AI Interview Wiki
主题综述

信念高于共识:最好的决策最有争议 · Conviction Over Consensus

主题综述

更新日志

主流共识

这五个人用各自的词讲同一件事:多数人同意的决定,不产生 alpha。

最硬的证据来自 Sequoia,因为它不是信仰而是十年的内部数据:

"We've been recording the number that everybody votes on every investment for more than a decade now. Our internal data shows that consensus versus non-consensus does not matter at all."
「十多年来,我们一直在记录每个人对每项投资的投票数。我们的内部数据显示,共识与非共识根本不重要。」
Pat Grady · Uncapped #36 Pat Grady & Alfred Lin from Sequoia
"Presence of conviction is what matters. And so if everybody is a six... If everybody's a six, probably shouldn't make the investment. It's consensus, but nobody has conviction."
「重要的是信念的存在……如果每个人都是六分,可能不应该进行投资。这是共识,但没有人有信念。」
Pat Grady · Uncapped #36 Pat Grady & Alfred Lin from Sequoia

Ben Horowitz 把同一个命题从投资搬到领导:

"If everybody agrees with the decision, then you didn't add any value because they would have done that without you. So the only value you ever add is when you make a decision that most people don't like."
「如果每个人都同意这个决定,那么你就没有增加任何价值,因为没有你他们也会这样做。所以你唯一能增加价值的时候就是当你做出一个大多数人都不喜欢的决定时。」
Ben Horowitz · Ben Horowitz Why Hesitation is a CEO's Worst Enemy

Keith Rabois 把它压成一句行业格言——光逆向不够,要逆向"且"对:

"You have to be contrarian and right, and not just bet on people... but you also have to be right, you know, a reasonable fraction."
「你必须是逆向投资者并且是对的,而不仅仅是押注于人,而且你还必须是正确的,而且概率要合理。」
Keith Rabois · Comparative Advantages | Keith Rabois

Sarah Guo 把它做成了公司的名字和品牌内核:

"Be willing to take risk on having an opinion is an important one for us... But part of the core ethos for me about venture is I want to understand the new thing and then we take risk on it. We published our LP letters recently. Some of those beliefs and predictions are going to be wrong and look very stupid."
「勇于冒险发表观点对我们来说很重要……但对我来说,风险投资的核心精神之一是,我想了解新的事物,然后我们冒险尝试。我们最近公布了我们的 LP 信。其中一些信念和预测可能是错误的,而且显得很愚蠢。」
Sarah Guo · Investing with Conviction | Sarah Guo

Vinod Khosla 在描述自己和 Founders Fund 共有的取向时,给出了最简洁的版本:

"Very, very similar goals in betting on really bold ideas, whether they're popular or not. Whether they're on trend or not."
「在押注真正大胆的想法方面,目标非常相似,无论它们是否受欢迎。无论它们是否符合潮流。」
Vinod Khosla · Uncapped #40 Vinod Khosla and Keith Rabois from Khosla

Sam Altman 从创始人一侧给出同一命题最极端的版本——而且他的对立面不是同行 VC,是整个领域的权威:

"Most of the world still does not understand the value of like a fairly extreme level of conviction on one bet."
「世界上大多数人仍然不理解在一个赌注上投入相当极端的信念的价值。」
Sam Altman · How To Build The Future: Sam Altman
"Everybody had some reason of, oh, it's not really learning. It's not really reasoning... And these were like the eminent leaders of the field. And more than just saying, you're wrong, they were like, you're wrong and this is... It was like a bad thing to believe or a bad thing to say."
「每个人都有理由说,哦,它不是真的在学习。它不是真的在推理……而这些人都是该领域的杰出领导者。他们不仅仅是说你错了,他们还说,你错了,而且这……相信或说出这种话是不好的。」
Sam Altman · How To Build The Future: Sam Altman

Shardul Shah 则把它压成一句资产类别的定义——共识本身就在错误的生意里:

"I'm not a value investor, right? I believe in the power law... So I'm not seeking average returns. I'm not seeking good deals. I'm looking for outliers."
「我不是价值投资者,对吧?我相信幂定律……所以我不是在寻求平均回报。我不是在寻找好的交易。我正在寻找异类。」
Shardul Shah · 20VC: Index's Shardul Shah on Why Market Size is a Trap
"I don't know why we're talking about averages. None of us are in the business of mean reversion."
「我不知道我们为什么要谈论平均数。我们没有一个人是在做均值回归的生意。」
Shardul Shah · 20VC: Index's Shardul Shah on Why Market Size is a Trap

共识到此为止:好决策最有争议;追求皆大欢喜等于自动放弃 alpha。

分歧在哪

把这句"信念高于共识"摊开,五个人各说各的——而且分歧恰恰藏在那个被他们当作共识的口号底下。

一、信念是天降的直觉,还是数据逼出来的纪律?

Keith Rabois 的信念几乎是瞬间的、近乎神启的:

"Airbnb, like instantly, Brian will tell you, like, three minutes after his monologue, three minutes in, I was like, this is the coolest thing since YouTube. Everybody else was totally wrong. There were people who thought they were cool, interesting founders... but nobody else thought it was a good idea."
「就像 Airbnb,立刻,Brian 会告诉你,在他的独白三分钟后,我就觉得,这是继 YouTube 之后最酷的事情。其他人都完全错了。有些人认为他们是很酷、很有趣的创始人……但没有其他人认为这是个好主意。」
Keith Rabois · Comparative Advantages | Keith Rabois

而且他自陈,最好的项目他事先就有把握——信念几乎是确定性的同义词:

"Ninety percent of the best stuff I've ever invested in as an angel or as a VC, I was dead sure of."
「作为天使投资人或风险投资人,我投资过的最好的项目,90% 我都非常有把握。」
Keith Rabois · Comparative Advantages | Keith Rabois

Sequoia 走的是反向路线:他们不信任直觉,把直觉当成需要排查的污染源。同样是逆向决策,他们事后做的是冷峻的归因:

"Every single one of them, if you play the five whys game... every single one of them comes down to some psychological bias or emotional trap. None of them come down to an error in calculations... We have a list of 40 of these things."
「如果你玩‘五个为什么’游戏……每一个案例最终都归结为某种心理偏见或情感陷阱。没有一个是因为计算错误……我们列出了 40 个这样的问题。」
Pat Grady · Uncapped #36 Pat Grady & Alfred Lin from Sequoia

Khosla 给出第三条路——信念来自把决策拆成可辩论的变量,既不是直觉也不是偏见清单:

"It's first principles thinking. If you can do first principles thinking, it's easy to know where you agree and where you disagree... We'll even come down to if X were true, then this is a good decision or a bad decision."
「就是第一性原理思考。如果你能进行第一性原理思考,就很容易知道你在哪些方面同意,在哪些方面不同意……我们甚至会深入到,如果 X 是真的,那么这是一个好的决定还是一个坏的决定。」
Vinod Khosla · Uncapped #40 Vinod Khosla and Keith Rabois from Khosla

三种"信念观":Keith 的瞬时确定、Sequoia 的去偏纪律、Khosla 的变量拆解。都打着 conviction 的旗号,获取信念的机制完全不同。

二、信念是个体的,还是机制的?

Ben Horowitz 把信念彻底个人化:领导力的本质就是一个人扛下别人都不喜欢的决定——价值恰恰来自孤身一人。

Sequoia 反而把它机制化——他们不要一个英雄,要的是投票分布里那个"九":

"If three people are nines and three people are one, we should probably make the investment, because the presence of the nines is a much more powerful signal than the presence of the ones."
「如果三个人是九分,三个人是一分,我们可能应该进行投资,因为九分的存在比一分的存在更具说服力。」
Pat Grady · Uncapped #36 Pat Grady & Alfred Lin from Sequoia

这是一个被低估的张力:Ben 的"多数人反对"和 Sequoia 的"强 yes 强 no"看似同盟,实则相反。Ben 要的是力排众议的个人;Sequoia 要的是容纳强烈分歧的合伙人制——他们甚至刻意保护下属去做自己看不懂的事:

"When somebody joins our team, we don't want David Kahn to become another Alfred or another Pat... if David goes off and does something that's not obvious to me or Alfred, I'm gonna be pretty curious. We know this guy's incredibly talented. What does he see that we don't see?"
「当有人加入我们的团队时,我们不希望 David Kahn 变成另一个 Alfred 或另一个 Pat……如果 David 去做一些我和 Alfred 不理解的事情,我会很好奇。我们知道这家伙非常有才华。他看到了我们没看到的东西吗?」
Pat Grady · Uncapped #36 Pat Grady & Alfred Lin from Sequoia

Khosla 又不一样——他和 Keith 的模式是:个人信念,但被一个能信任的搭档校准。Keith 形容那次拿不定主意时,是 Vinod 替他锁定了关键变量。信念在这里既不是孤胆也不是投票,而是双人协议。

David George 给出第四种,恰好卡在 Ben 的孤胆和 Sequoia 的投票之间:a16z 成长基金刻意废掉了投票委员会,改成"单人扣扳机"——鼓励充分分歧,但决定权归一个人:

"Most growth equity investment firms have an investment committee... You go, you present, you battle to get the votes... And what we decided to do at the firm in the growth fund was do it totally differently... single trigger puller... we openly expect disagreement. But once you disagree, you disagree and then you commit."
「大多数增长股权投资公司都有一个投资委员会……你去那里,你展示,你努力争取选票……我们在增长基金中决定做的事情是完全不同的……单人扣扳机……我们公开期待不同的意见。但一旦你不同意,你就是不同意,然后你承诺。」
David George · David George — Building a16z Growth

他给的理由正好是 Sequoia 投票制的反命题:投票会逼人拉票、把决策变成政治:

"You're never in this temptation to sell or to politic for a vote or try to influence someone's decision for the wrong reasons... So I think it allows us to more openly explore the merits of an investment."
「你永远不会受到诱惑去出售,或者为了选票而搞政治,或者因为错误的原因试图影响别人的决定……所以我认为这让我们能够更公开地探讨投资的优点。」
David George · David George — Building a16z Growth

于是"信念是个体还是机制"变成了四种答案:Ben 的孤胆、Sequoia 的投票分布、Khosla 的双人协议、David George 的"单人拍板+公开异议"。Sequoia 说票数分布是最可靠的信号,David George 却说投票机制本身会污染信念——两家都在管几百亿美元,给出的治理结论正相反。

三、信念要不要止损?逆向是永久立场还是中转站?

这是阵营内最尖锐、也最少被并置的分歧。

Keith Rabois 明确说:逆向有保质期,最终必须转化成共识,否则不算成功。

"You don't want to stay contrarian forever. To be really successful, eventually you need consensus that this is a good product, good thing to adopt, good thing to pay for. So you want to convert contrarian to consensus and then take advantage of that momentum."
「你不能永远保持逆向思维。要想真正成功,最终你需要获得共识,即这是一个好的产品,值得采用,值得付费。所以你要把逆向思维转化为共识,然后利用这种势头。」
Keith Rabois · Comparative Advantages | Keith Rabois

Alfred Lin 的表述则更像把逆向当成结构性常态,而非中转站——因为未来本来就不长得像过去:

"We're in the risk-taking business. We need that volatility, because the truth is not somewhere in the middle where everybody agrees. We're trying to build the future. The future does not look like the past, and so if everything looks consensus... when it's consensus, it all gets priced to consensus too."
「我们从事的是冒险业务。我们需要这种波动性,因为事实并非介于每个人都同意的中间立场。我们试图创造未来。未来看起来不像过去,所以如果一切看起来都达成共识……当达成共识时,一切都会被定价为共识。」
Alfred Lin · Uncapped #36 Pat Grady & Alfred Lin from Sequoia

两人没直接交锋,但放在一起就是真分歧:Keith 把"逆向→共识"看成必经的获利路径(共识是退出的流动性);Alfred 把共识本身看成定价已尽、超额已无的信号。一个把共识当终点,一个把共识当警报。

Sam Altman 从第三个角度切进这个分歧:他不问逆向何时转共识,而问信念何时该被杀死。他给的止损线是"数据"——OpenAI 也曾多次高信念且错,关键是数据一到就认:

"There is a difference between being high conviction just for the sake of it... Conviction is great until the moment you have data one way or the other. And there are a lot of people who hold on it past the moment of data."
「为信念而信念是有区别的……在你获得数据之前,信念是伟大的。很多人在数据出现后仍然坚持己见。」
Sam Altman · How To Build The Future: Sam Altman
"But there is a long moment where you have to be willing to operate without data. And at that point, you do have to just sort of run on conviction."
「但在很长一段时间里,你必须愿意在没有数据的情况下运作。在那一点上,你确实必须凭着信念前进。」
Sam Altman · How To Build The Future: Sam Altman

这是三条不同的止损观并置:Keith 说逆向到共识就退出,Alfred 说到共识就是警报,Altman 说到数据就该认错——三人切的都是"信念该在哪一刻让位",却指向完全不同的那一刻(退出流动性、定价饱和、证伪证据)。

值得注意的是,连"热门项目到底差不差"他们内部也不完全统一。Keith 和 Vinod 在 Khosla 那期斩钉截铁:

"I think the consensus ones aren't going to outperform the outliers at all."
「我认为共识投资的表现根本不会超过异类。」
Keith Rabois · Uncapped #40 Vinod Khosla and Keith Rabois from Khosla

但 Alfred Lin 复盘 DoorDash 时承认,真相更杂:有些轮次共识也赚钱,只是最好的两轮恰好非共识——他甚至承认种子轮自己因"信念不够"而错过:

"E, F, G, those were super consensus. They all made money. But the best rounds were the Series A and the Series C, and those were not consensus at all."
「E、F、G 轮,这些都非常一致。他们都赚了钱。但最好的几轮是 A 轮和 C 轮,而这些根本没有达成共识。」
Alfred Lin · Uncapped #36 Pat Grady & Alfred Lin from Sequoia

所以"非共识更好"在口号层面是共识,在账本层面是有条件的:共识也能赚钱,只是赚不到超额。

四、对谁该有信念:对人,还是对赌注?

还有一层措辞底下的错位。Keith 强调,逆向押的几乎总是"人";Vinod Khosla 则指出,越早期,共识热门越是关于人而非想法:

"I think at the seed level, consensus hotbeds are generally around people. Uh, not around the idea."
「我认为在种子轮阶段,热门共识通常围绕着人。而不是围绕着想法。」
Vinod Khosla · Uncapped #40 Vinod Khosla and Keith Rabois from Khosla

Sarah Guo 把信念落在"对世界的判断"——一封会被打脸的 LP 信、一个押在 AI 上的世界观。Ben Horowitz 落在"对处境的判断"——18 个月就带着 200 万收入流血上市,因为另一条路是破产。Sequoia 落在"对未来形状的判断"。同一个 conviction,标的物从人、到时代、到处境、到未来,各不相同。这意味着"信念高于共识"在不同人嘴里,赌的根本不是同一类东西。

Alex Sacerdote 又把"逆向"这件事本身挪了个位置——多数人以为超额来自早期押小众,他说最大的错配恰恰在人人都认识的大公司身上:

"It takes 100 people, 100 diversified PMs to realize Google's not a loser, it's a winner. And can we figure that out before 95% of those generalist PMs? And we've been able to do it."
「需要 100 个人,100 位多样化的基金经理才能意识到谷歌并不是失败者,而是赢家。我们能在这 95% 的通才基金经理之前弄明白这一点吗?我们已经能够做到。」
Alex Sacerdote · Alex Sacerdote — How to Invest Through Technology Cycles

在他那里,逆向不是"押别人没听过的人",而是"在别人都以为已经涨到头的巨头上,仍然持有多数人不敢持有的信念":

"If you don't have conviction every time with NVIDIA over the last four years, it's, oh, they had a great year. Oh, my God, it's got to be a bubble... The bear cases are not totally without merit, but if you can see the whole picture and understand how these things are unfolding and gain conviction in that..."
「如果在过去四年中的每一个时刻你没有对 NVIDIA 的信心,这就会变成,‘哦,他们过得很好。’‘天哪,这一定是泡沫。’……熊市的情况并非完全没有价值,但如果你能够看到全局,并理解这些事情是如何展开的,从中获得信心……」
Alex Sacerdote · Alex Sacerdote — How to Invest Through Technology Cycles

这是对整个口号的一个反直觉补充:逆向的对象不必稀有,可以是全世界报道最多的股票——难的不是发现无人知晓的东西,而是在众目睽睽下持有一个多数人反复喊"泡沫"的判断。

都没说透的

"What's the hardest thing? It's kind of identifying if you're delusional or you have conviction... You can be lazy. If you skip the details, if you don't do the work, if you stop asking questions and are surprised by new information, those are symptoms of being delusional."
「最难的是什么?就是分辨你是妄想还是有信念……你可能会变得懒惰。如果你忽略细节,如果你不努力,如果你停止提问,并且对新信息感到惊讶,这些都是妄想的症状。」
Shardul Shah · 20VC: Index's Shardul Shah on Why Market Size is a Trap

但注意他给的其实是"勤奋度"的代理指标(有没有偷懒、会不会被新信息吓到),不是"方向对不对"的判别——所以事前区分"高信念对"和"高信念错"这道题,仍然没被真正回答。

"It's hard because we are more consensus now."
「这很难,因为我们现在更有共识了。」
Vince Hankes · Uncapped #27 Vince Hankes from Thrive Capital

他把当年 GitHub 那笔的底气归给"地理上的局外身份"——不在硅谷回音壁里,信念才不被同化:

"We were able to buy all the shares because it was so non-consensus in the valley. But part of being in New York, part of being an outsider was like, we weren't basing our conviction off of what was happening kind of echo chamber of the valley."
「我们能够买到所有的股票,因为这在硅谷是如此的非共识。但身处纽约,作为一个局外人,我们并没有根据硅谷的回音壁来建立我们的信念。」
Vince Hankes · Uncapped #27 Vince Hankes from Thrive Capital

可如果逆向的来源是"局外/小/离中心远",那"信念高于共识"就有个内建的自毁机制:一旦你赢到成为中心,你恰恰就丧失了当初制胜的那个位置。没人给出规模化之后如何保住逆向的答案——连 Thrive 自己都只是承认这是个无解的"自我强化现实"。

我的看法

(以下是判断,非访谈结论)我的把握程度:中等偏上。我认为这批人真正的共识比口号窄得多——他们只在"共识不产生超额回报"上一致,而这句话几乎是同义反复(超额回报的定义就是跑赢平均)。更有信息量的是他们的分歧:Sequoia 的数据其实在悄悄修正这个口号——它说的不是"非共识更好",而是"信念的强度比方向更重要"(强 yes 强 no 都行)。这和"逆向且正确"是两个命题,常被混为一谈。我倾向于相信 Sequoia 的版本更可证伪、也更可靠:可经营的不是"逆向",而是"在场的高信念 + 系统性去偏"。Keith 那种"三分钟就 dead sure"的叙事,我判断是被幸存者偏差打磨过的,作为方法论不可复制。

这次新进来的五个人里,我认为信息量最大的是两处对口号的"祛魅":一是 David George 给出的机制解释——非共识跑赢,未必因为人群"愚蠢",而是因为"持续高增长在数学上就无法被建模"(苹果这种全世界报道最多的公司,四年后的共识预期都能差三倍)。这把"逆向"从一种性格美德降格成一个可解释的定价漏洞,反而更让我信服。二是 Sam Altman 的止损线("有数据之前信念才成立"):它是这批人里唯一把"信念"和"固执"在操作上分开的表述——判据是"数据到没到",不是"我信不信"。但我也注意到,连他都承认存在"必须无数据运行"的长窗口,而那段窗口里,他的规则和固执依然分不开。所以真正的难题(事前分辨对错)一个都没被解决,只是被更聪明地重新描述了一遍。

还想知道什么

取材

继续读
线索
阅读线索：agent 不到 9 nines 可靠，那应该卖给企业还是消费者
线索
阅读线索：如果 coding 是 commodity，AI startup 的护城河长在哪儿
线索
阅读线索：推理成本一年跌 10×，AI app 该按什么定价
线索
摘报
线索
摘报
线索
摘报