#416 坚持不懈创造AGI的传教士：德米斯·哈萨比斯

他会说，你会相信我接下来要说的话，然后人们真的就相信了。

He would kinda be like, you will believe the things I'm going to say, and then people did believe them.

所以他最初的构想是创办一家名为Elixir的游戏公司。

And so his first idea is to start this gaming company called Elixir.

就像他之前说的，游戏是通往人工智能的路径。

Like he said before, games versus path into AI.

他从这时起，后来也一直与风险投资家们多有摩擦。

He has all kinds of trouble now and later with venture capitalists.

当时他们表示愿意给他一些资金，但你必须放弃公司超过一半的股权。

At this point, they said they'll give him some money, but you have to give up more than half the equity of the company.

我之所以向你们提到这一点，是因为我认为这是他个性中最突出的方面之一。

And the reason I bring this out to you is because I think this is one of the most pronounced aspects of his personality.

他说，如果有一件事是Demis最厌恶的，那就是被任何人控制。

He says if there's one thing that Demis hated, it was to be controlled by anyone.

这就是他所说的。

This is what he said.

他们想用钱换取我们的灵魂。

They wanted our souls in exchange for the money.

这再次非常引人入胜。

Again, this is very fascinating.

我们稍后再谈这个。

We'll get to this later.

但德米斯把DeepMind卖给谷歌的一个原因，是为了避免陷入我認為他所說的、不斷向投資者籌資的「倉鼠輪」。

But one reason that Demis sells DeepMind to Google was to avoid having to be what I think he calls it, like, the hamster wheel of raising money from investors.

他正確地將這視為對其使命的巨大干擾。

He rightly saw that as just a giant distraction to his mission.

這引出了他一生和這本書中反覆出現的另一個主題：德米斯極其務實。

And that leads us to another reoccurring theme throughout his life and throughout the book is just Demis is insanely practical.

他還會從每一次實驗和經歷中汲取教訓。

He also is gonna learn from every single experiment, experience that he has.

因此，既然德米斯談到了創辦公司，他的抱負就一直是打造強大的AI，而不僅僅是設計電子遊戲。

And so since Demis talked about starting a company, his ambition had been to build a powerful AI, not just to design video games.

在创立Elixir时，他平衡了自己的雄心与务实的一面。

In founding Elixir, he was balancing his ambition against his practical side.

他最终的梦想是打造一个人工智能领域的曼哈顿计划。

His ultimate dream was creating a Manhattan project for artificial intelligence.

这个比喻，这个想法，嘿。

That metaphor, that idea, hey.

我们要打造一个人工智能领域的曼哈顿计划。

We're gonna create a Manhattan project for artificial intelligence.

这个说法在整本书中反复出现。

That's used constantly throughout the book.

就在这个过程中，有一句话特别精彩，因为我一直想强调，德米斯讨厌失败，他极度好胜。

And in the middle of this, there's this just a phenomenal line because I keep trying to hound the fact that Demis hates losing, that he is hypercompetitive.

他是这样描述失败带给他的感受的。

This is what he this is how he describes what losing feels like to him.

这感觉就像我的灵魂在燃烧。

It's like my soul is on fire.

因此，德米斯学会更务实的一种方式是通过犯错。

And so one way that Demis learns to be more practical is by making a mistake.

他有一个想法，想做一个叫《共和国》的游戏。

He has this he has this idea for this game called Republic.

《共和国》是一个过于雄心勃勃、基本上在技术上不可能实现的游戏创意。

Republic was this overly ambitious, basically techno technically impossible idea of a game.

这将导致他第一家公司的失败。

It's gonna lead to the failure of his first company.

在这里，他提出了一个非常精彩的观点。

And there's just this really great point that he makes here.

在接下来的几年里，《共和国》的发布日期一再推迟。

Over the next couple of years, Republic's release date was pushed back repeatedly.

作为愿景的守护者，德米斯一直抵制妥协，花了很长时间才意识到自己魅力所造成的陷阱。

As the keeper of the vision, Demis fought a rearguard action against compromise, and it took time for him to recognize the trap that his own charisma created.

谁能想到，你竟然会激励人们过度呢？

Who would have thought that you can actually inspire people too much, he said.

你可以，因为你可能会达到稀释团队的程度，而他们也会反过来稀释你。

Well, you can because you can get to the point where you're diluting your team, and then they are diluting you also.

我之所以做出这样的判断，是因为工程师们告诉我这是可能的，但他们之所以这么说，只是因为我过度激励了他们，德米斯说。

It's like I'm making this judgment that this is possible because the engineers are telling me it's possible, but they're only telling me it's possible because I've overinspired them, Demis said.

事实上，我们都没有获得真实的反馈。

So in fact, none of us were getting real feedback.

他的联合创始人谈到了如何与德米斯沟通、辩论并真正说服他。

His cofounder talks about on how to communicate and debate and really just persuade Demis.

你必须把对话推进到他越来越激动、越来越坚定地捍卫自己立场的地步。

You had to push the conversation to the point where he got more and more intense and defended his positions more and more strongly.

他越强硬，你就越接近成功。

The stronger he got, the closer you were.

然后，他最终可能会沉默下来。

Then eventually, he might go quiet.

那就是他吸收了信息的时候。

That's when he absorbed the message.

在这之后，他在思考接下来该做什么。

And so after this, he's thinking about what to do.

这正是他创立DeepMind之前的事，作者出色地描述了他是如何从一切经历中学习，以及这些经验是如何相互关联的。

This is right before he founds DeepMind, and the author does a great job of describing, again, just how he's learning from everything and how all these experiences fit together.

我惊叹于德米斯的经历和想法竟能如此完美地融合在一起。

I marveled how Demis' experience and ideas appear to slot together.

他对物理学的好奇促使他投身人工智能，因为人工智能是解锁科学的终极工具。

His curiosity about physics has spurred him to work on AI, the ultimate tool to unlock science.

他对人工智能的好奇引导他研究人脑——智能存在的实证。

His curiosity about AI had led him to investigate the human brain, the existence proof for intelligence.

他在模拟和电子游戏方面的工作，呼应了他对心智模拟的研究，以及《哥德尔、埃舍尔、巴赫》一书和神经科学对他的影响，这些都推动德米斯得出了相同的结论：信息是现实的基本单位。

His work on simulations and video games echoed his research on simulations in the mind and the influences of Immanuel Kant, of that book, Godel, Escher, and Bach, and neuroscience had pushed Demis towards the same bottom line, that information was the fundamental unit of reality.

一台未来的计算机，一个强大的人工智能，可能是无限的、无穷的。

A futuristic computer, a powerful AI might be limitless, infinite.

因此，这本书中出现了各种各样的有趣人物。

And so there is all kinds of interesting characters in this book.

我主要会聚焦在德米斯身上。

I'm gonna focus mainly on Demis.

考虑到人工智能在我们世界中变得越来越重要，我强烈推荐阅读这本书。

Considering how important AI is becoming in our world, I would highly recommend reading the book.

书中有很多关于他们如何实际构建这项技术的精彩故事，我认为你值得一读。

There is a lot of fascinating stories about how they actually built the technology that are in the book that I think you should read.

德米斯不断与其他人互动，他的故事交织在这些杰出的创业者和知名投资者之间。

And then Demis is constantly interacting with and his stories, you know, weaving in and out of all these other phenomenal entrepreneurs and well known investors.

第一个是彼得·蒂尔。

The first one is Peter Thiel.

德米斯需要资金来支持DeepMind，于是他想办法向彼得·蒂尔推销。

Demis needs money for DeepMind, so he finds a way to pitch Peter Thiel.

他们在一场会议上举办的派对上，德米斯没有像其他人那样在喧闹的派对中重复讲创业故事，而是用国际象棋吸引了蒂尔。

They're at this party at a conference that says instead of pitching Peter Thiel with yet another startup story and doing so in the middle of a crowded party, Demis hooked Thiel with chess.

于是他开始和彼得聊国际象棋，因为他知道彼得对国际象棋着迷。

So he starts talking about chess with t with Peter because he knew that Peter was obsessed with chess.

蒂尔随后邀请德米斯和他的联合创始人第二天到他家，详细解释他们雄心勃勃的事业。

Thiel then invites Demis and his cofounder over to his house the next day to explain their ambitious venture.

当他们这样做时，蒂尔开始觉得这个项目在科学上可能是A+，但在商业模式上可能是F，但他还有进一步的想法。

When they do so, says Thiel began to think this project was an a plus on the science and maybe an f on the business model, but he had also had a further thought.

德米斯是真正企业家的极端案例——他不是那种一开始就想通过创业致富、然后四处寻找一个看似合理的点子的逐利者，而是一名传教士，他感到必须解决某个特定的挑战，于是创办公司作为实现这一目标的方式。

Demis was an extreme case of an authentic entrepreneur, not a mercenary who starts with a desire to get rich from a startup then casts around for a plausible idea, but rather a missionary who feels compelled to work on a particular challenge then starts a company as a way of tackling it.

传教士的好处在于，他们从不放弃。

The good thing about missionaries is that they never quit.

即使必须昼夜不停地工作、不给自己发薪水，他们也会持续痴迷于这个问题。

Even if they have to work around the clock and pay themselves nothing, they will keep obsessing about the problem.

我总是说，人们并不是抽象意义上的创业者，但也许每个人心中都藏着一家伟大的公司，”彼得说。

I always say that people aren't really entrepreneurs in the abstract, but there's maybe one great company that somebody has in them, Peter said.

建造这家公司的使命，正是德米斯的宿命。

It was Demis' destiny to build this one.

因此，他正在向彼得·蒂尔筹款。

And so he's raising money from Peter Thiel.

他们制定了这份关于DeepMind的商业计划和演示文稿。

They build this business plan, this deck of DeepMind.

我想从这份商业计划和演示文稿中挑出几件我觉得有趣的事情。

There's just a couple things I wanna pull out from this business plan, from this deck that I thought was interesting.

在演示文稿中，引用了比尔·盖茨的一句话。

In the deck, it has a quote from Bill Gates.

在当时，成立一家人工智能公司的想法听起来很荒谬。

Remember the idea for an AI company is ridiculous at this time.

这是那句话。

This is the quote.

如果你在人工智能领域取得突破，让机器能够学习，那将价值十个微软。

If you invent a breakthrough in artificial intelligence so machines can learn, that would be worth 10 Microsofts.

接着，它阐述了他们想要实现的目标。

And then it lays out what they're trying to do.

正如商业计划所解释的，人类大脑的存储容量有限，人类的寿命也有限。

As the business plan explained, the human brain had limited storage capacity, and humans had limited lifespans.

将人类聚集在一起会导致收益递减，因为大型组织反应迟缓。

Grouping humans together resulted in diminishing returns because big organizations are sluggish.

总之，社会最紧迫挑战的复杂性超出了人类能力的范围。

In sum, the intricacy of society's most pressing challenges lay beyond the reach of human capabilities.

通用人工智能是解决这一问题的办法。

AGI is the solution to this problem.

或许更大胆的是，DeepMind声称，其极其雄心勃勃的AI构想反而更有可能取得进展。

Perhaps more audaciously, DeepMind asserted that its ultra ambitious conception of AI made progress more likely.

其他AI研究试图通过专注于狭窄任务来提高成功率，例如训练系统识别图像。

Other AI research sought to maximize the chances of success by focusing on narrow tasks training a system to recognize images, for example.

相比之下，DeepMind的目标是构建智能体，而不仅仅是系统。

In contrast, DeepMind was out to build agents, not merely systems.

区别在于，智能体将更具通用性和主动性。

The difference was that agents would be more general and proactive.

智能体不是由人类设计来掌握单一有限任务，而是会广泛而自主地学习，在与环境互动的过程中掌握多种问题。

Rather than being engineered by humans to master a single finite task, agents would learn broadly and autonomously, mastering a wide range of problems as they interacted with their environment.

复杂性的跃升是巨大的。

The jump in complexity was vast.

DeepMind 并非只想构建数字世界的房屋，而是立志建造一座城市。

Rather than building the digital equivalent of a house, DeepMind aspired to build a city.

所以，德米斯和他的三位联合创始人——德米斯、肖恩·莱格和穆斯塔法·苏莱曼。

And so Demis and one of his cofounders his three cofounders, it's Demis, Shane Legg, and Mustafa Suleyman.

因此，德米斯和穆斯塔法正在四处筹款。

And so Demis and Mustafa are out trying to raise money.

这本书里有太多离奇的故事了。

There's just so many bizarre stories in the book.

我听过太多创始人讲述的投资者噩梦，但大多数都没被公开提及，真可惜。

I've heard so many investor horror stories from founders, most of which are not repeated publicly, unfortunately.

只能说，外面确实有很多奇怪又令人不适的人。

Let's just say there's just a lot of creeps and weirdos out there.

所以当我读到这一部分时，我的笔记非常简单。

So when I got to this section, my note was very simple.

2010年9月，德米斯和穆斯塔法出现在一个非常奇特的投资委员会面前。

WTF says in September 2010, Demis and Mustafa appeared before a strange kind of investment committee.

大卫·加蒙表示愿意投入资金，但他只有一个条件：DeepMind必须按他的方式行事。

David Gammon declared himself ready to commit capital, but he'd only go forward if DeepMind did things his way.

寻求他支持的创业者必须前往他家，向加蒙、他的妻子和三个十几岁的儿子进行路演。

Entrepreneurs seeking his support required to visit his home and pitch to Gammon, his wife, and his three teenage sons.

每个家庭成员对是否投资都有平等的投票权。

Each family member would get an equal say on whether to invest.

我对德米斯说，如果你不能向我年幼的儿子解释清楚，你就别想得到他的投票。

I said to Demis, if you can't explain this to my younger son, you're not going to get his vote.

DeepMind商业计划的宏大科学愿景，与受邀和一个中学生聊天之间，存在着令人痛苦的巨大鸿沟。

There was a painfully large gap between the grand science of the DeepMind business plan and an invitation to chat with a middle schooler.

这简直太离谱了。

That is just flat out bizarre.

最终，德米斯从彼得·蒂尔那里筹集到了资金。

So eventually, Demis raises money from Peter Thiel.

这里有一些有趣的背景信息，我想你会感兴趣。

There's some interesting background here that think you'd be interested in.

一般来说，彼得·蒂尔认为担任董事会成员并不是他合伙人时间的明智用途。

As a general matter, Peter Thiel doubted that going on boards was a good use of his partner's time.

初创公司应该任其自生自灭。

Startups should be left to sink or swim.

他常说的是，风险投资的艺术在于支持反主流的想法，而不是辅导创业者。

The art of venture capital, he liked to say, was to back contrarian ideas, not coach company founders.

蒂尔持有一种不同寻常的观点，认为应避免集体决策。

Thiel had taken the unusual position that collective decision making should be avoided.

在他看来，如果投资决策基于投票，那么创始基金的投资组合将全是平庸的初创公司，因为没人会反对它们。

The way he saw things, if investments were chosen based on voting, the Founders Fund portfolio would consist of middle of the road startups to which nobody objected.

鉴于风险投资的所有利润都来自少数难以置信的突破性项目，这种共识型投资组合只会带来平庸的回报。

Given that all the profits in venture come from a few improbable moonshots, this sort of consensus portfolio would deliver mediocre performance.

创始基金汇出了230万美元，这让你明白为他的想法、为德米斯的想法融资有多困难。

Founders Fund wired 2,300,000.0 this gives you an idea of just how hard it was to raise money for his idea, for Demis' idea.

创始人基金向DeepMind转账了230万美元，并获得了公司不到一半的股权。

Founders Fund wired 2,300,000.0 to DeepMind, and they assumed ownership of a bit less than half the company.

当时没有其他资金可用。

There was no other capital available.

这是2010年12月。

This is December 2010.

彼得·蒂尔在这个故事中反复出现。

Peter Thiel reappears over and over again in the story.

他其实有一句我最喜爱的名言之一。

He actually has one of my all time favorite quotes.

这句话出自他的书《从零到一》。

It's in his book, Zero to One.

书中写道：我注意到的最强大的模式是，成功的人总能在意想不到的地方发现价值，他们通过从第一性原理而非固定公式出发思考商业来做到这一点。

It says the single most powerful pattern I have noticed is that successful people find value in unexpected places, and they do this by thinking about business from first principles instead of formulas.

这正是我的合伙人AppLovin在其新的广告平台Axon上所做的事情。

That is exactly what my partner AppLovin has done with their new advertising platform, Axon.

Axon是这一代最强大的广告平台。

Axon is the most powerful advertising platform in a generation.

Axon能让你获得完全专注的注意力。

Axon allows you to capture undivided attention.

Axon广告是全屏视频，平均观看时长达到三十五秒，留存率远超其他广告平台。

Axon ads are full screen videos that are watched for an average of thirty five seconds, retention that blows other ad platforms out of the water.

你可以在几分钟内上线Axon。

You can launch on Axon in minutes.

你设定目标，Axon帮你实现。

You set the goal and Axon achieves it.

无需复杂设置，也不需要专业知识，Axon能快速扩展。

No complex setup, no expertise needed, and Axon scales quickly.

它们能让你的广告触达超过十亿潜在客户。

They can put your ads in front of over a billion potential customers.

其他企业已经看到即时效果，日广告支出高达数十万美元，收入增长数百万，而大多数广告主甚至还没考虑过这个渠道。

Other businesses have seen immediate results scale to hundreds of thousands of dollars of spend per day and increase their revenue by millions, and most advertisers aren't even thinking about this channel yet.

不到1%的广告主能使用Axon。

Less than 1% of advertisers have access to Axon.

所以如果你想快速开始，可以访问 axon.ai/founders。

So you wanna get started quickly, and you can do that by going to axon.ai/founders.

就是 axon.ai/founders。

That is axon.ai/founders.

在我们继续本集之前，我得向你们介绍一下Vanta。

And before we get back into this episode, I need to tell you about Vanta.

Vanta。

Vanta.

Vanta。

Vanta.

Vanta能帮助你的公司证明其安全性，从而吸引更多客户使用你的产品或服务。

Vanta helps your company prove your secure so more customers will use your product or service.

你可以把Vanta看作是你随时在线的、由AI驱动的安全专家，它会随着你的业务一起成长。

You can think of Vanta as your always on AI powered security expert who scales with you.

你的业务增长得越多，安全需求就越复杂，而这种复杂性会演变成混乱。

The more your business grows, the more complex your security needs get, and that complexity turns into chaos.

Vanta 能够掌控这种混乱。

Vanta tames that chaos.

Vanta 自动化合规流程，持续监控你的控制措施，并为你提供合规与风险的单一信息来源。

Vanta automates compliance, continuously monitors your controls, and gives you a single source of truth for compliance and risk.

无论你是快速增长的初创公司还是大型企业，Vanta 都能轻松融入你现有的工作流程。

So whether you're a fast growing startup or an enterprise company, Vanta fits easily into your existing workflows.

这让你能够持续发展一家让客户信赖的公司。

This allows you to keep growing a company that your customers can trust.

许多公司除非你获得认证，否则不会签署合同，而这正导致你错失销售机会。

Many companies won't sign contracts unless you're certified, and this is causing you to lose out on sales.

这就是为什么平均而言，Vanta 客户在成为客户后报告的投资回报率达到 526%。

That is why the average Vanta customer reports a 526% return on investment after becoming a Vanta customer.

Vanta 将帮助你更快、更轻松地赢得信任、促成交易并保持安全。

Vanta will help you win trust, close deals, and stay secure faster and with less effort.

前往 vanta.com/founders，即可享受一千美元优惠。

Go to vanta.com/founders, and you'll get a thousand dollars off.

这就是 vanta.com/founders。

That is vanta.com/founders.

没人相信 DeepMind 股票的未来价值。

Nobody believed in the future value of DeepMind's stock.

该领域的权威人士认为，一支没有收入的研究团队只会短暂地做出一些有趣的成果，然后就会倒闭。

Prestigious figures in the field assumed that a research team with no revenues would do interesting signs for a couple years, and then they would go out of business.

而德米斯却完全不受影响。

And Demis is just completely undeterred.

他们招募 DeepMind 最初的一些员工的方式非常引人入胜。

This is very fascinating how they recruit some of the first employees to DeepMind.

这是一项科学挑战。

This is a scientific challenge.

作为科学类初创公司，你需要那些敢于探索未知领域的远见者。

As scientific startups, you need blue sky thinkers who wander into the unknown.

我们只想要真正坚定的信仰者。

We only wanted hardcore believers.

我们会去参加这些会议，告诉人们：我们正在创办一家通用人工智能公司。

We would go to these conferences and tell people, we're starting an AGI company.

80%的人会对我们翻白眼，真的翻白眼，然后转身离开。

80% of the people would roll their eyes at us, literally roll their eyes at us and turn around and walk away.

我们觉得这是一种非常高效的方式，来发现我们应该与谁交谈。

We figured that this was a very efficient way to discover who we should be talking to.

他们对求职者也有一个非常有趣的说辞。

They also had a really interesting pitch to recruits.

就是这个。

It was this.

学术界的氛围可能既枯燥谨慎，又令人恐惧地充满竞争。

The culture of academia could be both boringly cautious and terrifyingly competitive.

枯燥是因为它只追求渐进式的进步。

Boring because it pursued incremental advances.

令人恐惧的是，科学家们为了率先发表而互相倾轧。

Terrifying because scientists cut each other's throats to be the first to publish.

在DeepMind，我们承诺提供截然不同的体验：追求重大突破的激动人心之旅，以及几乎不存在竞争对手的环境。

At DeepMind, we're promising the opposite experience, the thrilling pursuit of the big leap and the near absence of rivals.

我们将去做那些无人认为可能实现的事情，因为根本没有竞争。

We are going to do stuff where there's no competition because no one thinks it's possible.

那些在没有证据之前就相信的人是有福的。

Blessed are those who believed before there was any evidence.

因此，他们做的第一件事之一——我认为这非常重要。

And so one of the first things they do I think this is really important.

传记的力量就在于此，因为你能看到他们思想和行为的演变过程。

The fact that this this to me is really the power of biography because you see that you see the evolution of their thinking and their behavior over time.

他正在学习。

He is learning.

他正在从自己前一家公司的失败实验中吸取教训。

He's learning from his failed experiment with his previous company.

所以，DeepMind 想要做的是创建一个能够在多种环境中制定计划并实现目标的智能体。

So DeepMind, what they wanna do is they're trying to create an agent that can make plans and achieve goals in multiple environments.

因此，他们最初选择了一个他们认为最适合测试智能体的环境。

And so they start out on what they think is the perfect environment for testing an agent.

那就是上世纪七十年代和八十年代由雅达利设计的所有电子游戏。

All the video games that were designed in the nineteen seventies and eighties by Atari.

而这正是原因所在。

And this is why.

考虑到那个时代视频图形的原始水平，破解雅达利游戏所需的计算能力是负担得起的。

Given the primitive state of video graphics in that era, the computing power required to crack Atari would be affordable.

由于雅达利发布了数十款游戏，智能体将有大量机会证明其通用性。

Given that Atari had released dozens of games, an agent would have plenty of opportunities to prove it could be general.

由于大多数雅达利游戏都有不断更新的分数系统，智能体将获得必要的反馈，从而学会如何玩得更好。

Given that most Atari games featured a constantly updating score, the agent would have the feedback it needed to learn how to play better.

在经历了与 Elixir 的经历后，德米斯已经成长了许多。

Demis had grown since his experience with Elixir.

在这两种情况下，德米斯都提出了宏大的抱负，但在DeepMind的情况下，他还找到了一条通向目标的路径。

In both cases, Demis had announced a maximalist ambition, but in the case of DeepMind, he had also figured out a ladder that led to his destination.

在Elixir时，他直接让公司投入开发有史以来最复杂的电子游戏，而这种过度扩张最终导致项目失败。

At Elixir, he had plunged his company straight into making the most complex video game ever and that Overreach had doomed the project.

在DeepMind，最终目标甚至更加宏伟，但德米斯在组建科学团队时，允许人们自由探索，而不是给他们设定严苛的目标。

At DeepMind, the ultimate goal was even grander, but Demis had let people tinker while he was building out the scientific team, not setting a demanding goal for them.

一旦团队组建完毕，德米斯便展现出卓越的判断力，同时不断宣扬他的愿景与使命。

And then once the team was assembled, Demis had shown exquisite judgment, and all the while he's preaching his vision, his mission.

德米斯·哈萨比斯认为，真正的通用智能将使几乎一切成为可能，其重要性将超越互联网、印刷术，甚至工业革命。

The way Demis Hassabis thinks true general intelligence would make almost anything possible, surpassing the Internet, the printing press, or even the industrial revolution in importance.

但他兼具宏伟抱负与务实精神。

But he has that combination of grand ambition and pragmatism.

他的日程安排体现了德米斯性格的两面性。

His schedule captured the two sides of Demis' persona.

当他彻夜未眠，阅读、思考、畅想时，他沉浸在自己宏大的抱负之中。

When he stayed awake into the small hours of the morning, reading and thinking and dreaming, he reveled in his maximalist ambition.

所以他的日程大概是晚上10点到凌晨4点。

So his schedule was something like at 10PM till like four in the morning.

那时候他就在做这些事。

That's when he's doing this.

然后他会睡几个小时，再去办公室。

Then he goes to sleep for a few hours and he goes to the office.

当他第二天到达办公室时，他说自己专注于攀登梯子的下一阶。

When he goes to the office, says when he arrived at the office the next day, he focused on getting to the next rung of the ladder.

同样，这种宏大的抱负与务实精神的结合。

Again, this this combination of grand ambition and pragmatism.

现在书中还有其他一些人物。

Now there's a bunch of other characters in the book.

其中一些是世界上最成功、最富有的人。

Some of them are successful and wealthiest people in the world.

埃隆·马斯克、拉里·佩奇，显然还有谷歌的创始人，都收购了DeepMind，贯穿了整本书。

Elon Musk is in all over this Larry Page, obviously, founder of Google, buys DeepMind, all over this book.

这本书里还有大量有趣的故事和轶事。

And so there's a ton of interesting stories and anecdotes in the book.

我只是想挑出其中一个。

I just wanna pull out one of them.

卢克·诺塞克是创始人基金的投资者，他希望投资DeepMind。

Luke Nosek was the investor at Founders Fund that wanted to make the DeepMind investment.

于是，卢克·诺塞克搭乘埃隆的私人飞机返回加州，同行的还有拉里·佩奇。

So Luke Nosek flew back to California on Elon's private jet accompanied by Larry Page.

飞机上的对话转向了人工智能。

The conversation on the flight turned to AI.

德米斯曾造访过埃隆在SpaceX的办公室。

Demis had visited Elon at SpaceX.

埃隆和德米斯曾讨论过，哪个目标更重要：太空探索还是开发通用人工智能。

Elon and Demis had discussed which mission mattered most, space travel or developing AGI.

埃隆表示，如果地球遭遇灾难，人类必须殖民火星。

Elon had declared that humans needed to colonize Mars in case disaster struck Earth.

德米斯反驳说，致命的AI机器人可能就是这样的灾难，但如果AI愿意，它显然可以跟随人类前往火星。

Demis had countered that killer AI robots might be one such disaster, but that the AI could obviously follow humans to Mars if it wanted to.

这两人形成了竞争性的友谊，而埃隆决定德米斯是对的。

The two men had forged a competitive friendship, and Elon had decided that Demis was right.

强大的人工智能可能确实比航天飞行更具深远影响。

Powerful artificial intelligence might indeed be more consequential than spaceflight.

埃隆承诺投资DeepMind。

Elon promised to invest in DeepMind.

现在回到这次由拉里·佩奇、埃隆和卢克·诺塞克共同乘坐的航班。

Now back to this flight that Larry Page and Elon and Luke Nosek are on.

我认为这就是拉里·佩奇得知DeepMind的方式。

I think this is how Larry Page found out about DeepMind.

于是埃隆说，我认为只有一家AI公司会成功。

So Elon says, there's only one AI company that I think is going to work.

我投资了这家公司，DeepMind。

I'm an investor in that company, DeepMind.

现在让我们回到另一个角色，彼得·蒂尔。

Now let's go back to another character, Peter Thiel.

这本书的这一部分，故事主要围绕着融资过程中的反复起伏展开。

A lot of this book at this part of the story, it's just a lot of back and forth and ups and downs in raising money.

请记住这一点，因为每个人都批评德米斯把公司卖给了谷歌。

Keep that in mind because everybody criticizes Demis for selling to Google.

但如果你正经历他所经历的一切，他会认为这是一场巨大的干扰，他需要一个拥有近乎无限资源的后盾。

But if you were going through what he was going through, he saw it as a giant distraction, and he needed a backer with essentially unlimited resources.

但关于董事会其他成员如何看待德米斯，还有一些有趣的故事。

But there's just some interesting stories about how these other players on the board were viewing Demis.

彼得·戴尔几乎没怎么接触过DeepMind团队，他本能地对一位象棋对手心存疑虑。

Peter Dale barely saw the DeepMind team, and he felt instinctively suspicious of a fellow chess player.

西奥认为，一个在成长岁月里总是用智力碾压对手的人，应当谨慎对待。

A man who had spent his formative years mentally crushing opponents should be treated with caution, Theo reckoned.

因此，在这些起起落落和艰难的融资过程中，德米斯收到了一封来自谷歌的邮件，他愿意与他们见面。

And so while these ups and downs and this this painful fundraising process is going on, Demis gets this email from Google, and he's willing to meet with him.

正因为如此，考虑到他与风险投资方紧张的关系，他非常渴望找到一个解决方案。

This is why given his testy relationship with his venture capital backers, he was eager.

一家资金雄厚的母公司可以让他摆脱无休止的融资谈判，这些谈判充斥着他的生活，分散了他对DeepMind研究的注意力。

A deep pocketed parent company could free him from the endless fundraising negotiations that cluttered his life and pulled his attention away from DeepMind's research.

我一直在不停地和投资者进行这些毫无意义的对话。

I was having these inane conversations nonstop with investors.

我感觉自己的大脑正在退化。

I felt my brain was atrophying.

我在谈论有史以来最重要的发明，但他们却总是回到‘产品在哪里？’这个问题上。

I'm talking about the biggest invention ever, and they keep coming back to where's the widget?

我心想：我要彻底革新所有产品，如果你愿意，我可以随便挑一个产品给你讲，但如果你还在问这个问题，显然你根本没理解重点。

And I'm like, I'm gonna revolutionize all widgets so I can pick you a random widget if you want me to, but you obviously haven't gotten the point if you're asking me this.

因此，融资过程糟糕透顶。

And so the fundraising was so bad.

有一度，他们几乎耗尽了资金。

At one point, they almost ran out of money.

有人会说，我们准备投入这么多资金，但后来却反悔了。

Somebody would say, you know, we're gonna we're in for x amount, and then they'd renege.

我想这促使德米斯最终选择了谷歌。

And, I think this just pushes Demis back into to choosing Google.

DeepMind 几乎濒临死亡的经历，迫使德米斯意识到，蓝天研究并不适合风险投资。

DeepMind's near death experience forced Demis to come to terms with the fact that Blue Sky Research was a poor fit for venture capital.

是时候寻找新的资助方了。

It was time to find a new backer.

因此，他解释了为什么选择将公司出售给谷歌，还描述了拉里·佩奇对他的邀约，我觉得这很有趣。

And so he describes why he'd sold to Google, and he describes Larry Page's pitch to him, which I thought was interesting.

德米斯的目标是创造通用人工智能。

Demis' goal was to create AGI.

那么，为什么还要坚持独立运营 DeepMind 的想法呢？

So why bother with the idea of an independent DeepMind?

谷歌显然是实现他抱负的绝佳之地。

Google was the obvious place to realize his ambition.

你为什么不利用我已经创造的东西呢？

Why don't you take advantage of what I've already created, Larry Page asked Demis.

这是一次他成功用于其他初创公司创始人的招聘演讲。

It was a recruitment pitch that he'd used successfully on other startup founders.

他基本上是在告诉我，也许你可以创建一家像谷歌那样的公司，但这会耗费你职业生涯中最宝贵的时光。

He was basically telling me, maybe you could build a company like Google, but it would take the best part of your career.

如果我的真正使命是构建通用人工智能，那我为什么不利用他积累的所有资源呢？

If my real mission was to build AGI, then why don't I use all the resources that he's accumulated?

我觉得这是一个相当有力的论点。

I thought that was a pretty good argument.

我是更愿意回顾自己建立了一家数十亿美元的公司，还是更愿意帮助解决智能问题呢？

Would I be happier looking back on building a multibillion dollar company or helping solve intelligence?

这是一个很容易的选择。

It was an easy choice.

当我们一起散步时，我觉得他会接受自己的这个提议。

When we went on that walk together, I felt he would have taken his own offer.

与DeepMind的风险投资方相比，这一点显而易见。

The contrast with DeepMind's venture capital backers was obvious.

德米斯曾努力说服创始人基金，让其相信DeepMind最终将改变世界上每一个产品。

Demis had struggled to persuade Founders Fund that DeepMind would end up changing every widget in the world.

但面对佩奇，他根本不需要做任何论证。

With Page, he didn't even have to make the argument.

我厌倦了四处奔波，试图证明我深知的这件事是史上最重要的，德米斯回忆道。

I was fed up with scrambling around trying to justify what I knew was the biggest thing of all time, Demis recalled.

我只是想，看吧。

I just thought, look.

我去谷歌吧。

I'll go to Google.

我会搞到海量的计算机，然后解决智能问题。

I'll get a shitload of computers, and then I'll solve intelligence.

于是，埃隆·马斯克听说了德米斯和谷歌之间的某些谈判。

And so then Elon Musk hears about some negotiations between Demis and Google.

于是他给德米斯打了电话，因为他想收购DeepMind，但谷歌没明白。

And so then he winds up calling Demis because he wants to buy DeepMind, Google doesn't get it.

当埃隆发现谷歌即将收购DeepMind时，他说德米斯不应该失去对公司的控制权。

Says when Elon found that Google was about to buy DeepMind, he said Demis shouldn't lose control of his company.

我们不能让一家大型企业掌控通用人工智能。

We can't have a giant corporation control AGI.

这对人类来说不是好事。

This is not a good thing for humanity.

于是埃隆给德米斯打电话，说：要不特斯拉收购你们？

And so then Elon calls Demis and says, how about if Tesla acquires you?

德米斯指出，特斯拉的现金流不足以支持DeepMind的研究。

Demis pointed out that Tesla was not generating enough cash to support DeepMind's research.

好吧。

Okay.

那要是SpaceX收购你们呢？

How about if SpaceX acquires you?

德米斯指出，SpaceX没有DeepMind未来所需的计算能力。

Demis points out that SpaceX didn't have the computer power that DeepMind was going to need.

因此，对于关注科技的人来说，这些人物不断在故事中出现。

And so for the people that follow tech, all these characters keep popping up in the story.

你有埃隆。

You have Elon.

你有拉里·佩奇。

You have Larry Page.

马克·扎克伯格也短暂地出现了。

Mark Zuckerberg pops up for a little bit.

从我们目前的故事时间点再过几年，萨姆·阿尔特曼出现了。

A few years from where we are in the story, Sam Altman pops up.

但在2014年1月，德米斯还是选择了谷歌，而他从未为此后悔过。

But in January 2014, Demis goes ahead with Google, and this is something he's never regretted.

谷歌以6.5亿美元收购了DeepMind。

Google bought DeepMind for $650,000,000.

德米斯获得了1.36亿美元。

Demis netted a $136,000,000.

谷歌收购后不久，DeepMind每年在员工薪酬上的支出达到2.6亿美元，是其成立前三年总支出的六倍。

Not long after the Google acquisition, DeepMind was paying $260,000,000 in staff costs annually, six times more than its total spending during its first three years of existence.

从德米斯的角度来看，这笔交易的好处是压倒性的。

From Demis' perspective, the advantages of the sale were overwhelming.

因此，在纪录片《思考的游戏》中，他谈到了这一点。

And so in the documentary The Thinking Game, he talks about this.

他说，我们的投资者并不想卖，但我们认为这对使命来说是最好的选择。

He says our investors didn't wanna but we decided this was the best thing for the mission.

在技术成熟之前，我们低估了自己的价值，未来你本可以以更高的价格出售DeepMind。

We were underselling in terms of value before it matured, and you could have sold DeepMind for more money in the future.

原因是，时间不容浪费。

The reason is because there's no time to waste.

在我还活着、大脑仍保持清醒的时候，有太多事情必须完成。

There are so many things that have to be done while I'm still alive and my brain is still in gear.

然后他对此提出了一个非常有说服力的理由。

And then he has a very compelling argument for this.

为了多活五年完成你设定的目标，你愿意用多少十亿来交换？

How many billions would you trade for another five years of life to do what you set out to do?

因此，他们正在研究的一个项目是打造AlphaGo。

And so one of the things that they're working on is they wanna build AlphaGo.

他们想构建一个能够击败围棋世界冠军的系统。

They wanna build a system that would defeat a world champion at Go.

德米斯与谷歌联合创始人谢尔盖·布林之间的这段对话，充分体现了德米斯的雄心。

This exchange between Demis and Sergey Brin, one the cofounders Google, says a lot about Demis' ambition.

德米斯告诉谢尔盖，他想打造一台能够击败围棋世界冠军的计算机。

Demis told Sergey that he wanted to build a computer that would defeat the world champion at Go.

布林显得难以置信。

Brin seemed incredulous.

那不会是不可能的吗？

Wouldn't that be impossible?

太好了，德米斯心里想。

Great, Demis thought to himself.

如果他认为这是不可能的，那我们要是做到了，就一定会非常了不起。

If he thinks it's impossible, it should be pretty impressive if we do it.

现在，这本书和纪录片中详细描述了他们是如何做到的。

Now there is a lot of detail in the book and in the documentary about how they do this.

但让我印象深刻的是，书中一个反复出现的有趣主题是：利用人工智能来提出非人类式的思路、策略或招法。

But what jumped out to me is one of the most interesting reoccurring themes in the book was this idea of using AI to come up with ideas or strategies or moves that aren't human like.

这种趋势发展到后来，甚至完全不再基于人类过去的做法。

And this progresses to the point where it's like they aren't even built on what humans have done in the past.

我做的很多标注都围绕着这一点。

A lot of my highlights center around this.

我觉得这个想法非常迷人。

I think this idea is fascinating.

因此，书的开头提到，如果你只是模仿人类的做法，是无法真正击败顶尖人类的。

And so at the beginning, it says if you pattern match what humans do, it's not going to take you all the way to beating the top human.

系统需要发现那些非人类风格的新着法。

The system needs to discover new moves which aren't human like.

我们需要构建一台能够搜索围棋中无限种变化并提出全新策略的机器。

We need to build a machine that would search the infinity of permutations in Go and come up entirely novel strategies.

我们早期的围棋系统下棋方式像人类一样。

The early version of our Go system played as a human would.

它重新发现了人类数千年积累的某些策略。

It rediscovered certain strategies that humans had learned over millennia.

随后它发现某些历史悠久的人类策略实际上可以被反制，于是便抛弃了它们。

Then it discovered that certain time honored human strategies can actually be counteracted, so it discarded them.

随着系统变得越来越强大，它的下法变得前所未见。

As the system became stronger, it played like nothing we've ever seen.

它发展出了一种完全陌生的风格。

It came up with a style that was completely alien.

书中有很多关于这一点的例子，我认为都非常非常有趣。

There are so many examples in the book about this I think are very, very fascinating.

所以现在我们已经推进了几年，一切在谷歌都进行得很顺利。

So it goes back to now we're couple years into this, and everything with Google was going well.

谷歌让他摆脱了筹款的恶性循环。

Google had liberated him from the fundraising hamster wheel.

谷歌允许他保留伦敦DeepMind的独立文化。

Google allowed him to retain DeepMind's independent culture in London.

谷歌甚至赋予了他的追随者特殊地位。

Google had even granted his followers privileged status.

在DeepMind工作的人可以进入全球任何一家谷歌办公室，但谷歌员工却被禁止进入DeepMind的办公场所。

People that worked at DeepMind could get into any Google office globally, but people that worked at Google were barred from DeepMind's premises.

在某些周里，DeepMind的一个研究团队消耗的计算资源可能超过谷歌全球Gmail网络的总和，而Gmail拥有九亿用户。

In some weeks, a single research team at DeepMind might gobble up more computational resources than Google's worldwide Gmail network, which had 900,000,000 users.

因此，埃隆和其他人看到了这一切，这促成了OpenAI的成立，以对抗DeepMind和谷歌。

And so Elon and others see what's going on, and this is gonna lead to the founding of OpenAI as a counter to DeepMind and Google.

在2015年初的某个时候，埃隆和德米斯共进午餐。

Sometime in early two thousand fifteen, Elon and Demis had lunch.

午餐时，埃隆继续抱怨，实质上指责DeepMind和谷歌不负责任。

Over lunch, Elon kept up his griping, effectively accusing DeepMind and Google of irresponsibility.

他们见面一个月后，埃隆收到了萨姆·阿尔特曼的一封邮件。

A month after their encounter, Elon received an email from Sam Altman.

他写给埃隆的邮件中说：我一直在思考，是否有可能阻止人类发展人工智能。

The email he wrote to Elon said, I've been thinking a lot about whether it's possible to stop humanity from developing AI.

我认为答案几乎肯定是不可能的。

I think the answer is almost definitely not.

如果这件事终究会发生，那么由谷歌以外的其他人率先去做似乎更好。

If it's going to happen anyway, it seems like it would be good for someone other than Google to do it first.

你可以看到，随着人们越来越认识到发展通用人工智能的重要性。

And you can see this as people begin to understand more and more the importance of developing AGI.

于是，有越来越多的人加入这场竞争。

You have all these other people that jump into this competition.

当一种具有无限潜力的技术浮现时，永远不可能就谁应该控制它达成安静的共识。

When a technology of infinite potential comes into view, there will never be a quiet consensus about who should control it.

鉴于如此重大的利益——权力、金钱、科学荣耀以及人类的未来——冲突是不可避免的。

With so much at stake, power, money, scientific glory, the future of humanity, conflict is unavoidable.

因此，德米斯非常务实。

And so Demis is practical.

他了解自己所打交道的人。

He knows the people he's dealing with.

他知道这类人迟早会想要自立门户。

He knows these kind of people are going to want to start their own thing.

埃隆和萨姆·阿尔特曼联手创立了OpenAI，一个非营利实验室，明确旨在打破谷歌DeepMind对AGI的垄断。

Elon and Sam Altman team up to launch OpenAI, a not for profit lab explicitly aimed at breaking the Google DeepMind AGI monopoly.

德米斯说，如果一些有影响力的人能够理解这项技术的影响，他们不会只是袖手旁观。

Demis said, if you have powerful people who are able to understand the impact of the technology, they're not just gonna sit on the sidelines.

他们不会满足于仅仅当你的顾问。

They won't be content to just be your advisers.

所以，显而易见的是，我们所谓的顾问实际上是我们真正的竞争对手。

So, obviously, what was going on was our supposed advisers were really our rivals.

于是，当我回到用人工智能生成这种外星般的、非人类想法的想法时，他们继续改进AlphaGo。

And so then when I go back to this idea of using AI to just generate this alien, like, nonhuman ideas, they continue to improve AlphaGo.

现在他们有了一个叫做AlphaGo Zero的想法。

Now they have this idea called AlphaGo Zero.

这个想法是，与其从专家人类对局开始训练代理，不如让它完全通过自我对弈来学习，通过尝试随机走法并发现哪些能产生奖励信号。

The idea is that rather than training the agent initial in expert human games, it would have to learn exclusively by playing against itself, by experimenting with random moves and discovering which ones generated a reward signal.

仅靠自我对弈，该系统远远超越了它的前身。

Learning only from self play, the system outclassed its predecessor by a mile.

通过摆脱人类智慧的束缚，这个模型发现了凡人棋手从未知晓的策略，达到了对围棋奥秘的新理解。

By unshackling itself from human wisdom, the model had discovered strategies unknown to mortal players, arriving at a new understanding of Go's mysteries.

人类从未意识到自己懂得有多么少。

Humans had not understood how little they had understood.

人工智能对数百年的人类智慧作出了裁决，肯定了一些结论，同时否定了另一些。

AI stood in judgment over centuries of human wisdom, vindicating some verdicts and tossing out others.

书中我最喜欢的一些部分，就是你能明显感觉到，他认为这就像他的宿命。

And then some of my favorite parts of the book is just you can just tell the that he believes that this is, like, his fate.

这是他的命运。

This is his destiny.

在向你引用德米斯的这段文字之前，其实有一段彼得·蒂尔的话让我觉得很有意思，他在谈论德米斯。

It's actually interesting before I read this this excerpt from Demis to you, which I found fascinating, there was an interesting, you know, few sentences from Peter Thiel, and he's he's talking about Demis.

他说，天才很少在普遍意义上出色。

He says geniuses are seldom brilliant in a general way.

他们往往特别适合某个特定的使命。

They tend to be brilliantly suited to a particular mission.

我的朋友丹尼尔·埃克是Spotify的创始人。

My friend Daniel Eck was the founder of Spotify.

他称之为创始人与使命的契合。

He calls this founder problem fit.

我觉得这是一个非常有趣的观点。

I think it's a really interesting idea.

让我们回到这个观点：德米斯，你完全可以看出来。

Let's go back to this idea that Demis, you could just tell.

他相信这是命运。

He believes this is fate.

这是宿命。

This is destiny.

他有一种美妙的表达自我的方式。

He's got a beautiful way of explaining himself.

人工智能的发展方式有点像工业革命。

The way AI has developed is a bit like the industrial revolution.

它以某种方式发展，但那有点幸运。

It developed in a certain way, but that was kinda lucky.

假设在工业革命初期，我们发现了能源和引擎，但想象一下，地下却没有煤炭或石油。

Suppose at the start of the industrial revolution, we had found out about energy and engines, but then imagine that there were no coal or oil in the ground.

毕竟，情况本不必如此。

After all, there didn't have to be.

死去的恐龙和古老的树木就这样静静地等待了六千万年，等着被挖出来？

Dead dinosaurs and ancient trees just waiting there sixty million years ready to be dug out?

如果你仔细想想，这有点不合理。

It's kind of unreasonable if you think about it.

为什么它们不在地下自然分解而变得无用呢？

Why wouldn't they just decay in the ground and become useless?

它们没有分解，这真是太巧了。

Quite convenient that they didn't.

这或许引出了另一个我们值得探讨的话题：这里究竟发生了什么？

And maybe that speaks to another conversation we could have about what's really going on here.

我们为什么会遇到这种巧合？

Why would we have this coincidence?

这里的类比是，互联网之于人工智能，就如同煤炭和石油之于工业革命。

The analogy here is the Internet has been for AI what coal and oil were for the industrial revolution.

你只需在地下钻个洞，就能得到黑色黄金。

You could just literally drill a hole in the ground and get black gold.

今天，我们只需下载整个互联网即可。

Today, we can just download all of the Internet.

这两种资源本不必存在。

Neither of these resources had to be there.

死去的恐龙就是互联网。

The dead dinosaurs are the Internet.

人类建造互联网时有着不同的目的，但令人惊讶的是，有一天我们突然意识到，我们拥有了相当于石油的东西。

Humanity built the Internet for a different purpose, and kind of amazingly, we woke up one day and realized that we've got the equivalent of oil.

所以，你和我之前讨论过，德米斯非常独立。

And so you and I have talked about that Demis is fiercely independent.

他是个特立独行、按自己方式做事的异类。

He's this outlier doing things his own way.

在某些时候，这可能是优势，而在另一些时候，这可能是劣势。

And at certain times, that could be a strength, and at certain times, it could be a weakness.

在这个故事节点上，OpenAI 尽管成立时间晚得多，却超越了 DeepMind，书中详细描述了它们当时采取的两条不同路径。

At this point in the story, OpenAI gets ahead of DeepMind even though it was founded much later, and the book goes into detail about the two different paths that they're taking at the moment.

但我认为，作者描述了理解德米斯的性格和人生经历，如何能解释他为何会犯下这个错误。

But I think the author describes how understanding Demis' personality and his life history could have accounted for him to make this mistake.

因此，他说在这一时期，DeepMind 应该像 OpenAI 那样转向语言模型，但 DeepMind 对自己的研究太过着迷。

And so he says at some point in this period, DeepMind should have pivoted to language models just as OpenAI did, but DeepMind was too excited by its own research.

它早已习惯于成为全球顶尖的 AI 实验室。

It was accustomed to being the world's top AI lab.

它几乎无法想象，一个模仿者可能会超越它。

It could scarcely imagine that a copycat outfit might overtake it.

此外，Demis 反对追随 OpenAI 的做法。

Besides, Demis rebelled against the prospect of following OpenAI's example.

他一生都在走自己的路。

All of his life, he had beaten his own path.

他童年时痴迷于国际象棋，未成年时为 Bullfrog 兼职，在剑桥对 AI 怀疑论共识表现出早熟的不耐烦，拥有非英式的创业热情，从游戏设计不可思议地转向神经科学——Demis 比硅谷大多数自诩的反叛者更加原创、更加特立独行。

His obsessive childhood chess, his underage moonlighting for Bullfrog, his precocious impatience with the AI skeptical consensus at Cambridge, his un British appetite for entrepreneurship, his improbable leap from game design to neuroscience, Demis was far more original and far more of a contrarian than most of the self identified contrarians of Silicon Valley.

然后，我最喜欢的一句格言是：行动表达优先级，因此我们通过观察他的行为来看什么对他最重要。

And then one of my favorite maxims is that actions express priority, so we see what's important to Demis by looking at his actions.

从大学时代起，他就痴迷于解决蛋白质折叠问题。

Since he was in college, he was obsessed with this idea of trying to solve protein folding.

再次强调，这正是他将获得诺贝尔奖的原因。

Again, this is what he's gonna win the Nobel Prize for.

在这个故事的节点上，他认为人工智能已经发展到足以首次解决这一问题，他解释了这为何如此重要。

And at this point in the story, he thinks that AI has progressed sufficiently to solve this problem for the first time in human history, and he talks about why this is so important.

蛋白质是生命的基本构建单元。

Proteins are the building blocks of life.

它们为器官和肌肉、激素和头发、血液和脑细胞提供结构和功能。

They provide the structure and also the function of organs and muscles, hormones and hair, blood and brain cells.

解决蛋白质折叠问题或预测蛋白质所形成的复杂结构，本身就很迷人，而且几乎必然能推动医学进步。

The idea of solving protein folding or predicting the complex shapes that proteins assume was both fascinating in its own right and almost certain to unlock medical advances.

了解蛋白质的结构将帮助研究人员设计出能够与其表面结合的药物分子。

Knowing the structure of proteins would help researchers to come up with drug molecules that could bind to their surfaces.

了解这些结构的形成方式，可能为帕金森病或阿尔茨海默病找到治愈方法，因为这两种疾病都被认为与错误折叠的蛋白质有关。

Knowing how those structures formed might unlock cures for Parkinson's or Alzheimer's since both diseases were thought to be linked to incorrectly folded proteins.

当他们最终解决这个问题时，他们将把这项技术及所有研究成果免费向全世界开放。

And when they wind up solving this, they make the technology and all their research free to the world.

当DeepMind的项目完成后，该公司将成果无偿奉献给科学界，让全球的研究人员都可以免费使用这一发现。

And when DeepMind's project was completed, the company gifted its results to science, allowing researchers all over the world to make free use of its discovery.

然后，德米斯与历史上大多数伟大企业家的共同点在于，当你做出伟大的成就时，你不会躺在功劳簿上睡大觉，也不会止步不前，而是立刻去投身新的事情。

And then one thing that Demis has in common with most of history's greatest entrepreneurs is like when you do something great, you don't sleep on the wind, you don't rest on the laurels, you just go and do something else.

它说，当德米斯解决了一个重大问题时，他并不会花太多时间沉浸于成就的喜悦中。

It says, when Demis solves something big, he doesn't pause to spend much time savoring the achievement.

因此，德米斯谈到自己很不讲理，总是不断追求更多。

And so Demis talks about the fact that he's unreasonable, that he's always pushing for more.

公司内部有些人确实提到，他永远不满足。

There's people inside the company actually talk about the fact that he's never satisfied.

他们总是想要更多。

They always wants more.

他们对此有一句非常精彩的话。

They have this great line this great line about it.

他们称之为‘德米斯驱动开发’。

They call it Demis driven development.

在DeepMind，我们有一种叫做‘德米斯驱动开发’的做法。

At DeepMind, have something called Demis driven development.

如果与德米斯的评审会议安排在周二，你就会被催促在周一之前完成下一轮升级。

If a review meeting with Demis has been scheduled for Tuesday, you were urged to complete the next round of upgrades by Monday.

无论有多少升级成果提交，德米斯总是想要更多。

No matter how many upgrades arrive, Demis wants more of them.

因此，在真正获得证据之前十多年，德米斯就反复强调一个观点：他坚信人工智能能够带来科学上的奇迹。

And so something that Demis would repeat way a decade, more than a decade before he actually had proof of this, is the fact that he really believed that AI could offer scientific miracles.

他谈到，信念会带来可实现性。

And he talks about the fact that belief comes to affordability.

他说，如果团队的领导者认为这个问题根本不可能解决，那你就绝对无法攻克它。

He says, you definitely can't crack a hard problem if the person leading the team thinks it's not possible.

因此，人工智能可能带来的奇迹之一，就是他们在AlphaFold上所做的工作。

And so one example of what the miracles that AI could offer is what they did with AlphaFold.

数百位学术科学家花费了几十年时间研究蛋白质折叠。

Hundreds of academic scientists had spent decades on protein folding.

DeepMind团队最多时只有大约20人，他们是如何击败所有人的？

How was it that the DeepMind's team, numbering perhaps 20 at its peak, had defeated all of them?

于是，这本书转入了ChatGPT的发明过程，这时Demis意识到自己正为生存而战。

And so then the book gets into the invention of ChatGPT, and this is where Demis realizes he's in a fight for his life.

这是一场战争。

This is war.

据说，在ChatGPT发布前一晚，OpenAI的团队打赌到周末结束时会有多少人尝试这个工具。

Says the night before ChatGPT's release, OpenAI's team placed bets on how many people might try the tool by the end of the weekend.

有人猜几千人，其他人猜几万人。

Some guessed a few thousand, others guessed tens of thousands.

为确保万无一失，公司准备了足以支持十万用户的服务器容量。

To be safe, the company readied enough server capacity for a 100,000 users.

五天内，他们就收集了一百万用户。

Within five days, they collected a million users.

两个月内，用户数已达到惊人的一亿，使其成为有史以来增长最快的消费类应用。

Within two months, it had amassed an astonishing 100,000,000, making it the fastest growing consumer application ever.

这是德米斯的回应，再次说明，他在描述别人写的这本书时，表现出一种病态的竞争心态。

This is Demis' response, and, again, he is patholog the way he was describing the book by somebody else is pathologically competitive.

2023年4月，我拜访了德米斯，问他感觉如何。

At the April 2023, I visited Demis and asked how he was feeling.

这是一场战争，他回答道。

This is wartime, he answered.

OpenAI 和微软已经 literally 把坦克停在了草坪上。

OpenAI and Microsoft have literally parked the tanks on the lawn.

因此，德米斯也花时间思考他竞争对手的动机。

And so Demis also spends time thinking about the motivations of the people that he's competing with.

我认为这非常重要。

And I think this is really important.

我认为史蒂夫·乔布斯说得最好。

I think Steve Jobs put it best.

史蒂夫说：‘我年纪越大，就越坚信动机至关重要。'

Steve said, the older I get, the more I'm convinced that motives make so much difference.

史蒂夫会把惠普作为他的标杆例子，指出在创始人经营公司时，他们的首要目标是制造出色的产品，而不是成为最大或最富有的公司。

Steve would use HP as his North Star example, that their primary goal, back when the founders were running it, obviously, that their primary goal was to make great products, not be the biggest or the richest.

对乔布斯来说，这种区别绝非表面现象。

And that distinction to Jobs wasn't cosmetic.

它是根本性的。

It was foundational.

因此，德米斯在谈论萨姆·阿尔特曼时，提到了保罗·格雷厄姆的见解。

And so Demis is talking about Sam Altman, and it says Demis recalled what Paul Graham wrote.

保罗·格雷厄姆是阿尔特曼最亲密的职业导师之一。

Paul Graham was one of Altman's closest professional mentors.

保罗写道：萨姆非常擅长获得权力。

And Paul wrote, Sam is extremely good at becoming powerful.

你即使把他空投到一个满是食人族的岛上，五年后回来，他也会成为国王。

You could parachute him into an island full of cannibals and come back in five years, he'd be the king.

德米斯说：对于任何试图构建通用人工智能的人来说，这都是一个值得思考的问题。

Demis said, I think there's a question for anyone trying to build AGI.

你构建它的理由是什么？

What are your reasons for building it?

我的理由是科学性的。

My reasons are scientific.

有些人显然是为了其他目的在构建它。

Some are definitely building it for other purposes.

德米斯不仅愤怒，而且极具竞争性。

Demis was not just furious, he was ferociously competitive.

这是对OpenAI和ChatGPT所做之事的回应。

And so this is the response to OpenAI and ChatGPT that they do.

拉里·佩奇坚持认为，谷歌必须尽一切可能赶上来，否则将一无所有。

Larry Page insisted that Google should do everything conceivable to catch up, otherwise it would be nowhere.

德米斯开始引导他的团队转变思维方式。

Demis set about preparing his troops to think differently.

他宣布，DeepMind广泛的基础研究项目必须进行缩减。

He declared that DeepMind's broad portfolio of blue sky research bets would have to be paired back.

公司将会停止发布竞争对手可以复制的关键性研究。

The company would stop publishing mission critical research that competitors could copy.

它将专注于工程，而不仅仅是科学。

It would focus on engineering and not just science.

研究人员必须从和平时期的心态转向战时心态。

Researchers would have to make the mental shift from peacetime to wartime.

他们还计划将谷歌大脑和深度思维合并。

They also planned to merge Google Brain and DeepMind.

就连谷歌也承担不起重复研究团队的奢侈。

Not even Google could afford the luxury of duplicate research teams.

他们开始研发下一代语言模型，该模型将被称为Gemini。

They began to work on their next generation language model, which would be called Gemini.

谷歌将把其研究、计算能力和营销力量集中到一个聊天机器人上。

Google would put its research, computing power, and marketing muscle behind a single chatbot.

关于这段时期对公司有多宝贵，这位深度思维的研究人员有一句精彩的评价。

And there's a great quote from this researcher on DeepMind about how valuable this time was to the company.

我认为，我们可能需要暂时做第二名，这样才能激发自己的紧迫感。

My view is that we probably needed to be second for a while just to light a fire under our own ass.

没有什么比公开的羞辱更能激发行动了。

There's nothing like public humiliation for galvanizing action.

然后你会发现，德米斯天生就是为这个时代而生的。

And then you see Demis was just built for this time.

他的一生都在为这一刻做准备。

His entire his entire life, he's prepared for this.

他就是热爱竞争。

He just loves competition.

德米斯如鱼得水。

Demis was in his element.

他职业生涯中最伟大的赛事才刚刚开始。

The greatest tournament of his career was just getting started.

一切都是竞争，竞争会带来这种疯狂的冲刺，德米斯说。

Everything is competitive, and competition brings this mad rush, Demis said.

我心中一直有这个念头。

I've always got this in the back of my mind.

他只是在描述自己生命中的这段时光。

And he's just describing this time in his life.

这一年过得很艰难。

It's been a hard year.

部分原因是，现在每个人都知道了我二十多年甚至更久以来一直知道的事：人工智能是史上最重要的事情。

It's partly because everyone knows now what I've known for twenty years or more that AI is the most important thing ever.

风险资本家们正在资助任何能动的东西。

Venture capitalists are funding anything that moves.

中层工程师们收到了创办初创公司的邀约，尽管他们并不适合经营公司。

Mid level engineers are getting offers to do startups even though they're not suited to running a company.

世界上最大牌、最有抱负、最凶猛、最激进的人才都涌入了这个领域。

You've got the biggest titans, the most ambitious, most ferocious, most aggressive people in the world crowding into the sector.

DeepMind 坚持了严格的统一性。

DeepMind embraced a strict unity.

所有团队成员都全力以赴改进单一模型。

All team members poured their energies into improving one single model.

接下来，他们采用了 meritocracy（ meritocracy 译为“唯才是举”）。

Next, they embraced meritocracy.

任何团队成员都可以提出对模型的改进并进行测试。

Any team member was welcome to propose an improvement to the model and test it.

如果改进提升了性能，就会被纳入所有人共同构建的主代码中。

If the upgrade boosted performance, it was added to the master code on which everyone was building.

资历、个人魅力、关于某项技术为何有效的华丽理论主张，这些都不影响最终进入程序的内容。

Seniority, force of personality, dazzling theoretical claims as to why something should work, none of that affected what went into the program.

只有测量结果才重要。

Only measurement mattered.

因此，由于德米斯负责谷歌所有人工智能项目，他说，这个项目规模太大，我已经不再写代码了。

And so since Demis is running all of Google's AI, he says, the project is so big that I don't code anymore.

我不再直接设计东西，所以我的技能更多体现在同时在脑海中管理一百个不同项目上。

I don't design things directly, so my skills are more about holding a 100 different projects in my mind.

在时间间隔极短的复杂任务之间频繁切换。

Context switching between complicated things with negative minutes of time between.

划分职责。

Laying out division.

在实现宏大目标的过程中选择正确的中间目标，并培养他人承担任务。

Picking the right intermediate targets on the way to the big goal, and nurturing people to take things on.

我最常使用的词是‘不懈’——不懈的进步、不懈的发布、一个持续创新的生产机器。

The word I'm using the most is relentless, relentless progress, relentless shipping, a relentless production machine for innovation.

在导致谷歌DeepMind诞生的那场混乱的‘包办婚姻’之后不到两年，德米斯的团队就缩小了技术差距。

Less than two years after the messy shotgun marriage that created Google DeepMind, Demis' team had closed the technical gap.

这是一个了不起的成就。

It was a considerable achievement.

因此，他们开始讨论对未来的展望。

Hence, they talk about where they see this going.

人类总是不断构想长期目标，并规划下周和下个月需要做什么来实现这些目标。

Humans are forever conceiving long term objectives and planning what they need to do next week and next month in order to realize them.

未来的人工智能也会以同样的方式行事。

Future AIs would behave in the same way.

举个例子，为了完成研发超导体以解决能源短缺这个任务，人工智能可能会列出一份阅读清单、开展实验、发明新型材料等等，用一年甚至更长的时间来推进目标落地。

Task, for example, to help solve energy scarcity by inventing a superconductor, an AI might draw up a reading list, conduct experiments, invent novel materials, and so on, pursuing its goal over the space of a year or more.

这一切都归根于戴密斯本身就是一名使命者，他始终在践行自己的使命，并且对自己的动机有着无比清晰的认知。

And it all goes back to the fact that Demis is a missionary, that he's on a mission, and that he's very clear about what his motives are.

我做这件事是为了追寻知识，发展科学。

I'm doing this for knowledge and science.

这是我毕生的事业。

This is my whole life's work.

我必须去做所有该做的事。

I have to do what's necessary.

这个使命已经刻进了我的骨子里。

The mission is in me.

它已经和我融为一体了。

It is infused in me.

你无法将它与我分开。

You can't separate it from me.

我绝不否认它可能很固执或难以相处。

I'm definitely not denying it can be strong willed or difficult.

我认为我必须如此。

I think I have to be.

如果我像风中的芦苇一样摇摆，我就无法胜任领导者的职责。

If I was like a reed in the wind, I wouldn't be doing my job as a leader.

德米斯的核心理念是，金钱和权力本身并不是目的。

Demis' core theme is that money and power were not ends in themselves.

它们是通往科学知识的手段。

They were a means to scientific knowledge.