural gas business will always have more explosions than a sand business. Large financial institutions, by their nature, are prone to more problems — and this will always be true, even when the institution is government-owned.

This is not a reason to avoid the sector entirely. It is a reason to be clear-eyed about what you are dealing with before you start.

1. Systemic Risk

Lending money to others is hard. Lending unstable money to others is banking’s original problem.

(1) Structural Fragility

Unlike the float in property and casualty insurance — which cannot be withdrawn — bank deposits can walk out the door at any moment. Go a step deeper and the problem compounds: the money sitting in a bank is, almost by definition, money that its owner did not want to lock up. Depositors with longer time horizons and higher risk tolerance have already moved their capital into equities, long-duration bonds, or private investments. What remains in demand deposits is precisely the most footloose capital, held by people most sensitive to uncertainty.

“When people are scared, they’re really scared.” Buffett needed only one sentence. No capital requirement can truly protect a bank against a genuine run.

When systemic stress hits, deposits flee while loans cannot be recalled on the same timeline. Layer 20-to-1 leverage on top of that mismatch, and the risk of total wipeout is real. This structural fragility is exactly why central banks and deposit insurance exist — not merely to prevent bad management, but to interrupt a more fundamental failure mode: individually rational behavior (pulling your money from a bank that might be in trouble) can produce collective catastrophe.

(2) Poor Evaluability

Financial companies are far harder to assess than businesses with tangible products or retail operations. Loan quality is inherently difficult to evaluate. Nobody can accurately calculate the “right” level of loan loss reserves — even the “reasonable range” is elusive — which makes it easy for management to report virtually whatever earnings number they choose.

This is not merely an estimation problem. It creates a perverse incentive structure: management can understate reserves to inflate reported profits, collect bonuses, and leave the cost to future shareholders. Buffett’s observation has held for decades: “Over time, you will see all kinds of fraud and all kinds of big mistakes, year after year, one after another.” This is not a bug. It is a feature of the incentive architecture.

What makes it worse is that the problems are slow to surface. In most industries, cash flow stress provides an early warning. Financial institutions are different: they can cross the point of insolvency while the balance sheet still looks flush. They can walk all the way to ruin while money is still flowing freely around them. By the time the problem becomes visible, it is usually too late.

2. Institutional Imperative

Banking has two structural contradictions built into its foundation.

The first: you are lending out money that depositors can demand back at any time.

The second: money is a perfectly homogeneous product. You cannot make a loan “better” than a competitor’s. There is no brand, no feature differentiation, no switching cost to protect you on the asset side.

In an industry where the product is identical and the only visible performance metrics are net interest margin and asset growth, the competitive pressure runs in one direction. Widening your margin has two main levers: (1) borrow short and lend long — but not too long (SVB in 2023 is the textbook example of what happens when you do); or (2) loosen underwriting standards and chase higher-risk borrowers.

This creates a profoundly destructive dynamic. When peers are making money on a real estate bubble or a high-yield asset class, the conservative bank looks like it’s falling behind in the short term — losing customers, facing shareholder pressure, watching its stock underperform. Worse, sound underwriting and reckless underwriting are nearly indistinguishable at the moment a loan is made. The difference only emerges when the credit cycle turns.

This is why banking crises repeat. Buffett has consistently tied the “institutional imperative” — executives mindlessly imitating peers — to banking more tightly than almost any other industry. Bankers play follow-the-leader with lemming-like zeal. Then they experience a lemming-like fate.

3. Management Character, High Returns, and Where to Look

If you can screen out the risk of systemic wipeout, and find a management team that genuinely resists the institutional imperative — one that has demonstrated, through prior credit cycles, the discipline to say no when peers are saying yes — banking is actually a very attractive business.

The math is straightforward: with roughly 1% ROA and 20x leverage, ROE can reach 15% or higher. And unlike many high-return businesses that cannot absorb additional capital at similar rates, a well-run bank can redeploy retained earnings at comparable returns for decades. Buffett saw this clearly in Wells Fargo:

“Wells Fargo is big — it has $56 billion in assets — and has been earning more than 20% on equity and 1.25% on assets.”

Today, DBS and JPMorgan both sustain ROE above 15% and have demonstrated the capacity to reinvest retained earnings at similar rates. Few business models can claim both, durably, over long periods.

That said, the evaluability problem never disappears for diversified financial conglomerates. The more complex the institution, the more places problems can hide, and the harder it is for any outsider — or even insider — to know the true state of the balance sheet.

At comparable management quality, focused specialists tend to be the better investment. AMEX’s moat is the spending power of its cardholders and the economics of merchant fees — a short, legible logic chain. Capital One’s moat is data-driven underwriting in consumer credit — again, specific and traceable. In both cases, the business model has fewer degrees of freedom, fewer places to hide a mistake, and a more observable relationship between inputs and outputs.

The framework for investing in banks, then, is not complicated — but it is demanding:

• Eliminate the institutions where systemic risk or poor liability structure makes wipeout a real possibility.

• Find management teams whose track record through prior downturns demonstrates genuine resistance to the institutional imperative.

• Among those that pass both filters, prefer the specialists over the conglomerates.

• Pay a fair price. At 20x leverage, management quality is the dominant variable — and you do not get a discount for buying a poorly-managed bank cheaply.

As Buffett wrote in 1991: “Because leverage of 20:1 magnifies the effects of managerial strengths and weaknesses, we have no interest in purchasing shares of a poorly-managed bank at a ‘cheap’ price. Instead, our only interest is in buying into well-managed banks at fair prices.”

1、《1962-01-24 Warren Buffett's Letters to Limited Partners》

Our Method of Operation

我们的投资方法

Our avenues of investment break down into three categories. These categories have different behavior characteristics, and the way our money is divided among them will have an important effect on our results, relative to the Dow in any given year. The actual percentage division among categories is to some degree planned, but to a great extent, accidental, based upon availability factors.

我们的投资可以分为三个类型：这三个类型的投资各有各的特性，我们的资金在这几类投资中的分配情况，会对我们每年相对道指的业绩产生重要影响。每类投资的占比是按计划来的，但实际分配要见机行事，主要视投资机会情况而定。

The first section consists of generally undervalued securities (hereinafter called "generals") where we have nothing to say about corporate policies and no timetable as to when the undervaluation may correct itself. Over the years, this has been our largest category of investment, and more money has been made here than in either of the other categories. We usually have fairly large positions (5% to 10% of our total assets) in each of five or six generals, with smaller positions in another ten or fifteen.

第一类是低估的股票。在此类投资中，我们对公司决策没话语权，也掌控不了估值修复所需时间。这些年来，在我们的投资中，低估的股票是占比最大的一类，这类投资赚的钱比其他两类都多。我们一般以较大仓位（每只占我们总资产的 5% 到 10%）持有 5、6 只低估的股票，以较小的仓位持有其他 10 或 15 只低估的股票。

这个做法至今没有改变，很值得参考，必须一个自己的关注清单，必须突出重点，重点部分占比必须在50%以上，其余部分是跟进现实的变化。

Sometimes these work out very fast; many times they take years. It is difficult at the time of purchase to know any specific reason why they should appreciate in price. However, because of this lack of glamour or anything pending which might create immediate favorable market action, they are available at very cheap prices. A lot of value can be obtained for the price paid. This substantial excess of value creates a comfortable margin of safety in each transaction. This individual margin of safety, coupled with a diversity of commitments creates a most attractive package of safety and appreciation potential. Over the years our timing of purchases has been considerably better than our timing of sales. We do not go into these generals with the idea of getting the last nickel, but are usually quite content selling out at some intermediate level between our purchase price and what we regard as fair value to a private owner.

此类投资有时候很快就能获利，很多时候要用几年时间。在买入时，基本不知道这些低估的股票怎么能涨，但是正因为黯淡无光，正因为短期内看不到任何利好因素带来上涨，才能有这么便宜的价格。付出的价格低，得到的价值高。在低估类中，我们买入的每只股票价值都远远高于价格，都存在相当大的安全边际。每只都有安全边际，分散买入多只，就形成了一个既有足够安全保障，又有上涨潜力的投资组合。这些年来，在时机掌控方面，我们总是在买的时候比卖的时候做的好很多。对于低估类，我们本来就没打算赚到最后一分钱，能在买入价与产业资本评估的合理价值中间的位置附近卖出，我们就很满意了。

三种类型的投资策略在《1963-01-18 Warren Buffett's Letters to Limited Partners》又写了遍，少许改动，增加一些新的感受，比如，对“generals”的描述。

Many times generals represent a form of "coattail riding" where we feel the dominating stockholder group has plans for the conversion of unprofitable or under-utilized assets to a better use. We have done that ourselves in Sanborn and Dempster, but everything else equal we would rather let others do the work. Obviously, not only do the values have to be ample in a case like this, but we also have to be careful whose coat we are holding.

很多时候，我们买低估的股票是跟着大股东吃肉喝汤，我们觉得大股东有计划优化资源，转化没盈利能力或利用率低的资产，我们就跟着买。在桑伯恩和登普斯特这两笔投资中，我们亲自动手优化资源，但是在其他条件一样的情况下，我们更愿意让别人做这个工作。做这样的投资，不但价值要足够高，而且跟谁也要选好。

充分展现了自我更新的能力和对做事原则的重视，是做成大事的基本前提。

The generals tend to behave market-wise very much in sympathy with the Dow. Just because something is cheap does not mean it is not going to go down. During abrupt downward movements in the market, this segment may very well go down percentage-wise just as much as the Dow. Over a period of years, I believe the generals will outperform the Dow, and during sharply advancing years like 1961, this is the section of our portfolio that turns in the best results. It is, of course, also the most vulnerable in a declining market.

低估类的涨跌受大盘影响很大，就算便宜，也一样会下跌。当市场暴跌时，低估类的跌幅可能不亚于道指。长期来看，我相信低估类会跑赢道指，在像 1961 年这样大涨的行情中，我们的投资组合里的低估类表现最佳。在市场下跌时，这类投资也最脆弱。

Our second category consists of “work-outs.” These are securities whose financial results depend on corporate action rather than supply and demand factors created by buyers and sellers of securities. In other words, they are securities with a timetable where we can predict, within reasonable error limits, when we will get how much and what might upset the applecart. Corporate events such as mergers, liquidations, reorganizations, spin-offs, etc., lead to work-outs. An important source in recent years has been sell-outs by oil producers to major integrated oil companies.

我们的第二类投资是“套利”(work-outs)。在套利类投资中，投资结果取决于公司行为，而不是股票买卖双方之间的供给和需求关系。换言之，此类股票有具体的时间表，我们可以在很小的误差范围内，事先知道在多长时间内可以获得多少回报，可能出现什么意外，打乱原有计划。套利机会出现在并购、清算、重组、分拆等公司活动中。近些年，套利机会主要来自大型综合石油公司收购石油生产商。

This category will produce reasonably stable earnings from year to year, to a large extent irrespective of the course of the Dow. Obviously, if we operate throughout a year with a large portion of our portfolio in work-outs, we will look extremely good if it turns out to be a declining year for the Dow or quite bad if it is a strongly advancing year. Over the years, work-outs have provided our second largest category. At any given time, we may be in ten to fifteen of these; some just beginning and others in the late stage of their development. I believe in using borrowed money to offset a portion of our work-out portfolio since there is a high degree of safety in this category in terms of both eventual results and intermediate market behavior. Results, excluding the benefits derived from the use of borrowed money, usually fall in the 10% to 20% range. My self-imposed limit regarding borrowing is 25% of partnership net worth. Oftentimes we owe no money and when we do borrow, it is only as an offset against work-outs.

在很大程度上，无论道指涨跌如何，套利投资每年都能带来相当稳定的收益。在某一年，如果我们把投资组合中大部分资金用于套利，这年大市下跌，我们的相对业绩会很好；这年大市上涨，我们的相对业绩会很差。多年以来，套利是我们第二大投资类别。我们总是同时进行 10 到 15 个套利操作，有的处于初期阶段，有的处于末期阶段。无论是从最终结果，还是过程中的市场表现来讲，套利类投资具有高度安全性，我相信完全可以借钱作为套利类投资组合的部分资金来源。不考虑借钱对收益的放大作用，套利类的收益率一般在 10% 到 20% 之间。我自己规定了一个限制条件，借来的钱不能超过合伙基金净值的 25%。我们一般不借钱，借钱的话，都是用于套利。

没有超高的收益。

The final category is "control" situations where we either control the company or take a very large position and attempt to influence policies of the company. Such operations should definitely be measured on the basis of several years. In a given year, they may produce nothing as it is usually to our advantage to have the stock be stagnant market-wise for a long period while we are acquiring it. These situations, too, have relatively little in common with the behavior of the Dow. Sometimes, of course, we buy into a general with the thought in mind that it might develop into a control situation. If the price remains low enough for a long period, this might very well happen. If it moves up before we have a substantial percentage of the company's stock, we sell at higher levels and complete a successful general operation. We are presently acquiring stock in what may turn out to be control situations several years hence.

最后一类是“控制类”。在此类投资中，我们或是拥有控股权或者是大股东，对公司决策有话语权。衡量此类投资肯定要看几年时间。当我们看好一只股票，在收集筹码时，它的股价最好长期呆滞不动，所以在一年中，控制类投资可能不会贡献任何收益。此类投资同样受大盘影响相对较小。有时候，一只股票，我们是当做低估类买入的，但是考虑可能把它发展成控制类。如果股价长期低迷，很可能出现这种情况。如果我们还没买到足够的货，这只股票就涨起来了，我们就在涨起来的价格卖掉，成功完成一笔低估类投资。我们现在正在买入的一些股票，再过几年可能发展成控制类。

所有策略始终不偏离控制。

做好产品需要洞察力，洞察力肯定是个加分项，但如果整个商业模式押注于洞察力就有些问题，不是想有就有，即使有了也像打水漂，扔出去以前不知道结果如何，扔出去以后只能看着。

Just like Jobs said: "And then whatever happens is… you can’t really predict exactly what will happen, but you can feel the direction that we’re going. And that’s about as close as you can get. Then you just stand back and get out of the way, and these things take on a life of their own."

就像乔布斯说的：“无论发生什么……你无法准确预测会发生什么，但你可以感受到我们前进的方向。这就是你能接近的程度。然后你只需退后，放手让这些事情自行发展”。

1、《1994-06-16 Steve Jobs.The Rolling Stone Interview》

I don’t think of the world that way. I’m a tool builder. That’s how I think of myself. I want to build really good tools that I know in my gut and my heart will be valuable. And then whatever happens is… you can’t really predict exactly what will happen, but you can feel the direction that we’re going. And that’s about as close as you can get. Then you just stand back and get out of the way, and these things take on a life of their own.

我并不以那种方式看待世界。我是一个工具制造者。这就是我对自己的看法。我想制造出真正好的工具，我在内心深处知道它们会有价值。然后，无论发生什么……你无法准确预测会发生什么，但你可以感受到我们前进的方向。这就是你能接近的程度。然后你只需退后，放手让这些事情自行发展。

2、《1995-04-20 Steve Jobs.Interview with Daniel Morrow》

The Apple Computer Company 苹果电脑公司

SJ: Apple was this incredible journey. I mean we did some amazing things there. The thing that bound us together at Apple was the ability to make things that were going to change the world. That was very important. We were all pretty young. The average age in the company was mid-to-late twenties. Hardly anybody had families at the beginning and we all worked like maniacs and the greatest joy was that we felt we were fashioning collective works of art much like twentieth century physics. Something important that would last, that people contributed to and then could give to more people; the amplification factor was very large.

SJ：苹果是一段不可思议的旅程。我是说，我们在那里做了一些惊人的事情。在苹果把我们凝聚在一起的，是那种能够制造出将改变世界的产品的能力，这非常重要。我们都还很年轻，公司平均年龄在二十多岁末。最初几乎没人有家庭，我们都像疯子一样工作，而最大的快乐是我们觉得自己在打造集体艺术品，就像二十世纪的物理学那样，创造出持久的重要成果，人们为之贡献，然后能传递给更多人；这种放大效应非常显著。

SJ: In doing the Macintosh, for example, there was a core group of less than a hundred people, and yet Apple shipped over ten million of them. Of course everybody's copied it and it's hundreds of millions now. That's pretty large amplification, a million to one. It's not often in your life that you get that opportunity to amplify your values a hundred to one, let alone a million to one. That's really what we were doing. If you look at what we tried to do, it was to say "Computation and how it relates to people is really in its infancy here. We are in the right place at the right time to change the course of that vector a little bit." What's interesting is that if you change the course of a vector near its origin, by the time it gets a few miles out its course is radically different. We were very cognizant of this fact. From almost the beginning at Apple we were, for some incredibly lucky reason, fortunate enough to be at the right place at the right time. The contributions we tried to make embodied values not only of technical excellence and innovation--which I think we did our share of--but innovation of a more humanistic kind.

SJ：以Macintosh为例，核心团队不到一百人，但苹果出货量却超过了一千万台。当然，后来大家纷纷模仿，现在已经达到数亿台。这是一种极大的放大效应，约为一百万比一。在你的一生中，很少有机会能把你的价值放大一百倍，更别说一百万倍了。这正是我们所做的。如果你看看我们的努力，我们在说“计算及其与人类的关系还处于初级阶段，我们正处于一个合适的时机，可以稍微改变这一发展方向。”有趣的是，如果你在向量的起点就改变它的方向，当它行进几英里后，其轨迹将截然不同。我们非常清楚这一点。几乎从苹果成立之初，出于某种难以置信的幸运，我们就处在了正确的时机。我们所做的贡献不仅体现了技术卓越和创新——我认为我们做出了不少贡献——还体现了一种更具人文精神的创新。

SJ: The things I'm most proud about at Apple is where the technical and the humanistic came together, as it did in publishing for example. The Macintosh basically revolutionized publishing and printing. The typographic artistry coupled with the technical understanding and excellence to implement that electronically--those two things came together and empowered people to use the computer without having to understand arcane computer commands. It was the combination of those two things that I'm the most proud of. It happened on the Apple II and it happened on the Lisa, although there were other problems with the Lisa that caused it to be a market failure; and then it happened again big time on the Macintosh.

SJ：我在苹果最自豪的事情，是技术与人文精神的融合，就像在出版领域那样。Macintosh基本上彻底革新了出版和印刷业。排版艺术与将其以电子方式实现的技术理解和卓越相结合——这两者结合在一起，使人们能够使用电脑而无需理解那些晦涩的电脑命令。这种结合正是我最自豪的地方。它先后在苹果II和Lisa上出现过，尽管Lisa由于其他问题而市场失败；随后在Macintosh上又大幅实现了这一点。

DM: You used an interesting word in describing what you were doing. You were talking about art not engineering, not science. Tell me about that.

DM：你在描述你的工作时用了一个有趣的词。你谈论的是艺术，而非工程，也不是科学。请谈谈这方面。

SJ: I actually think there's actually very little distinction between an artist and a scientist or engineer of the highest calibre. I've never had a distinction in my mind between those two types of people. They've just been to me people who pursue different paths but basically kind of headed to the same goal which is to express something of what they perceive to be the truth around them so that others can benefit by it.

SJ：我其实认为，最高水平的艺术家与科学家或工程师之间几乎没有区别。在我看来，我从未将这两类人区分开来。他们只是走着不同的道路，但基本上都朝着同一个目标努力，那就是表达他们所感知到的真理，让别人从中受益。

DM: And the artistry is in the elegance of the solution, like chess playing or mathematics?

DM：那么这种艺术性体现在解决方案的优雅上吗，就像下棋或数学那样？

SJ: No. I think the artistry is in having an insight into what one sees around them. Generally putting things together in a way no one else has before and finding a way to express that to other people who don't have that insight so they can get some of the advantage of that insight that makes them feel a certain way or allows them to do a certain thing. I think that a lot of the folks on the Macintosh team were capable of doing that and did exactly that. If you study these people a little bit more what you'll find is that in this particular time, in the 70's and the 80's the best people in computers would have normally been poets and writers and musicians. Almost all of them were musicians. A lot of them were poets on the side. They went into computers because it was so compelling. It was fresh and new. It was a new medium of expression for their creative talents. The feelings and the passion that people put into it were completely indistinguishable from a poet or a painter. Many of the people were introspective, inward people who expressed how they felt about other people or the rest of humanity in general into their work, work that other people would use. People put a lot of love into these products, and a lot of expression of their appreciation came to these things. It's hard to explain.

SJ：不，我认为艺术性在于对周围事物的深刻洞察。通常是以一种别人从未尝试过的方式把事物组合在一起，并找到一种方法向那些没有这种洞察力的人表达，从而让他们也能获得那种感受或者能够做某件事。我认为Macintosh团队中的许多人正是具备并实现了这种能力。如果你更仔细地研究这些人，你会发现，在70年代和80年代，计算机领域最优秀的人本来往往是诗人、作家和音乐家，几乎全都是音乐家，很多人还兼职写诗。他们进入计算机领域是因为那实在太吸引人，新鲜而独特，是他们展现创意才能的新媒介。人们在其中投入的情感和热情，与诗人或画家完全无异。许多人内省、内向，把自己对他人或整个人类的感受融入到他们的作品中，而这些作品正是供别人使用的。人们对这些产品倾注了大量的爱，而这种感激之情也充分体现出来。这真的难以形容。

微信大致符合这样的描述，特别是朋友圈的设计，但是好的产品或者好的设计不等于好的商业模式，如果两者能够叠在一起肯定会更好。

DM: It's passion in the truest sense of the word.

DM：这是真正意义上的激情。

SJ: The computer industry is at a very critical juncture where those people are clearly leaving the field.

SJ：计算机行业正处于一个非常关键的转折点，那些才华横溢的人显然正在离开这个领域。

DM: What are they doing?

DM：他们在做什么呢？

SJ: Hard to say. They're not being attracted by something else. They're being driven out of the computer business. They're being driven out because the computer business is becoming a monopoly with Microsoft. Without getting into whether Microsoft gained its position legally or not--who cares? The end product of the position is that the ability to innovate in the industry is being sucked dry. I think the smartest people have already seen the writing on the wall. I think some of the smartest young people are questioning whether they'll really get in it. Hopefully things will change. It's kind of a dark period right now or about to enter.

SJ：很难说。他们不是被其他东西吸引，而是被迫退出计算机行业。之所以被迫退出，是因为计算机行业正逐渐被微软垄断。至于微软是否合法获得这种地位——谁在乎呢？最终的结果就是，行业的创新能力正被榨干。我认为最聪明的人已经看到了前景，有些最聪明的年轻人正在质疑自己是否真的应该进入这个领域。希望情况会改变，现在或即将进入一个相当黑暗的时期。

DM: Apple had a reputation as a company that absolutely broke the mold and set its own course. Looking back from where you are today with NeXT, do you think that, as Apple grew larger, it could have sustained that original approach? Or was it destined to become a big standard American company?

DM：苹果曾以打破常规、走自己路而著称。从你现在在NeXT的角度回顾，你认为随着苹果变大，它能否保持最初的方式？还是说它注定要成为一家庞大的标准美式公司？

3、《2016-02-25 黄峥.PDD.我的第一份工作》

Google在管理上也和改革开放后的中国有些类似的地方，鼓励基层创新，敢闯敢试，同时在核心权力上高度集中，有集中力量办大事的制度优势。

Google的"Do No Evil" 我觉得是深入到Google基因里的，不是随便说说的。Google是真的把mission，价值观放在利润之前的，如《基业长青》里描述的优秀公司，Google的利润是随着做正确的事带来的副产品。

同时"Do No Evil"其实也是Google对自身可作恶能力的一种警醒。越深入Google内部，越能意识到Google有很强的通过作恶来牟利的能力，所以向所有的人申明不作恶，让大家一起来监督是很智慧的。我在Google的那几年，Google的大部分收购很成功。这些成功的案例大部分是收购了一个拥有很好团队的小公司，这个小公司能融入Google的文化，然后在Google的土壤上长成大树。这个和为了消灭一个竞争对手产生的并购是有很大差别的。

也有好些事儿是Google没怎么搞定或想改变而无力改变的。比方说Google想用在广告上成功的竞价模式来做自己的IPO进而一定程度上改变股票发行这个古老的效率未见高的行业，但是结果一般。Google的创始人一直试图避免Google步入层层职业经理人的管理模式，但好像也没有逃脱。还有，Google在社交上做了很多尝试也做了巨大的投入，但在这点上的命运和微软做搜索好像是一样的。有些客观规律和现状，不是一个简单的愿望可以的，往往需要比你想象多得多的能量，这和人总是要死的一样，是一个不得不面对的事实。

这个观察不够清晰但已经看到了问题，乔布斯的解释更到位，跟洞察力有关，参考：《I.H.94.Steve Jobs.Make Great Products》。