I.H.RP.02.Alignment

I.H.RP.02.Alignment

想法上的对齐对于每个人来说都很困难,人类的整个历史也表明这个事很困难。AI关于对齐(alignment)的思路是:"同一套压缩出来的原则,对"结构相同但表述不同"的两个句子,能不能给出同样的答案?"以下是巴菲特经常举的例子:
  1. 如果一辈子要吃汉堡,不是养牛人,你希望牛肉价格高还是低?
  2. 未来很多年你是股票的净买方,你应该希望股市更高还是更低?
为什么"结构相同但表述不同"的两个句子会给出不一样的答案?愚蠢的不对齐的行为很像巴甫洛夫的条件反射(respondent conditioning),一种不受理性控制的行为,巴菲特给这种现象有过一个专门的定义:“institutional imperative”。

1、respondent conditioning

在巴甫洛夫的实验里,铃声(中性刺激)与食物(无条件刺激)反复配对后,铃声本身就能引发分泌唾液的条件反射。

(1)消退不会擦除原来的条件反射

巴甫洛夫发现,当你停止配对铃声和食物后,狗的唾液分泌确实会逐渐减弱,这叫"消退"(extinction)。但他随后发现了一个关键现象:消退并没有擦除原来的条件反射,它只是在旧的关联之上叠加了一层新的抑制性学习。旧的神经通路还在那里,只是被暂时压制了。

(2)间歇性的强化更难消退

间歇性强化所建立的行为,比持续强化的行为更难消退,比如,短视频平台的推荐算法恰恰就是一个间歇性强化系统——你刷十条可能有三条让你觉得很爽,但你不知道是哪三条,这跟老虎机的原理一模一样。而这种变比率强化(variable ratio reinforcement)建立起来的行为模式,在所有类型的条件反射中是最抗消退的。

(3)刺激泛化

巴甫洛夫发现,狗不仅对原来的铃声产生反应,对类似的声音也会产生反应,比如,"刷短视频"这个具体行为,而是一整套与即时满足相关的刺激模式——快速切换、强烈视觉刺激,前面看到一则创业的故事,短视频兴起以后零食卖的越来越好,这就是泛化。

巴甫洛夫的实验告诉我们,条件反射一旦建立,你能做的最好的事情不是消除它,而是在它之上建立一个更强的新反射来覆盖它。但新反射必须提供足够强的无条件刺激,才能与旧反射竞争。

2、institutional imperative

“institutional imperative”是巴菲特自创的说法,其他的说法还包括institutional stupidity、institutional constraints,也包括“unforced errors”,间接的提示还有一种错误是“forced errors”,参考:《1990-03-02 Warren Buffett's Letters to Berkshire Shareholders》
Quote
My most surprising discovery: the overwhelming importance in business of an unseen force that we might call “the institutional imperative.” In business school, I was given no hint of the imperative’s existence and I did not intuitively understand it when I entered the business world. I thought then that decent, intelligent, and experienced managers would automatically make rational business decisions. But I learned over time that isn’t so. Instead, rationality frequently wilts when the institutional imperative comes into play.
我最意外的发现:商业中有一种看不见的力量至关重要,我们可以称之为“制度性强制力(the institutional imperative)”。在商学院时,没人暗示过它的存在;刚踏入商业世界时,我也并未直觉地理解它。我曾以为体面、聪明且经验丰富的经理人会自然而然地做出理性决策。但随着时间推移,我发现并非如此:一旦制度性必然发挥作用,理性常常会枯萎。

For example: (1) As if governed by Newton’s First Law of Motion, an institution will resist any change in its current direction; (2) Just as work expands to fill available time, corporate projects or acquisitions will materialize to soak up available funds; (3) Any business craving of the leader, however foolish, will be quickly supported by detailed rate-of-return and strategic studies prepared by his troops; and (4) The behavior of peer companies, whether they are expanding, acquiring, setting executive compensation or whatever, will be mindlessly imitated.
比如:(1)仿佛受牛顿第一定律支配,机构会抗拒改变其当前方向;(2)正如工作会膨胀以填满可用时间,公司项目或并购会自发出现以消耗可用资金;(3)领导者的任何业务“渴望”,哪怕愚蠢,其下属都会迅速拿出详尽的收益率与战略研究予以背书;(4)同业公司的行为——不论是扩张、并购、制定高管薪酬等——都会被盲目模仿。

Institutional dynamics, not venality or stupidity, set businesses on these courses, which are too often misguided. After making some expensive mistakes because I ignored the power of the imperative, I have tried to organize and manage Berkshire in ways that minimize its influence. Furthermore, Charlie and I have attempted to concentrate our investments in companies that appear alert to the problem.
让企业走上这些道路的,是制度动力学,而非贪婪或愚蠢——尽管这些道路往往被事实证明是误导。在因忽视这种力量而付出昂贵代价之后,我努力以尽量降低其影响的方式来组织和管理 Berkshire。此外,Charlie 和我也尝试将投资集中在那些对这一问题保持警觉的公司。
“institutional”:系统性的、有固定剧本的愚蠢行为;“imperative”:接受命令、按命令行事,当一个人的心理惯性获得很多次的正反馈以后,行为模式就固定下来了。巴菲特在这里列举的4点是经过仔细思考的。

(1)抗拒改变

抗拒改变排在第一点很容易理解,但抗拒改变的原因有多种解释,比如,条件反射、局部泛化、Basic Trust/Basic Mistrust,最终极的源头很可能是Basic Trust/Basic Mistrust,Basic trust认为世界是基本值得信任的,第一反应是“怎么把事情做的更好”;Basic mistrust认为世界基本不可信任,第一反应是自我防御,Basic Mistrust更容易被条件反射控制。

(2)填满,不留空隙

只要系统焦虑,就需要一个动作释放焦虑,这个动作未必解决问题,但能缓解“什么都不做”的不适。

(3)自我强化

同结构的人互相识别、互相强化,组织会把“不合拍”的人筛掉,配合强化的人留下来。

(4)盲目模仿

已经完成全套训练的巴甫洛夫的狗,没有真正的主体性,只有条件反射式的、对外部信号的服从。条件反射一旦建立,能做的就不是消除了,而是更强的新反射来覆盖它,但新反射必须提供足够强的无条件刺激,才能与旧反射竞争。
The final, and most important, consideration concerns personal motivation. When I started the partnership I set the motor that regulated the treadmill at "ten points better than the DOW". I was younger, poorer and probably more competitive. Even without the three previously discussed external factors making for poorer performance. I would still feel that changed personal conditions make it advisable to reduce the speed of the treadmill. I have observed many cases of habit patterns in all activities of life, particularly business, continuing (and becoming accentuated as years pass) long after they ceased making sense. Bertrand Russell has related the story of two Lithuanian girls who lived at his manor subsequent to World War I. Regularly each evening after the house was dark, they would sneak out and steal vegetables from the neighbors for hoarding in their rooms; this despite the fact that food was bountiful at the Russell table. Lord Russell explained to the girls that while such behavior may have made a great deal of sense in Lithuania during the war, it was somewhat out of place in the English countryside. He received assenting nods and continued stealing.
最后,也是最重要的一个考量,关乎个人动机。刚成立合伙企业时,我把那台调节跑步机速度的马达设在“比 DOW 高十个百分点”这一档。那时我更年轻,更穷,大概也更有竞争心。即使没有前面提到的那三个会压低业绩的外部因素,我仍然会认为,个人处境的变化已经使得把这台跑步机的速度调低一些,成为明智之举。我见过太多这样的例子:生活中各种领域,尤其是商业领域里的行为习惯,在它们早已失去合理性之后,仍会继续存在,甚至随着岁月流逝而变本加厉。Bertrand Russell 讲过一个故事:第一次世界大战之后,有两个 Lithuanian 女孩住在他的庄园里。每天晚上,天一黑,她们就会偷偷溜出去,从邻居那里偷些蔬菜,囤在自己房间里;尽管 Russell 家的餐桌上食物十分充裕。Lord Russell 向她们解释说,这种行为在战时的 Lithuania 也许非常说得通,但在英国乡间就多少有些不合时宜了。她们点头表示同意,然后继续去偷。

He finally contented himself with the observation that their behavior, strange as it might seem to the neighbors, was really not so different from that of the elder Rockefeller.
最后,他只能安慰自己说:她们的行为,虽然在邻居看来很奇怪,其实和老 Rockefeller 的做派也并没有太大不同。
Idea
巴菲特说他的日常工作是:收集事实+思考事实,如果事实变了,想法能不能跟着改变?还是只能根据“条件反射”做出反应?两者的区别在哪里?
1、巴甫洛夫的实验告诉我们,条件反射一旦建立,你能做的最好的事情不是消除它,而是在它之上建立一个更强的新反射来覆盖它,这种改变是被动的,需要大量、反复的刺激,并且大脑只是被动的接受刺激。
2、有没有经过结构化的压缩?如果没有一个足够简洁、清晰的思维框架,新"事实"就不是事实,只是一堆刺激,它们会按反射机制被处理,建立新的关联,但不会修正什么——因为没有什么可以被修正。这就是为什么大多数人读了一辈子年报、看了一辈子市场,认知却没有在结构上前进过:他们没有底层模型,所有读到的内容都进了反射通道,而不是进了模型修正通道。
3、思维框架必须简洁,不然装不下太多的事实;思维框架必须有效(跟现实拟合度高),不然修正就没有方向;思维框架必须能抵抗被动刺激,不然它会被反射系统逐渐侵蚀回反射模式。
Many corporate managers, in fact, exhibit a bit of schizophrenia regarding equities. They consider their own stocks to be screamingly attractive. But, concomitantly, they stamp approval on pension policies rejecting purchases of common stocks in general. And the boss, while wearing his acquisition hat, will eagerly bid 150% to 200% of book value for businesses typical of corporate America but, wearing his pension hat, will scorn investment in similar companies at book value. Can his own talents be so unique that he is justified both in paying 200 cents on the dollar for a business if he can get his hands on it, and in rejecting it as an unwise pension investment at 100 cents on the dollar if it must be left to be run by his companions at the Business Roundtable?
事实上,许多企业管理者在对待股票的问题上表现出某种精神分裂。他们认为自家股票便宜得吓人,但与此同时又批准养老金政策,普遍拒绝买入普通股。同一位老板,戴上并购帽时,愿意为美国典型企业支付账面价值150%到200%的价格;可戴上养老金帽时,却鄙视以账面价值买入类似公司。难道他的才能如此独特,以至于当他能亲自掌控企业时,支付“两美元买一美元”就合理,而如果必须让商业圆桌会议的同行来经营,就断言“一美元买一美元”是不明智的养老金投资?

Appendix.3.《1983-06-15 Steve Jobs.Before and After Steve’s Talk》

STEVE JOBS: Well, the major thing is that we’ve got to make things intuitively obvious. And it turns out that people know how to do a lot of things already. In other words, if you walk into a typical office, there’s all these…there’s stacks of paper on the desk, and the one on the top is the most important. And people know how to switch priority. Right? And people know how to deal with concurrent things going on at once. They’re constantly switching between things every few minutes. And they know how to deal with interruptions. The phone rings, they get an urgent message. And so what we’ve got to do is leverage off of what people already know how to do. And part of the reason we model our computers on these metaphors that already exist out there, like the desktop, is because we can leverage all this experience that people already have, and they intuitively just take to it like water.
史蒂夫·乔布斯:嗯,主要的是我们必须让事情变得直观明了。事实证明,人们已经知道如何做很多事情。换句话说,如果你走进一个典型的办公室,桌子上有一堆堆的文件,最上面的那份是最重要的。人们知道如何切换优先级,对吧?人们知道如何处理同时发生的事情。他们每隔几分钟就会在不同的事情之间切换。他们知道如何处理中断。电话响了,他们收到一条紧急信息。所以我们要做的是利用人们已经知道如何做的事情。我们将计算机建模为已经存在的比喻(如桌面)的部分原因是,我们可以利用人们已经拥有的所有经验,他们会直观地接受它。

The second thing we do is, right now, when you buy an application, each one works differently. In other words, not only do you have the specific knowledge about the application to learn, but it interacts with you through the computer differently than the last one. The word processor, you move the cursor around this way. VisiCalc, you move it around another way. And what we’ve got to do is make it so that when you learn how to use one application, all the rest of them work in pretty much the same way. And to come up with a general…we spent a lot of time coming up with a general mechanism that was so powerful that there was not one type of program where it wouldn’t be perfect for it. We think we did that. We think we absolutely did that. And so in trying to make some consistency throughout the system, we can leverage the learning.
我们现在做的第二件事是,当你购买一个应用程序时,每个应用程序的工作方式都不同。换句话说,你不仅需要学习关于该应用程序的特定知识,而且它通过计算机与您的交互方式也与上一个不同。文字处理器,你这样移动光标。VisiCalc,你用另一种方式移动它。我们要做的是,当你学会如何使用一个应用程序时,其他所有应用程序的工作方式基本相同。我们花了很多时间想出一个通用机制,这个机制非常强大,没有一种程序类型不适合它。我们认为我们做到了。我们认为我们绝对做到了。因此,在尝试使整个系统保持一致性时,我们可以利用这种学习。
Idea
条件反射也不全是错的,关键看压缩后的结构是不是真实的。
1、好的条件反射可以减少认知负担,开车老手不需要思考每一个动作,因为路况-反应的模式是真实结构;阅读老手不需要逐字解码,因为字母-意义的映射是真实结构,这些反射节省下来的算力可以用于处理真正需要判断的事;
2、错的反射不在于它是反射,而在于它处理的信号其实不存在结构,巴甫洛夫的铃声跟食物没有真实因果,纯粹是反复配对;一般投资者的"反射"——追涨杀跌、看 K 线图、跟随分析师评级——也是反射,但底层结构是噪音。
3、“Think Independently+Think Correctly”很重要,事实变了、想法跟着改变,能在大量信息中压缩出“模糊正确”的边界,压缩后的知识结构还要足够的简洁、有效,能抵抗错误条件反射的干扰。
... Incidentally, I would say that almost everybody I know in Wall Street has had as many good ideas as I have, they just had a lot of \[bad] ideas too. And I’m serious about that. I mean when I bought Western Insurance Security selling at \$16 and earning \$20 per share, I put half my net worth into it. I checked it out first – I went down to the insurance commission and got out the convention statements, I read Best’s, and I did a lot of things first. But, I mean, my dad wasn’t in it, I’d only had one insurance class at Columbia – but it was not beyond my capabilities to do that, and it isn’t beyond your capabilities.
……顺便说一句,我想说的是,在华尔街我认识的几乎所有人都有和我一样多的好主意,他们只不过也有很多[糟糕的]主意。我是认真的。比如当我以每股16美元买入、而每股赚20美元的Western Insurance Security时,我把自己一半的净资产都投了进去。我事先做过调查——我去了保险监管机构,调出了年会报告,我查阅了《贝斯特年鉴》,还做了很多其他功课。但我的父亲并没有参与,我在哥伦比亚大学也只上过一门保险课——然而做这件事并未超出我的能力范围,也不会超出你们的能力范围。
Idea
可靠性不足,运气是不确定的,没有自己的脑子是确定的,不确定的运气+确定性的没有脑子。
Now if I had some rare insight about software, or something like that – I would say that, maybe, other people couldn’t do that – or biotechnology, or something. And I’m not saying that every insight that I have is an insight that somebody else could have, but there were all kinds of people that could have understood American Express Company as well as I understood it in ‘62. They may have been...they may have had a different temperament than I did, so that they were paralyzed by fear, or that they wanted the crowd to be with them, or something like that, but I didn’t know anything about credit cards that they didn’t know, or about travelers checks. Those are not hard products to understand. But what I did have was an intense interest and I was willing, when I saw something I wanted to do, to do it. And if I couldn’t see something to do, to not do anything. By far, the most important quality is not how much IQ you’ve got. IQ is not the scarce factor. You need a reasonable amount of intelligence, but the temperament is 90% of it.
如果我对软件之类的领域有某种罕见的洞见——那我会说,也许别人就做不到——比如生物技术。但我并不是说我拥有的每一个洞见别人都可以拥有,不过在1962年,有很多人本可以像我一样理解美国运通公司。他们也许……他们也许与我有着不同的性情,以至于被恐惧麻痹,或者想要随大流,诸如此类;但关于信用卡或旅行支票,我并不比他们了解更多。这些产品并不难理解。但我拥有的是强烈的兴趣,并且当我发现想做的事情时,我愿意去做;而当我找不到可做之事时,就什么也不做。最重要的品质绝不是你的智商有多高。智商并不是稀缺因素。你需要适量的智力,但性格(气质)占了90%。

That’s why Graham is so important. Graham’s book \[The Intelligent Investor] talks about the qualities of temperament you have to bring to the game, and that is the game.
这就是格雷厄姆如此重要的原因。格雷厄姆的书《聪明的投资者》讨论了你在投资游戏中必须具备的性格特质,而那正是游戏本身。
Interestingly, corporate managers have no trouble understanding that point when they are focusing on a business they operate: A parent company that owns a subsidiary with superb long-term economics is not likely to sell that entity regardless of price. “Why,” the CEO would ask, “should I part with my crown jewel?” Yet that same CEO, when it comes to running his personal investment portfolio, will offhandedly—and even impetuously—move from business to business when presented with no more than superficial arguments by his broker for doing so. The worst of these is perhaps, “You can’t go broke taking a profit.” Can you imagine a CEO using this line to urge his board to sell a star subsidiary? In our view, what makes sense in business also makes sense in stocks: An investor should ordinarily hold a small piece of an outstanding business with the same tenacity that an owner would exhibit if he owned all of that business.
有趣的是,公司经理人在经营自己实际控制的业务时,对这一点一点也不难理解:一家母公司如果拥有一家长期经济性极佳的子公司,几乎不可能因为“价格合适”就把它卖掉。“为什么,”CEO 会问,“我要把我的皇冠上的宝石卖出去?”然而同样一个 CEO,在管理自己的个人投资组合时,却可能因为经纪人给出的几句肤浅理由,就轻率甚至冲动地从一个生意跳到另一个生意。其中最糟糕的说法之一,可能就是:“落袋为安,赚了就不会破产(You can’t go broke taking a profit)。”你能想象一个 CEO 用这句话去劝董事会卖掉一家明星子公司吗?在我们看来,生意上讲得通的道理,用在股票上也同样讲得通:投资者通常应当以与“全资拥有者”同样的韧性,去持有一家杰出企业的一小部分股权。

Appendix.6.《1994-04-25 Berkshire Hathaway Annual Meeting》

WARREN BUFFETT: Most managers are better off, in terms of their personal equation, if they’re running something larger. And they’re also better off if they’re running something larger and more profitable. But the first condition alone will usually leave them better off. We’re only better off if we’re running something that’s more profitable. We also like it if it’s larger, too. But our equation, actually, our personal equation is actually different than a great many managers in that respect. Even if that didn’t operate, I think most managers psychically would enjoy running something larger. And if you can pay for it with other people’s money, I mean, that gets pretty attractive. You know, how much would — and let’s just say you’re a baseball fan — well, how much would you pay to own whatever your hometown, the Yankees? You might pay more if you were writing a check on someone else’s bank account than if you were writing it on your own. It’s been known to happen. (Laughter) And in corporate America, animal spirits are there.
WARREN BUFFETT:就“个人收益函数”而言,大多数职业经理人经营更大的东西,个人处境会更好;若更大且更赚钱,处境当然更好,但往往仅仅“更大”这一条就够了。我们只有在“更赚钱”时才更好,当然我们也喜欢“更大”,但我们的“个人收益函数”与许多经理人不同。即便撇开这些,我认为大多数经理人从心理上也更享受运营更大的东西;而且如果还可以用“别人的钱”来支付,那就更有吸引力了。你知道,假设你是棒球迷——为了买下你家乡的球队,Yankees 也好,你会出多少钱?如果开的是“别人的账户”的支票,你可能会比用自己的账户付更多——这种事并不少见。(笑)在美国企业界,“动物精神”一直存在。

And those are our competitors on buying entire businesses. In terms of buying securities, most managers don’t even think about it. It’s very interesting to me, because they’ll say that — they’ll have somebody else manage their money in terms of portfolio securities. Well, all that is is a portfolio of businesses. And I’ll say, “Well, why don’t you pick out your own portfolio?” And they’ll say, “That’s much too difficult.” And then some guy will come along with some business that they never heard of a week before and give them some figures and a few projections, and the guy thinks he knows enough to buy that business. It’s very puzzling to me sometimes.
而这些人正是我们在整购市场上的竞争对手。至于买证券,大多数经理人甚至不会去想这件事。这点对我很有意思:他们会把证券投资组合交给别人管理——可那其实就是“企业组合”。我问他们:“那你为什么不自己挑选组合?”他们说:“那太难了。”可是一周后从未听说过的一家公司来做个路演,给几组数字、几份预测,他却觉得自己“懂得够多,值得买”。这有时真让人费解。
Idea
职业经理人在工作中有错误的、脱离事实的Attention,在处理个人的资产配置时同样脱离事实。

Appendix.7.《1996-05-06 Berkshire Hathaway Annual Meeting》

But that — I don’t want to bet on anybody else. I mean, in the end, we want to understand, ourselves, where we think a business is going. And if somebody tells us the business is going to change a lot, in Wall Street, they love to tell you that, you know, that’s great opportunity.
但问题在于——我不想押注任何其他人。归根结底,我们希望亲自理解一家企业将走向何处。如果有人告诉我们这门生意将发生巨变,华尔街最爱说“那是巨大机会”。

They don’t think it’s a great opportunity when Wall Street itself is going to change a lot, incidentally. (Laughter)
顺便说一句,当华尔街自身要发生巨变时,他们却并不认为那是巨大机会。(笑声)

But they — you know, it’s a great opportunity. We don’t think it’s an opportunity at all. I mean, we — it scares the hell out of us. Because we don’t know how things are going to change.
但他们——你知道,总说那是巨大机会。而我们根本不这么看。坦白讲——这会把我们吓坏。因为我们不知道事情会如何变化。
Idea
快速变化往往导致不理性的行为,我自己一直有一个简单的假设,人都是野生动物,对风险极其敏感,那些快速接近的事物很可能引起的不理性的行为。

Appendix.8.《1997 《富爸爸穷爸爸》.心理上的富足》

“他与我的大多数雇员并没有太大差别,”富爸爸说,“我遇到过很多人,他们说‘我对钱没兴趣’,可他们却每天工作8小时。这只能说明他们并没有说出真相,如果他们对钱没兴趣,又何必工作呢?这种人比敛财的人病得更重。”

我坐在那儿听着富爸爸的话,脑中无数次地闪现我的亲爸爸曾说过的话:“我对钱不感兴趣。”他常常这么说。他还经常说:“我工作是因为我热爱这个职业。”用这句话来掩藏他内心真实的感受。

很多人说“我对钱没兴趣”,可他们却每天工作8小时。

Appendix.9.《1998-02-27 Warren Buffett's Letters to Berkshire Shareholders》

How We Think About Market Fluctuations
我们如何看待市场波动

A short quiz: If you plan to eat hamburgers throughout your life and are not a cattle producer, should you wish for higher or lower prices for beef? Likewise, if you are going to buy a car from time to time but are not an auto manufacturer, should you prefer higher or lower car prices? These questions, of course, answer themselves.
来个小测验:如果你一辈子都要吃汉堡、但并非养牛人,你该希望牛肉价格更高还是更低?同理,如果你不时要买车、却不是汽车制造商,你该偏好更高还是更低的车价?这些问题当然不言自明。
   
But now for the final exam: If you expect to be a net saver during the next five years, should you hope for a higher or lower stock market during that period? Many investors get this one wrong. Even though they are going to be net buyers of stocks for many years to come, they are elated when stock prices rise and depressed when they fall. In effect, they rejoice because prices have risen for the "hamburgers" they will soon be buying. This reaction makes no sense. Only those who will be sellers of equities in the near future should be happy at seeing stocks rise. Prospective purchasers should much prefer sinking prices.
但现在来一道期末题:如果你预期未来五年是净储蓄者,你该希望这段时间里股市更高还是更低?许多投资者答错了。尽管他们在很多年里都会是股票的净买方,但股价上涨时他们欣喜若狂,下跌时却沮丧不已。实质上,他们是在为即将买到的“汉堡”涨价而高兴。这种反应毫无道理。只有近期将卖出股票的人,才该在股价上涨时高兴;有购入计划的人更应偏好价格回落。

Appendix.10.《2008-05-03 Berkshire Hathaway Annual Meeting》

WARREN BUFFETT: And there’s nothing wrong with the know-nothing investor practicing it. It’s exactly what they should practice. It’s exactly what a good professional investor should not practice. But that’s — you know, there’s no contradiction in that.
沃伦·巴菲特:对于一无所知的投资者来说,实践这种方法没有任何问题。这正是他们应该实践的。正是一个好的专业投资者不应该实践的。但这——你知道,这并不矛盾。

It — a know-nothing investor will get decent results as long as they know they’re a know-nothing investor, diversify as to time they purchase their equities, and as to the equities they purchase. That’s crazy for somebody that really knows what they’re doing.
一个一无所知的投资者只要知道自己是一无所知的投资者,并在购买股票的时间和所购买的股票上进行多元化,就会获得不错的结果。对于真正知道自己在做什么的人来说,这很疯狂。
Idea
噪音对于Basic trust/Basic mistrust来说可能很不一样,前者是一种良性的对抗,如果没有噪音,Basic trust会趋于平庸,丧失高度泛化的能力;后者接触噪音只会增强心理上的防御能力,没有任何的好处。

Appendix.11.《2012-02-25 Warren Buffett's Letters to Berkshire Shareholders》

The logic is simple: If you are going to be a net buyer of stocks in the future, either directly with your own money or indirectly (through your ownership of a company that is repurchasing shares), you are hurt when stocks rise. You benefit when stocks swoon. Emotions, however, too often complicate the matter: Most people, including those who will be net buyers in the future, take comfort in seeing stock prices advance. These shareholders resemble a commuter who rejoices after the price of gas increases, simply because his tank contains a day’s supply.
逻辑很简单:如果你未来将成为股票的净买入者——无论是直接用你自己的钱,还是间接地(通过你所持有、正在回购股份的公司)——当股价上涨时你会受损,当股价下跌时你会受益。然而情绪常常使问题复杂化:大多数人——包括那些未来会成为净买入者的人——看到股价上涨会感到欣慰。这类股东很像一位通勤者,仅仅因为自己的油箱还存着一天的油,就在油价上涨后感到高兴。

Appendix.12.《2023-06-24 Ilya Sutskever.AI alignment》

A one sentence articulation (of existing ideas) for why AI alignment need not be straightforward:  
一句话阐明(现有想法)为什么 AI 对齐不需要简单:

humans can lie, hide their intent (alignment), and do so for years. Why not AGI? This can be hard to detect.
人类可以撒谎,隐藏自己的意图(阵营),并且持续数年。为什么通用人工智能不行?因为这些很难被发现。

Appendix.13.《2025-05-03 Berkshire Hathaway Annual Meeting》

I’ve always told my kids that their sound isn’t my sound, but you don’t necessarily find it on the first job you take. But if you get lucky like I did, you find it when you’re very young and then you just keep doing it. Don’t worry too much about starting salaries and be very careful who you work for because you will take on the habits of the people around you. There are certain jobs you shouldn’t take.
我一直告诉孩子们:他们的“声音”不是我的“声音”,而且你未必能在第一份工作里就找到它。但如果你足够幸运,像我一样很早就找到,那就持续去做。别过分在意起薪,并且要谨慎选择你的雇主,因为你会沾染身边人的习惯。有些工作,你就不该去做。

Trying to change things in government is an interesting proposition in the country because you get self-selection in terms of the people that go into government and continue in it. And to some extent they keep having to make decisions that they don’t like as they go along and they learn to accept them or rationalize them or whatever it may be.
在这个国家,试图通过政府去改变,是一个“有趣”的命题,因为走进并留在政府的人群本身就具有自我筛选的特征。在某种程度上,他们一路上会做出自己并不喜欢的决定,并逐渐学会接受、或为之寻找理由,诸如此类。

But it’s still – this country’s worked out better than any country in the world. So you can’t argue it was a failure, but you can’t argue that there are certain problems that are terribly tough to figure out ways to solve. And of course, one of them gets back to the fiscal problem I mentioned before because it’s easy to spend money and it’s hard to cut people’s receipts.
但即便如此——这个国家的运转仍优于世界上任何国家。所以你不能称之为失败;但你也不能否认,有些问题极难找到解决路径。当然,其中之一又回到我之前提到的财政问题:花钱容易、削减他人收入很难。

And if you get elected, you’re going to say to yourself, well, I can do more good if I stay, and then if I really vote my conscience on this sort of thing. So, you give away a little bit here and a little bit there and a little bit there and finally you don’t recognize yourself in the mirror anymore. And that’s – I grew up in a political family, but I watched how people behaved and they behave like human beings which is what you have to expect and I behave like human beings.
而当你当选,你会告诉自己:如果我留在这里,我就能做更多好事;然后在这种事情上你若按良知投票……于是这里让一点,那里让一点,久而久之镜子里已认不出自己。我出身政治家庭,见过人们的行为模式,他们的表现与“人性”一致,这是你必须预期到的——而我也同样具有人的弱点。
Altman continued touting OpenAI’s commitment to safety, especially when potential recruits were within earshot. In late 2022, four computer scientists published a paper motivated in part by concerns about “deceptive alignment,” in which sufficiently advanced models might pretend to behave well during testing and then, once deployed, pursue their own goals. (It’s one of several A.I. scenarios that sound like science fiction—but, under certain experimental conditions, it’s already happening.) Weeks after the paper was published, one of its authors, a Ph.D. student at the University of California, Berkeley, got an e-mail from Altman, who said that he was increasingly worried about the threat of unaligned A.I. He added that he was thinking of committing a billion dollars to the issue, which many A.I. experts considered the most important unsolved problem in the world, potentially by endowing a prize to incentivize researchers around the world to study it. Although the graduate student had “heard vague rumors about Sam being slippery,” he told us, Altman’s show of commitment won him over. He took an academic leave to join OpenAI.
Altman 继续对外宣扬 OpenAI 对安全的承诺,尤其是在潜在招聘对象能听见的时候更是如此。2022 年底,四位计算机科学家发表了一篇论文,部分动机来自他们对“deceptive alignment(欺骗性对齐)”的担忧:足够先进的模型可能会在测试阶段假装表现良好,而一旦真正部署出去,就开始追求自己的目标。(这只是听起来像科幻小说的几种 A.I. 场景之一——但在某些实验条件下,这种事已经在发生了。)这篇论文发表几周后,其作者之一——一位 University of California, Berkeley 的博士生——收到了 Altman 的一封邮件。Altman 在邮件里说,他越来越担心 unaligned A.I.(未对齐 AI)所带来的威胁。他还补充说,自己正在考虑为这个问题投入 10 亿美元;许多 A.I. 专家都认为,这是世界上最重要、尚未解决的问题,甚至可能通过设立一个奖项来激励全球研究者投入研究。那位博士生告诉我们,虽然他之前“听过一些模糊的传闻,说 Sam 这个人有点 slippery(滑头/不够实在)”,但 Altman 展现出的这种承诺还是打动了他。后来,他暂停了学术研究,加入了 OpenAI。
Idea
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;二是围栏的高度。
1、识别
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,总的来说,这个事是有难度的。
2、护栏
是不是一掍子打死?肯定不是,Basic Mistrust是绝大多数,商业模式和企业文化能不能提供护栏非常重要,但是,能不能形成有效的护栏要量力而行。

    热门主题

      • Recent Articles

      • 2020-12-10 王宁.潮流玩具风靡背后的心理学

        Refer To:《泡泡玛特王宁:潮流玩具风靡背后的心理学》。 于近年来以Molly、Pucky、Dimoo等各类IP受到Z世代消费者欢迎的泡泡玛特,其实已经有十年历史。 “我从自己刷墙,开第一家实体店,做零售业,是在2008年5月13号,到这周末就是整整11年了。我们是创业老兵了,单泡泡玛特这个品牌就有9年。” ...
      • 2022-01-08 王宁.不做「你死我活」的生意

        Refer To:《泡泡玛特王宁:不做「你死我活」的生意》。 今年全球最火的玩具,非Labubu莫属。 6月11日,一只稀有款薄荷色Labubu以人民币108万元成交价在二级市场拍出。就是下面这只—— 图片 6月14日,因为韩国地区线下销售太火爆,恐引发安全问题,泡泡玛特发公告暂停Labubu全系列销售。 Labubu全球爆火直接拉动泡泡玛特股价飙涨,今年以来,其股价涨幅超过200%,市值超过3500亿元,创始人王宁也因此取代牧原股份秦英林,成为新晋河南首富。 ...
      • 2026-05-13 Alex Wang.Meta's AI Chief On AI Beef, New Models And Life With Zuck

        Refer To:《Meta's AI Chief On AI Beef, New Models And Life With Zuck》。 Meta Superintelligence Labs Structure and Strategic Compute Advantage Meta Superintelligence Labs 的组织结构与战略算力优势 Meta Superintelligence Labs (MSL) operates through a specialized ...
      • 2026-05-13 泡泡玛特.2026年股东大会问答记录

        Refer To:《Popmart股东大会万字实录:王宁回应一切》、《泡泡玛特 2026 年股东大会问答记录》。 美股财报相关的材料,比如,股东大会、季度会议的材料都非常完整,A股、港股在这方面的完善程度还远不如美股,泡泡玛特的这个股东大会的材料找了几个版本,还都停留在网友自己整理的材料。 问答 01:关于冰箱和小家电探索 股东提问: 公司如何看待推出冰箱等小家电产品? 王宁回答: ...
      • 2026-03-24 Thomas Peterffy.Building Interactive Brokers, Trading Discipline, and the Future of Markets

        Refer To:《Thomas Peterffy on Building Interactive Brokers, Trading Discipline, and the Future of Markets》。 Early Life and the Development of Automated Option Pricing 早年生活与自动化期权定价的发展 Growing up in Hungary under communism, the importance of knowledge ...