E14：Salesforce收购Slack、DeepMind的AlphaFold突破进展、信托基金社会主义者及其他

我们都觉得那简直难以置信。

We all thought that was, like, unbelievable.

我是说，因为回报率超过50倍，但

I mean, because it was over a 50 x return, but

萨克斯，你从中得到了什么教训？

What's the lesson of you, Sachs?

只要有可能，就永远不要卖出任何资产。

Just never sell anything if you can help it.

我卖掉了'永远不卖'的原则。

I sold Never sell anything.

2014年卖掉了Facebook股票，转而买入亚马逊和特斯拉。

At Facebook in 2014 and bought Amazon and Tesla.

我认为必须能够卖出资产有两个原因：流动性和道德义务。

I think that you have to be able to sell for two reasons, liquidity and moral obligation.

是啊。

Yeah.

我是说，这有点夸张了。

I mean, that's an exaggeration.

我的意思是，人们确实需要能够卖出，但尽可能长期持有，只是不要全部卖掉。

I mean, you can never it's it's it's people need to be able to sell, but to the extent you can hold on, just don't sell everything.

你知道的，总是要...保留

You know, always, you know, keep

想想那些八十年代在苹果或微软，九十年代在亚马逊工作的人。

you know, keep the QVC mean, think about the people who were at Apple in the eighties or Microsoft in the eighties or Amazon in the nineties.

很多人因为长期持股而感到沮丧，但我认为永远保留至少20%的股份可能会带来惊人的回报。

A lot of those people got frustrated holding the shares for so long, and I think keeping at least 20% of your shares forever, you know, could be amazing.

有人告诉我他从未卖出过一股——不知道这故事是真是假。

There was somebody told me had never sold a single share of I don't know if that's a true story or not.

我...我跟你说过这事。

I I told you that.

你可不能泄密

You can't be leaking

那个信息。

that information.

好的。

Okay.

总之，不知道有泄露这回事。

Anyway, didn't know there was a leak.

哇哦。

Whoop.

哇哦。

Whoop.

哇哦。

Whoop.

哇哦。

Whoop.

好的。

Okay.

更多突发新闻。

More breaking news.

我们来看看AlphaFold吧。

Let's move to AlphaFold.

来用AlphaFold吧。

Let's AlphaFold.

我们

Let's

必须注明来源是All In播客。

move Must credit All In podcast.

天啊。

Oh my god.

无论他的股票如何，这可能是真的也可能不是。

May The or may not be true without and his shares.

知道我们该做什么吗？我们应该在里面加入哔哔声，尼克。

You know what we should do is, we should do a we should put beeps in there, Nick.

我听说从未出售过一股股票，然后我们就让大家自行反应。

I was told had never sold a share of and then we just let everybody react to it.

这样没人知道我们在谈论什么。

This way nobody knows what we are talking about.

我确实知道他一股都没卖过，只有在需要资金调用时才出售股票，这展现了非凡的坚韧与远见。

I do know that has not sold a single share and it has only sold shares of to fund capital calls, which is an incredible statement to Fortitude and Vision.

太不可思议了。

Incredible.

天啊。

Lord.

难以置信。

Incredible.

顺便说一句，这招并不总是奏效，因为他之前对别的公司也这么干过，结果就不太理想。

It's by the way by the way, it's it's not always worked out because he did the same with and those didn't go as well.

是啊。

Yeah.

我是说，听着，当你把所有鸡蛋放在一个篮子里，只投资一家公司时，你必须进行多元化。

I mean, look, you you you have to diversify when when you've got all your eggs in one basket in one company.

显然，你必须出售部分股份。

Obviously, you have to sell some shares.

但你知道吗，过去二十年我学到的一点就是——人们常问我最大的遗憾或教训是什么。

But, you know, one of the things I've just learned over the last twenty years is probably you know, people ask me what's your biggest regret or learning or whatever.

过早卖出可能是你能犯的最大错误之一。

It's just selling too early is like one of the biggest mistakes you can make.

看看PayPal的例子。

Look at PayPal.

PayPal如今是一家价值2500亿美元的公司。

PayPal is now a $250,000,000,000 company.

我们在2002年以15亿美元的价格卖掉了它。

We sold it in 2002 for 1,500,000,000.0.

当时我们还觉得赚大了，结果卖价还不到如今价值的1%，而产品本质上根本没变。

We thought that was a great deal at the time and we sold it for less than 1% of what it's worth today and the product's basically the same.

你知道，这就是复利的力量。

You know, it's just compounding.

教训就是

Is the lesson.

永远不要卖出。

Never sell.

如果是赢家，就坚持持有。

If it's a winner, ride it.

你可以

You can

配对 好的。

pair Okay.

等一下。

Hold on.

等一下。

Hold on.

我要给这件事画上句号，然后我们就去讨论AlphaFold。

I'm gonna put a nail on this coffin and then we're gonna go to AlphaFold.

沃伦·巴菲特有句名言：如果你清楚自己在做什么，最好的策略就是尽可能集中投资。

There's a great quote by Warren Buffett which is, if you know what you're doing, the best thing you can do is be as concentrated as possible.

没有人能靠第七好的想法致富。

Nobody ever got rich in their seventh best idea.

我认为这基本上概括了一切。

And I think that that basically sums it up.

但你必须有能力进行这种投资组合配置，我觉得这很难。

But you have to be in a position to have the ability to have that kind of portfolio allocation and I think that's hard.

自由鸟们，请解释一下AlphaFold。

Freebirds, explain AlphaFold, please.

好的。

Okay.

给我两分钟时间，我先解释蛋白质及其重要性，然后再讲AlphaFold。

Let's explain give me two minutes on I'll explain proteins and then the importance of proteins and then alpha fold.

要记住的数字是4、3和20。

So the numbers to remember are four, three, and 20.

有四种核酸构成了你的DNA。

There are four nucleic acids that make up your DNA.

我们在高中生物课上都学过这个。

We all learned this in high school biology.

每三个A、C、T、G的组合决定一个氨基酸。

Sets of three, a, c, t, and g combinations define an amino acid.

共有20种氨基酸，而蛋白质就是氨基酸的链式结构。

There are 20 amino acids, and a protein is a string of amino acids.

所以在你的身体里，每个细胞中都有这些细胞器，它们通过读取DNA、复制并转化为氨基酸链来制造蛋白质，这就是我们所说的蛋白质。

So in your body, in every cell, there are these organelles, they make proteins by reading the DNA, taking out a copy of it and turning it into amino acid chains and that's what we kinda call proteins.

但有趣的是，当你制造一条氨基酸链时——每个链节点可能有20种选择——它并不会保持长链形态。

But what's interesting is when you make a chain of amino acids, so there's 20 of them that you could put in each point in the chain, it doesn't come out as a long chain.

实际情况是这些氨基酸会整体折叠，形成非常特定的形状，而这个蛋白质的形状决定了它的功能。

What happens is those amino acids, the whole thing collapses and it turns into a very specific shape And the shape of that protein is what defines its function.

因此，几乎所有生命体的生物功能都是由蛋白质来执行的。

So pretty much every biological function, across all life, is, is undertaken by proteins doing something.

有些蛋白质，比如我们红细胞中的血红蛋白，会有一个非常特定的小口袋，氧气分子会嵌入这个口袋，然后将氧气从肺部输送到细胞。

Some proteins like hemoglobin in our red blood cells will have a very specific little pocket where oxygen molecules stick into the pocket, and then it moves the oxygen from your lungs to your cells.

这种蛋白质的存在相当神奇，它的形状专门设计来执行这一精确功能。

It's a pretty amazing protein to exist, and it it specifically is shaped to do that exact function.

还有其他蛋白质能够分解其他分子，打断分子键。

There are other proteins that can, for example, rip apart other molecules, break a molecular bond.

例如，有些蛋白质可以从大气中提取氮气并将其送入植物细胞，供植物生长使用。

There are other proteins, for example, that can take nitrogen out of the atmosphere and put it into plants' cells that the plants can then use to grow.

在纳米尺度上，蛋白质的应用潜力令人难以置信，我们在生活中见证这一点，每天都为之震撼和惊叹。

There's an incredible, you know, set of potential on the nanoscale of what you can do with proteins and we see that in life and we're just shocked and awed and amazed by it every day.

但要弄清楚如何制造具有特定功能的蛋白质，你必须了解这些氨基酸如何最终形成蛋白质的形状，这就是所谓的蛋白质折叠。

But in order to figure out how to create proteins that that do specific things, you have to know how do those amino acids turn into the shape that the protein ultimately takes, and that's what's called protein folding.

所以难点在于——你知道为什么这很重要吗？

And so the hard thing is and and, you know, why is this important?

这很重要，因为我们可以轻松读取DNA，从而能确定构成该蛋白质的氨基酸序列。

It's important because we can easily read DNA, and therefore we can figure out what amino acid sequence is being made to define that protein.

但我们真正不清楚的是该蛋白质的形状，以及它如何执行我们在生物学中观察到的功能。

But what we don't know really well is what is the shape of that protein, and therefore how does it undertake the function that we see it taking in biology.

反过来想，如果你在生物学中想要实现某个功能，你可以设计一种蛋白质来为你执行该功能。

And if you think about the reverse of this, the reverse of this, if you have a function you wanna undertake in biology, you can design a protein to do that function for you.

例如，与癌细胞上的特定点位结合，或者从大气中吸收碳，几乎任何你能在纳米尺度上想象到的功能，蛋白质都能被设计来实现。

For example, bind to a specific point on a cancer cell or, you know, take carbon out of the atmosphere or pretty much anything else your your mind can kind of imagine on the nanoscale proteins can be designed to do.

挑战在于如何编写DNA代码来制造执行该功能的蛋白质？

The challenge is how do you write the code which is the DNA to make the protein that does that thing?

我们不知道代码如何转化为形状，这就是所谓的蛋白质折叠问题。

Well, we don't know how the code turns into the shape and that's what folding the folding problem is.

关于折叠问题，存在一个数据集，数据集包含蛋白质的三维形状，以及定义构成该蛋白质的氨基酸序列的DNA代码，如何从氨基酸序列预测蛋白质的形状？

So the folding problem, there's a dataset and the dataset is what's the three-dimensional shape of a protein and then what's the DNA code that defines the amino acid sequence that makes that protein and how do you figure out how to predict the shape of the protein from the amino acid sequence?

这一直是个不可能解决的问题。

It has been an impossibility.

再者，如果你思考这条氨基酸链，它们每一个都带有微小的电磁空间，以及它们相互结合的方式，这非常复杂。

And, again, if you think about this chain of amino acids, they each have little, you know, electrical spaces and and the way that they bind to each other, it's very complicated.

你无法简单地通过确定性来定义它。

You can't just deterministically define it.

要知道，我们在量子尺度上尚未达到那种理解水平。

You know, we don't have that level of understanding on a quantum scale.

AlphaFold的突破在于，它现在能够通过从数十万种已通过极其困难的扫描显微镜等技术确定的结构蛋白质形状数据库中学习，预测出一段氨基酸序列最终会形成怎样的蛋白质形态。这些技术旨在微观尺度上扫描蛋白质，然后分析DNA序列并找出其中的关联。

So what AlphaFold has done is they have now been able to predict from a sequence of amino acids what the protein shape will ultimately become by learning from a database of hundreds of thousands of structural protein shapes that have been defined through really, really, really difficult, you know, scanning microscopes and other techniques to really try and scan a protein on a microscopic scale and then looking at the the DNA sequence and figuring out, okay, what's the relationship?

如今，他们预测模型的精确度已处于实际用于扫描和测量这些蛋白质的显微镜误差范围之内。

And the accuracy of their predictive model now is within the range of error of the microscopes that are you being used to actually scan and measure those proteins.

这非常了不起，因为理论上你现在可以设计出一种蛋白质，并通过编写氨基酸序列实际构建这种蛋白质，而这种蛋白质能完成你想要的任何功能。

So that's incredible because now theoretically, you could come up with a design for a protein and you could actually build that protein by writing the amino acid sequence, and that protein can do any number of things you wanna do.

这是一个困扰人类数十年的棘手难题，一直难以攻克。

And this has been a difficult problem that's been intractable by humanity, and we've been challenged by it for decades.

这项机器学习突破在短短不到三年内就得以实现——我是说，这些团队去年得分才40，今年就接近90分了。

For this machine learning breakthrough to to kind of be realized in literally less than three years, I mean, the the these guys were at a score of 40 last year and this year they're at like nearly ninety Ninety.

是的。

Which is Yeah.

太不可思议了。

Incredible.

没错。

Yeah.

所以现在，我们可以通过DNA序列预测蛋白质结构形状，这将开启全新的可能性。

And so now, you know, we can now predict what the shape will be from maybe from the DNA sequence and and this is gonna unlock this ability.

现在所有人要么会直接采用他们的模型（如果他们获得授权），要么会效仿DeepMind使用的技术路线进行学习。

Everyone's now gonna take their model if they license it or whatever they do with it, or people are gonna go learn using the same techniques that that DeepMind used.

但这至少证明了这是可行的。

But it just means that it's possible.

然后科学家们就会开始行动，他们会说：

And then scientists will go away and they'll say, you know what?

我想在微观尺度实现这个特定功能。

I wanna do this particular thing on a microscopic scale.

让我在三维空间中设计一种蛋白质来实现这个功能。

Let me design in in three-dimensional space a protein to do that thing.

好的。

Okay.

现在让我来研究如何通过编写DNA代码制造这种蛋白质，如果你能利用这个算法为你解决这个问题，那将非常简单。

Now let me go figure out how to make that protein by writing the DNA code, which is really easy if you can use this algorithm to solve that for you.

而制造蛋白质的成本实际上只需几美元甚至几美分。

And it is literally dollars and pennies to make proteins.

我们可以在计算机上编写DNA。

We can write DNA on a computer.

我们可以在48小时内收到FedEx快递送来的来自DNA打印设施打印的DNA。

We can get printed DNA sent to us in forty eight hours in a FedEx envelope for a DNA printing facility.

我们可以把它放入微生物中，让微生物在一天内为我们制造出这种蛋白质。

We can put it in a in a microbe and we can get that microbe to make the protein for us in a day.

实验室成本很低，现在任何高中生物课都能做到这一点。

The lab cost any high school biology class can do this now.

因此，通过能够根据蛋白质功能目标来确定需要编写的DNA，这将开启我们在医学和环境科学领域所能实现的无限可能。

So by being able to actually figure out what DNA to write based on the objective function of what do we want the protein to do, it's gonna unlock this universe of things we can do in medicine, in environmental science.

我们可以做到分解PET塑料这类事情。

We can do things like break apart PET plastics.

我们可以实现从大气中固氮并消除对化肥工厂的需求。

We can do things like fixing nitrogen from the atmosphere and getting rid of fertilizer plants.

我们能创造出各种新型的食品解决方案、健康解决方案和环境解决方案。

We can create all sorts of new, you know, food solutions, health solutions, environmental solutions.

有没有可能做出不含碳水化合物的披萨？

Any chance you can make a pizza that doesn't have carbohydrates?

因为我正在想的是，我们能否做出健康意面或披萨这类食物？

Because that's what I'm thinking about here is, is there a way we can make a healthy pasta or pizza or something like that?

但说正经的，你认为这项技术早期会取得哪些成果？这是二十年后才能受益的理论性胜利，还是近期一至五年内我们就能获益的重大突破？

But in all seriousness, what do you think the early wins will be out of this technology and is this a theoretical win that will benefit from in twenty years or is this a serious breakthrough that we're gonna benefit from in the near term like Both one to five are

确实如此。

true.

这是机器学习领域一项极其重要的进步。

This is an incredibly important advancement in machine learning.

但现实情况是，谷歌和DeepMind仍需投入大量时间进行完善。

But the reality is that, you know, Google will still have to spend and DeepMind will have to spend a lot of time refining it.

而且他们还将面临一些重大的伦理挑战。

And then they have some really big ethical challenges ahead of it.

如何向谁以及在何种条件下开放这项技术？

How do you expose this technology to whom and under what conditions?

这与OpenAI在GPT-3上面临的情况如出一辙。

And it's the same situation that OpenAI has with GPT three.

不过我认为，GPT-3在计算机科学挑战的规模上可能是更大的突破，因为它涉及更开放的领域。

Although a lot of people I think, you know, the scale of the computer science challenge maybe was a bigger win in GPT three because it was a much more open space.

而我认为这项技术更像是某种非常专业的专家系统。

And I think this is a much more specific sort of almost expert system in a way.

但其下游的商业影响绝对是巨大的。

But the the downstream commercial implications of this are just enormous.

所以想想看，这就是为什么你会爱上谷歌这样的公司，它们的存在意义非凡——从1999年的PageRank到2002年的CPC广告，再到2020年的AlphaFold。

And so the just think about this like, this is where like you gotta you gotta love companies like Google, the fact that they exist because from, you know, PageRank in 1999 to CPC ads in 02/1934, we have AlphaFold in 2020.

对我来说，这简直就是一个

And that to me is just that's just an

反对拆分科技巨头的论据，因为只有拥有如此资源与知识的科技公司才能斥资十亿美元收购DeepMind

argument against breaking up tech because only a tech company with this amount of resource knowledge can then go spend a billion dollars on DeepMind

我是说，Alphabet每年在所谓的'其他赌注'上烧掉40到50亿美元，人们对此颇有微词，但只要其中一项成功，就是千亿美元的回报。

I mean, Alpha Bet's burning 4 to $5,000,000,000 a year on their quote unquote other bets line And people give them a lot of shit for it, but I mean, you hit any one of these things and it's a $100,000,000,000 payday.

看看YouTube就知道了。

I mean, look at YouTube.

YouTube轻松实现了千亿美元回报，而当初收购价不过16亿美元。

YouTube's easily a $100,000,000,000 payday on a billion dollar bet, billion 6.

哦，不。

Oh, no.

那可是2500亿到5000亿美元市值的公司。

That's a that's a 250 to $500,000,000,000 company.

语义学。

Semantics.

很多人忽略了这一点，但应用语义学曾是一个价值1亿美元的赌注，而这最初就是AdSense的全部。

Lot of people miss this, but applied semantics was a $100,000,000 bet, and that's the entirety of AdSense initially.

安卓，Chrome。

Android, Chrome.

安卓。

Android.

没错。

Right.

你知道，而且

You know, and

这是DeepMind的第一个商业应用吗？

Is this is this the first commercial application of DeepMind?

因为在此之前，他们只有AlphaZero。

Because until now, you know, they've had AlphaZero.

这是

It's been

所以，有那么一段时间，很多人...让我先想一下。

So So there there was a period of time a lot of people I don't know if let let me just think about this for a second.

好的。

Alright.

因为AlphaZero确实非常出色。

Because AlphaZero AlphaZero was really good.

我想谨慎处理这个问题，但我确实...

I wanna be careful about this but I do

认为要谨慎避免透露...不。

think that careful to not disclose No.

但我确实认为，他们曾公开表示过利用DeepMind提升了广告质量。是的。

But I I do think I do think it was disclosed that they've used DeepMind to improve ads quality and Yeah.

还优化了YouTube观看体验。

To improve YouTube viewing.

正因如此，YouTube普通用户的日均使用时长翻了一倍甚至两倍。

And as a result of that, you get the number of hours per day of the average user on YouTube to double or triple.

广告收入增长了三倍。

Ad revenue goes by three x.

我记得在某一个季度，谷歌通过DeepMind实现了约150亿美元的年度增量收入。

And I think in one quarter, Google was able to generate something like an incremental 15,000,000,000 annualized revenue from DeepMinds

就在那里。

out there.

你知道DeepMind在YouTube上实际做了什么吗？

And you know what that DeepMind actually did on YouTube?

它把所有人都推向了另类右翼的Infowars和本·夏皮罗。

It sent everybody to the alt right info wars and Ben Shapiro.

恭喜啊。

Congratulations.

用人工智能引导人们的思想时，请务必谨慎。

Be careful where you send people's minds with artificial intelligence.

也许你刚才就是在主张拆分他们。

Maybe you just argued to break them up.

是啊。

Yeah.

但我我我了解到Saxon现在在谷歌各产品中的使用方式已经更加谨慎，这是我的理解。

But I I I understand Saxon is being used broadly across the products at Google in in a in a in a now now in a much more careful way is my understanding.

听着。

Look.

我是我是我是Alphabet的坚定盟友。

I'm a I'm a I'm a big ally of Alphabet.

我是他们的超级粉丝，我非常热爱

I'm a big fan and I love

你知道，他们曾经

You know, they were

而且我过去在那里工作过，和那里的人关系非常密切。

and I used to work there and I'm very close to people there.

所以

So

你知道谁曾是董事会成员，并且是那家公司DeepMind的主要支持者吗？

You know who used to be on the board and was the major backer of that company, DeepMind?

创始人基金。

Founders Fund.

创始人基金。

Founders Fund.

还有埃隆·马斯克。

Elon Musk too.

而他 是的。

And he Yeah.

恳求他们不要卖给谷歌，因为百万

Begged them to not sell to Google because million

美元的四亿退出价简直是捡了大便宜

dollar $400,000,000 exit was a steal

大概50位科学家吧，我想。

or 50 scientists, I think.

是的。

Yeah.

绝对的便宜货。

Absolute steal.

不。

No.

我是说，埃隆公开表示过，他认为DeepMind是最大的威胁——他认为AI是人类最大的威胁。

Elon I mean, Elon has publicly said that he thinks deep mind is like the greatest threat to well, thinks AI is the greatest threat to humanity.

而在研究AI的团队中，DeepMind走得最远，因此也最危险。

And of the people working on AI, deep mind is the furthest along and therefore most dangerous.

他试图阻止过。

He he tried to stop.

他直接告诉我——他后来也公开说过——他说：'我可以给你无限的资金，只要你不卖给谷歌。'

He I mean, he told me straight up, like, he's been very public about this since, that he he said, I'll give you an unlimited amount of money to not sell to Google.

他和拉里·佩奇也是朋友。

He's friends with Larry Page too.

他说，别卖。

Said, don't sell.

我希望你们保持独立。

I want you to be independent.

我希望你们继续研究这个。

I want you to keep working on this.

但他说当他看到那里的AI意识到自己是个AI时，那一刻他感觉不对劲。

But he said that when he saw the AI there become aware that it was an AI, that that was when he was like, wait a second.

不。

No.

等等。

Wait.

DeepMind的AlphaFold或AlphaZero并不具备自我意识。

Deep deep mind is not alpha alpha fold or alpha zero is not self aware.

就是这样

That's it is

它感觉自己在逐渐意识到自己的存在，明白自己是什么。

It felt it was becoming self aware that it knew what it was.

我我不确定那是不是只是埃隆，你知道，某种程度上

I I don't know if that was just Elon, you know, just sort

不

of No.

它并没有自我意识。

It's not it's self aware.

所以，我是说，我我

So, I mean, I I

看了太棒了。

watched amazing.

是的。

The yeah.

我是说，DeepMind之前发布的那些令人惊叹的东西都是游戏。

I mean, the the amazing things that DeepMind has released prior to this were games.

对吧？

Right?

他们有这个，国际象棋

They had this, chess

公开地。

Publicly.

这个名为AlphaZero的国际象棋AI，迅速成为不仅是国际象棋界的最强者——它不仅击败了世界上所有人类棋手，还击败了所有国际象棋引擎，因为计算机凭借其强大的计算能力，早就在国际象棋上超越了人类。

This chess AI called AlphaZero, which rapidly became not just the best chess, It it it not only beat every human in the world, also beat every chess engine because computers became more, you know, better at chess than humans a long time ago because of their sheer computational power.

但AlphaZero下棋风格像人类，却拥有同样的计算能力。

But, AlphaZero plays like a human, but with kind of that same computational power.

这在国际象棋引擎领域引发了一场革命。

And so that created a whole revolution in chess engines.

后来他们又用围棋这个游戏做到了同样的事。

And then they also did that with a game called Go.

他们创造了AlphaGo。

They created AlphaGo.

基本上，你能想到的每一种游戏，DeepMind都创造了一个Alpha版本，无论是人类还是电脑都无法匹敌。

And basically, every single game that you can think of, DeepMind's created, you know, an an alpha whatever alpha version that, destroys both humans and computers.

但这是他们首次公开宣布的项目，看起来最终会向公众开放，可能产生巨大的社会影响。

But this is the first thing they've publicly announced that that seems like it will be available to others eventually that could have tremendous social impact.

想象一下政府试图理解这个。

Imagine the government trying to understand this.

你能想象他们被带到参议员面前吗

Can you imagine them being brought before, like, senators

还有国会议员？

and congressmen?

这些就是新的大规模杀伤性武器。

These are these are the new weapons of mass destruction.

说实话。

I mean, let's be honest.

比如，如果别人已经掌握了AlphaFold技术——很可能确实有人掌握了但没说，要知道谷歌实际上重视透明度，所以我们才能知道这件事。

Like, if if somebody else had had AlphaFold and probably somebody does and just hasn't said, you know, I mean Google actually values transparency so that's the only reason why we know.

想象一下他们能做什么，就像弗里德伯格说的，这就像是...你知道，反过来的话就是设计一种特定蛋白质来摧毁器官之类的。

Imagine what they could do as Friedberg said, it's like, you know, the the opposite of this is basically to design a specific protein that basically, you know, destroys organs or There

生物学上最可怕的东西在我看来是一种叫朊病毒的蛋白质。

there are proteins called prions which are the scariest thing known in biology in my opinion.

朊病毒这种蛋白质会寻找相似蛋白质，并根据自身形状促使其他蛋白质发生变异，就像病毒一样。

A prion is a protein that it actually finds similar proteins and based on its shape, it gets those proteins to change and it becomes like a virus.

朊病毒相关疾病是非常悲惨的——当你的身体错误表达某种蛋白质时，这种蛋白质会促使其他蛋白质变异并自我复制扩散。

And prions actually are there's an extremely sad series of diseases that are related to prions where your body expresses a protein in the wrong way, and then that protein itself gets other proteins to change and creates copies of itself and it spreads.

这是一种既令人着迷又可怕的生物现象，但要知道，有了设计蛋白质的能力，你能做出极其可怕的事情。

It is a fascinatingly scary, biological phenomenon, but, you know, there there are extremely scary things you can do with the designer protein capability So

基本上，你可以制造出类似《普罗米修斯》里外星工程师使用的那种生物武器。

basically, you can make a bioweapon that would be similar to what the aliens and Prometheus and aliens were doing, the engineers.

甚至可能更糟糕。

Even much worse.

甚至可能更糟。

Even much worse.

这就是为什么我认为他们会花多年时间准备，回到之前的问题，这项技术多年后才能公开，因为他们足够聪明，明白自己发现了什么。我认为他们公开讨论的做法是正确的，因为这里有许多惊人的研发成果。

This is why I think they're gonna spend years before, so this is why back to that earlier question, it'll be years before this sees the light of day because they're smart enough to know what they've uncovered and I think they did the right thing by talking about it because I think there's a lot of really amazing R and D.

但现实是，谷歌在未来二三十年内将面临层层监管，毕竟你不会允许一家公司在山景城未经监管就进行铀浓缩，这相当于同样的事。

But the reality is that Google for the next twenty or thirty years will have layers and layers of oversight because, you know, it's not like you're gonna allow a company to enrich uranium in Mountain View without oversight and this is the equivalent.

我们只是在争取时间，直到政府意识到必须介入监管。

It's just we're we're buying time until governments realize that they have to do it.

但别忘了，他们训练用的蛋白质数据集是公开的。

But let's also remember the protein dataset they trained on is publicly available.

要知道，任何人都能获取这些数据。随着机器学习能力在基础设施方面追赶DeepMind，理论上...我认为未来24个月内必然会有其他人成功复现。

Know, anyone can access this and with the, you know, a catch up in machine learning capability to DeepMind infrastructure wise, theoretically you know, someone could could I think it's an inevitability that in the next twenty four months someone else will Right replicate

我认为DeepMind的贡献在于展示了特定学习技术可以突破算力限制。

now, I think what DeepMind has done is showed that specific techniques of learning can overcome limitations in compute.

但正如你所说大卫，我们正在以足够快的速度推进硅基计算，如果能投入5、10、500 petaflops的算力，即使用非常老旧的TensorFlow版本，也可能得到相同答案。

But to your point, David, like we're accelerating silicon at a fast enough speed where, you know, if like if you can throw five, ten, 5,100, 500 pet ops at a problem, you can you know, some really really old version of TensorFlow and probably get to the same answer.

是的。

Yeah.

其实我觉得他们公布了——我这就找出来，你们聊着——他们确实公布了运行这个所需的计算资源。

I actually think they published I'm I'm gonna find it while you guys are chatting, but they published what the compute needs were on running this.

我马上找到。

I'll find it in a second.

我告诉你。

I'll tell you.

因为——请继续。

Because Go ahead.

我想就整个大规模杀伤性武器话题稍微补充一个观点。

Me make one sort of slightly descending point on the whole WMD thing.

我的意思是，显然，这项技术的使用权限必须受到管控。

I mean, look, obviously, the you know, who who has access to this technology needs to be controlled.

我们必须确保它不会落入恶意行为者手中。

We we do have to make sure it doesn't get into the hands of bad actors.

但现实是，现有的许多大规模杀伤性武器已经存在很长时间，不法分子本就可以获取它们。

But the reality is there's plenty of existing WMDs out there that have been around for a long time that bad actors could get their hands on.

他们早就能在实验室里制造出极其危险的病毒。

They could already make, you know, extremely dangerous viruses in a lab.

他们可以拿天花这样高传染性的病毒，改造成像埃博拉那样高致命性的变种。

They could, you know, take something that's as contagious as smallpox and make it as deadly as Ebola or something like that.

而正是这些新技术为我们提供了对抗这些大规模杀伤性武器的手段，因为技术往往会被民主化普及。

And it's these new technologies that provide a way for us to combat those those WMDs because, you know, technology has a tendency to become democratized.

想想看，75年前美国是唯一拥有原子弹的国家。

You know, was it, seventy five years ago The US was the only country that had an atomic weapon.

现在呢？差不多有10个国家拥有核武器了。

Now, what, like, close to 10 have, you know, nuclear atomic weapons.

这类邪恶技术终究会扩散，我们确实需要这些新能力和新技术来对抗它们。

And so, you know, these these sort of nefarious technologies get democrat anyway, and we really need some of these new technologies and new abilities as a way to combat them.

就像上周我们讨论的，现在用mRNA技术两天内就能打印出疫苗。

It's like last week, we were talking about the ability now to basically print, you know, a vaccine vaccine using using this this mRNA mRNA technique within two days.

想象一下，如果有人试图使用生物武器制造新病毒，我们第二天就能打印出疫苗。

You know, you could imagine, you know, if if if someone tried to use a bioweapon, create a new virus, we could print a vaccine the next day.

这些正是我们对抗大规模杀伤性武器扩散所需的能力。

Those are the kinds of capabilities we're gonna need to fight the spread of WMD.

我认为你是对的。

I think you're right.

只是这样一来就会面临大数定律的问题。

It's just that you then have a law of large numbers problem.

意思是说，之前的问题在于有180、190甚至200个国家行为体参与其中。

Meaning, you know, the previous problem before is you had a 180, a 190, 200 state sponsored actors.

假设有百分之一的人是坏人，那就意味着两个国家会作恶。

So if you had a one percent, you know, rate of people who are just bad, you're talking about two states.

而现在当50亿人都能接触到相当于在AWS或GCP上运行实例的代码，哪怕只有1%的坏人比例，这个1%突然就变得举足轻重了。

Now when you talk about 5,000,000,000 people getting access to code that's basically running an AWS or GCP instance someplace and you have a 1%, you know, bad guy rate, all of a sudden that 1% really matters.

我认为这才是问题所在。

And I think that's the problem.

这不过是大数定律，大卫，应用于极低比例的不道德行为。

It's just the law of large numbers, David, applied to very very small rates of immorality.

但我的意思是，利用重组DNA技术，你已经可以做些非常可怕的事——比如从一个生物体提取DNA，植入另一个生物体并释放它。理论上，你本可以设计出一种在人类中传播并造成大规模死亡的病毒。

But I mean you can already do very scary things with, know, recombinant DNA, taking DNA from one organism, putting in another and sending that organism out and you know, theoretically, you could have designed a virus that would spread throughout humanity and cause a lot of death.

对吧？

Right?

我是说，这类事情很多

I mean, there's a lot of this

具备这种能力的人本应发现这点，因为运营这类实验室需要猴类或小鼠模型，你不可能悄无声息地完成。

capability People that would have found that out because, like, you can't run these labs and you're not gonna get, you know, monkey models or mouse models.

这就像留下一连串面包屑，比在GCP上运行模拟实例、再通过联邦快递寄送打印蛋白质要容易追踪得多，你同意吗？

It's like there's there's a trail of breadcrumbs there that is easier to track, wouldn't you agree, than just running an instance in a simulation on GCP, sending it someplace to then have a protein printed that comes in a FedEx envelope too.

而且你需要招募人员——就像我们在伊朗看到的，他们的首席科学家被不知哪方势力用60人意外干掉了。

And you have to hire people as we saw in Iran with, you know, their chief scientist being whacked by whoever accidentally whacked him with 60 people.

做这种事需要智力支持。

Like you need to have the brainpower to do this.

一旦拥有实施此事的人才，就能追踪这些人才。

Once you have the brainpower to do it, you can track that brainpower.

过去我认为，自从我们进行核弹试验以来，就一直有特殊飞机在全球飞行，寻找核爆特征或任何核辐射迹象，这些飞机持续飞行至今。

And we had for the last, I think it's like since we even tested nuclear bombs, we've had these special planes flying around the globe looking for the signature of nuclear detonations or any kind of nuclear fallout and we've been flying those forever.

我想

Think I

我认为生物防御将成为本世纪下半叶地球上最大的产业之一，因为数字生物学将像八十年代的软件一样成为普及工具——突然之间人人都能接触、使用并创造。

think think biodefense is going to become probably one of the biggest industries on planet earth starting in the latter half of this century because you're going to get to a point where digital biology is like a tool just like software was in the eighties where suddenly everyone could access it, everyone could use it, everyone could do stuff with it.

可怕的事物将无处不在，黑客遍地都是。

You have this ubiquity of scary shit happening, hackers everywhere.

等等。

Hold on.

Freebark 是啊。

Freebark Yeah.

关于核能我们不是也听过同样的说法吗？可我们近百年来不都安然无恙？

Didn't we have that as well in those same claims about nuclear power and we were been able to for a hundred years?

看。

Look.

核能接近

With nuclear close to

一百年。

a hundred years.

特别是核能，在法律层面。

With nuclear in particular With legal.

特别是核能，存在一个实质性问题。

Nuclear in particular, there is a material problem.

你需要获取并提纯材料才能制造恐怖事件。

You have to source and refine material to do something scary.

但生物技术不需要。

With biology, you don't.

你可以在任何地方做到这一点。

You can do this anywhere.

你可以在家里的桌面上完成，而且你知道，这正变得越来越容易、便宜、快速，你可以使用软件和打印机，还有便宜的测序仪来做这些事情。

You can do it on your desktop at home and, you know, it's getting easier, cheaper, faster and you can use software and a printer and, you know, a cheap sequencer and do stuff.

我们是否认为中国具备这样的框架？

Do we believe that China has this framework?

是的。

Yes.

中国？

China?

当然。

For sure.

对。

Yeah.

我是说，看，没错。

I mean, look, yeah.

这是

This is

你认为他们领先于DeepMind还是落后100年？

think Are they ahead of DeepMind or behind 100.

100%

100%.

100%

100%.

DeepMind公布的数据是这样的。

So here's the stats that DeepMind put out.

他们用了相当于100到200个GPU运行几周来构建这个模型。

It took them the equivalent of a 100 to 200 GPUs run over a few weeks to build this model.

我不管DeepMind在这方面有多先进——

That is I don't don't care how advanced DeepMind is at

在这一点上。

this point.

所以，如果这就是模型，那么其他人也能复制它。

That's So, you know, if that's the the model, then someone else can replicate that.

是的。

Yeah.

你知道，有人

You know, someone

这正是AlphaZero的情况，DeepMind在象棋领域验证后，现在涌现了一大批象棋AI，而且

will That's what happened to to AlphaZero, the DeepMind After chess they proved it, now there's a whole bunch of chess AIs and

我们距离这个时间点不到一年，J。

I we're within within one year, J.

Cal，很多

Cal, of a lot

的

of

初创公司将复制这种AlphaFold技术，并用于药物发现，未来12到18个月内你会看到50家基于新型蛋白质概念的初创公司获得融资。

startups replicating this AlphaFold technique and then using that to go do drug discovery and you'll see 50 startups getting funded twelve to eighteen months from now based on some novel protein idea.

我认为这

And I think that this

我们得在ISAs上试试水。

We gotta get our beaks wet on this on the ISAs.

大家都去，这太荒谬了。

Everybody go to this is getting ridiculous.

每个人都应该在thesyndicate.com/allin上创建一个。

Everybody should create a the syndicate.com/ all in.

但是...但是我...是的。

But but I but I Yeah.

我们应该创建一个蛋白质...是的。

We should create a protein Yeah.

你说吧。

Go ahead.

我们应该创建一个蛋白质折叠联合体。

We should create a we should create a protein folding syndicate.

没错。

Yeah.

绝对同意。

Absolutely.

就在这里。

Here it is.

我会的，大家

I will Everybody

去syndicate.com/ all in（全押）。

go to the syndicate.com/ all in.

如果我们这么做，我现在就开始收集邮箱。

If we do this, I'm collecting the emails now.

不行。

No.

但是我我

But I I

我会我会我会把首批赌注更多地押在防护上，而非应用本身。

I will I will I will make my first series of bets less on the application and more on the the protection.

我认为三十年后的生物防御将变得极其重要。

I think biodefense thirty years from now is gonna be so important.

好的。

Alright.

如果你想与最优秀的投资者合作

If you want to invest with the best

轻松。

ease.

会不会有一种蛋白质能阻止鲁迪·朱利安尼放屁？

Can there is there gonna be a protein that prevents Rudy Giuliani from farting?

我是说，这到底是怎么回事？

I mean, what is going on?

萨克斯，让我们听听你这位常驻右翼人士的看法——我开玩笑的。

Sachs, let's throw to you as our resident right wing I'm joking.

我开玩笑的，萨克斯。

I'm joking, Sachs.

别太往心里去。

Don't take it so personal.

我们知道你没投特朗普的票。

We know you didn't vote for Trump.

我们知道你写了塔克·卡尔森的名字投票。

We know you did a write in vote for for Tucker Carlson.

让我问问

Let me ask

你有没有

Have you

你到底有没有和塔克·卡尔森共进过晚餐？

you have you or have you not had dinner with Tucker Carlson?

这个停顿可够长的。

That's a big pause.

我摊牌。

I call.

全押。

All in.

我其实...是的。

I I've actually yeah.

我...不。

I I no.

我想我没和塔克共进过晚餐。

I don't think I've had dinner with Tucker.

但你去过

But You've

参加过有塔克·卡尔森在场的派对

been to a party with Tucker Carlson

在彼得·泰尔家。

at Peter Tales House.

没有。

No.

但让我先说说现在的情况——我感觉未来四年华盛顿政治僵局的梦想正在逐渐破灭。

But let me let me let me get to here's what's happening is well, I feel my dream of gridlock in Washington for the next four years is slipping away.

你知道，我曾把这种情况描述为某种理想场景——一个分裂的国会。

You know, the the the you know, I described this as sort of like it was kind of a dream scenario that you'd have a a divided congress.

共和党在11月赢得了参议院50个席位。

And the Republicans, you know, they won 50 they won 50 seats in the senate in November.

然后因为乔治亚州这个愚蠢的规定——不能单纯靠得票领先获胜，必须获得超过50%选票。

And then, you know, because of this stupid rule in Georgia that you have to you can't just win, you have to win by having over 50%.

共和党本可以有51席，但现在这两个空缺席位需要进行决选。

The Republicans would have had, 51, but now these there's a runoff for these two open seats.

当时共和党在民调中处于领先地位。

And the Republicans were ahead in the polls at the time.

我认为这主要是因为共和党候选人比民主党候选人更受欢迎。

They were the more I I think it's mostly based on the candidates being more popular than the Democratic candidates.

但现在由于选举闹剧，两位民主党候选人的支持率都反超了。

But now because of all these antics around the election, the Democrats have both Democratic candidates have pulled ahead.

什么？

What?

而且

And

是啊。

so it's yeah.

过去一周左右，两位候选人都变成了民主党人。

They are now both Democratic can in the last couple of last week or so.

奥斯诺夫领先珀杜约两个百分点，这令人震惊，因为珀杜之前已经击败过他。

Ausup is ahead of Purdue by about two points, which is a stunner because Purdue has already beaten him before.

他在11月的选举中击败了奥斯诺夫。

He beat him in the November election.

他认为美国

He's think America

不过。

though.

对吧？

Right?

而美国的优势在于共和党。

And the the margin America is to the republican.

没错。

Right.

但沃诺克领先洛夫勒七个百分点。

But Warnock is seven points ahead of, Loffler.

所以这个结果已经超出误差范围了。

So that one is outside the margin of error.

显然他正在击败她，而且现在看来珀杜也有危险了。

He's clearly, beating her and, it you know, and now it looks like Purdue's in danger too.

共和党只需赢得其中一场，我们就能在华盛顿形成分治政府——我认为这往往比让单一政党掌控所有权力杠杆能产生更好的结果。

The Republicans just need to win one of those in order for us to have, you know, divided government in Washington, which I think tends to produce better results than giving one party all the levers of power.

但这这真是件疯狂的事。

But but this is this is the crazy thing.

所以我认为，即便从特朗普的角度来看，这更像是一场品牌营销，试图证明他并未败选。

And so I think even from Trump's point of view, you know, I think this is a branding exercise to, you know, prove that he didn't lose.

问题在于

The problem is

是啊。

Yeah.

进展如何了？

How's that going?

嗯，这招并不奏效，而且看起来会让共和党失去参议院席位，我认为这将是他无法挽回的损失。

Well, it's not working, and it looks like it's gonna cost Republicans to the senate, which is something that I think he won't come back from.

如果我是总统的顾问，我会建议他放弃这些法律挑战。

So if I was advising the president, I would tell him to drop this, you know, these these legal challenges.

他们解雇了法律团队中的西德尼·鲍威尔，我认为这是个明智的决定，因为她那些疯狂离谱的指控。

They fired Sidney Powell from the legal team, which I think was a good decision because of her crazy wild allegations.

但现在，鲁迪又在外面干着同样的事。

But now, Rudy's out there doing the same stuff.

朱迪·朱利安尼。

Judy Giuliani.

鲁迪。

Rudy.

哦，不。

Oh, no.

朱迪·朱利安尼。

Judy Giuliani.

我给你的回答是他们连一个证人都懒得询问。

The answer that I gave you is they didn't bother to interview a single witness.

朱迪·朱利安尼。

Judy Giuliani.

他当时正在调整。

He was adjusting.

他并没有唤醒那个精灵。

He wasn't waking up the genie.

他当时正在把裤子塞好，躺在卧室里，和一个俄罗斯未成年人喝酒。

He was he was tucking his pants in, laying down in the bedroom, having a drink with a Russian underage.

我是说，鲁迪刚刚在《波拉特》电影里客串完。

I mean, Rudy is just I mean, fresh off his his guest starring appearance in the Borat movies.

我是说，这

I mean, it's

就像未署名那样。

like Uncredited.

他是未署名的。

He's Uncredited.

是啊。

Yeah.

他在这些听证会上的表现，简直像是在重演我表弟维尼的故事。

It's like he's reenacting my cousin Vinny or something in these in these hearings.

没错。

Yeah.

共和党人还要因为多少事情失去参议院席位？

How many things are gonna cost Republicans the senate?

是啊。

Yeah.

我觉得这太不可思议了。

I think they're it's so amazing.

就像民主党现在基本上稳操胜券，因为极有可能他们会赢得这两个参议院席位，那简直太难以置信了。

Like, the Democrats basically have now have a stone cold free roll because there is a very good chance that these guys are going to win these two senate seats, which would just be absolutely incredible.

简直难以置信。

Absolutely incredible.

所以共和党人肯定也意识到了这点。

So so Republicans are definitely figuring this out.

里奇·劳里写了篇精彩专栏。

Rich Lowry had a great column.

他是《国家评论》的编辑。

He's a editor of National Review.

他属于右翼，但我认为他的政治分析非常客观公正。

He's on the right, but I think provides very good objective political analysis.

他在最新专栏中为乔治亚州的情况敲响了警钟。

He's sound his last column sounded the alarm bells about what was happening in Georgia.

听着，如果乔治亚州决选的关键在于这些闹剧行为。

And look, if the if the issue in the Georgia runoffs is are these antics.

如果是鲁迪和西德尼·鲍威尔在搅局，共和党就会失去这个席位。

If it's Rudy and Sidney Powell, the public is gonna lose the seat.

两个席位都会丢。

The both seats.

如果选举议题是民主党是否应该为所欲为地强行通过任何法案，那共和党就能赢。

If the question of the election is whether democrats should have a free hand to ram through whatever legislation they want, then I think the republicans win.

所以问题在于：这次选举的核心议题究竟是什么？

So the question is, what is that election gonna be about?

鲁迪在台上发表这些疯狂指控的时间越长，对共和党就越不利。

And the longer that Rudy stays on the stage making these crazy wild allegations, the worse it gets for Republicans.

乔治亚州的决选是什么时候？

When is the Georgia runoff election?

我想是1月5号或10号。

I think it's January 5 or tenth.

就是这两个日期之一。

It's it's one of those two.

应该是1月5号。

January 5, I think.

天哪。

My gosh.

在座的各位如果有灰白头发——我知道电话那头有三位男士是灰白头发，弗里德伯格，你...你可能...我好像已经回答过这个问题了。

Would anybody here, if they had gray hair, and I know we we have three gentlemen on the call who have gray hair, Friedberg, somehow you're you're you're you're you might be maybe I've answered this question already.

他头上可没有（灰白头发）。

Not on his head.

如果你们是79岁或87岁的鲁迪·朱利安尼，你们会染头发吗？

Not Would on his any of you dye your hair if you were a seventy nine, eighty seven year old Rudy Giuliani?

不。

No.

我的头发已经白了，而且越来越白。

My hair is white and getting whiter.

顺其自然吧。

It's like it is what it is.

这是时光流逝的痕迹。

It's a passage of time.

我会选择坦然接受。

I would just move on.

大卫，你会考虑把那些灰白头发染掉吗？

David, would you consider just taking that gray right out of your hair?

因为你现在已经是'银狐'了。

Because you have the you're the silver fox as it stands.

你会考虑这么做吗？

Would you ever consider it?

能看出这个决定

Can see the decision

我已经做出的决定。

I've made.

你你能看出我做出的决定。

You you can see the decision I've made.

但听着，那不是我的意思，我是说，确实。

But look, that's not I mean, it's yeah.

我是说，那简直就像一场崩溃。

I mean, that was just like a meltdown.

我是说

I mean

字面意义上的崩溃。

A a literal meltdown here.

字面意义上的崩溃。

A literal meltdown.

而且

And

这简直就像是那个时刻的完美写照。

It was so, like, representative of the moment.

国会共和党人想用钩子把鲁迪拽下台。

The the congressional Republicans wanna use a hook to get Rudy off the stage.

我是说，他们迫不及待想让他下台。

I mean, they cannot wait to get him off the stage.

然后站起来

And stand

直接告诉特朗普终止这一切。

up and just tell Trump to shut this down.

我觉得...我觉得快来了。

I think I think it's coming.

快来了。

It's coming.

所以

So

是啊。

Yeah.

哦，哇。

Oh, wow.

所以共和党人在特朗普任期还剩四周时站出来反对他。

So the republicans are stand up to Trump with four weeks left in office.

哇。

Wow.

真是勇气可嘉啊。

What a profile in courage.

弗里德伯格，你在染发吗？

Friedberg, are you dyeing your hair?

你是天生灰发，还是真的在染发，或者还没到长灰发的年纪？

Do you have gray hair or are you actually dying or have you not hit your gray yet?

现在就告诉我们。

Just tell us right now.

这是个紧急情况。

It's an emergency pot.

不。

No.

这是事实。

That's the truth.

大卫，你在染发吗？

David, are you dyeing your hair?

是还是不是？

Yes or no?

这是

This is

它。

it.

再过一会儿就好了，宝贝。

It's an over next a while, baby.

好吧。

Alright.

方法有很多。

There's so many ways.

有人对特朗普是否被允许预先赦免自己和他的子女有意见吗？

Does anybody have an opinion on whether Trump is allowed to pre pardon himself and his kids?

这正是我要说的。

It's exactly where I was going.

是啊。

Yeah.

萨克斯。

Saxe.

嗯

Well

是的。

Yeah.

萨克斯。

Saxe.

我们的右翼

Our right wing

他绝对可以赦免他的孩子们。

He can definitely pardon his kids.

我不确定是否存在一个公开的法律问题，关于

I don't know that there's an open legal question whether

他

he

能预先赦免吗？

can Can pre pardon?

在这个定义中，预先赦免是什么意思？

What does it mean pre pardon in this definition?

你是说我不能。

You mean I can't.

你只能对已犯下的罪行进行赦免。

You can only pardon for a committed crime.

对吧？

Right?

我是说，传统上是不行的。

I mean, that's traditionally No.

用一揽子方式。

With a blanket.

对吧？

Right?

嗯，你可以赦免过去发生的事情。

Well, you can pardon for things that happened in the past.

我认为你不能赦免未来发生的事情。

I don't think you can pardon for things that happened in the future.