AI正在扼杀软件吗？——与Bret Taylor对话

他们真棒。

Good for them.

我想说，如果非得挑一件事，你知道，我很好奇。

I I would say if there's one thing that is, you know, I'm curious.

我真的对新技术感到好奇。

I I really am curious about new technologies.

如果你对任何你看到的事物都立即持负面反应，那么要接受自己业务或技术上的变化就会很难。

I think it is hard to embrace change in your own business or change in technology if you immediately are reflexively negative about anything that you see.

我鼓励那些对某项技术持怀疑态度的人，去找一个你信任、观点与你不同的人。

I encourage people who are skeptical about a technology to find someone whose opinion that you trust, who has a different opinion than you.

和他们一起吃午餐或吃晚餐，说：说服我，让我知道我错了。

Have lunch or have dinner with them and say, convince me that I'm wrong.

我对一些我曾持怀疑态度的技术都这样做过，我会找一位对它非常看好的创业者，然后说：我只是想了解你看到了什么。

And I've done this with a few different technologies that I was skeptical about and I'd find, you know, a fellow entrepreneur that was really bullish on it and I'd say, I just wanna I wanna understand what you see.

我这么说是因为，并不是每一个被炒作的技术都值得被如此吹捧。

And I say that just because I not every hyped technology is, you know, worth the hype.

但如果聪明人都对它感兴趣，通常说明他们看到了某些非常重要的东西。

But there's usually if smart people are interested in it, there's usually something really important that they see.

我认为，如果你拥有这种好奇心，就更有可能将其中有益的部分应用到你的事业或生活中。

And I think if you have that curiosity, you'll be more likely to be able to apply what's good about it to your business or to your life.

我确实有这种好奇心。

And I definitely have that.

我常开玩笑说，我从未遇到过一个悲观的创业者，而我自己也绝对不是这样的人。

I joke I've never met a pessimistic entrepreneur and I'm definitely not one either.

是的。

Yep.

这很有趣，因为我们刚刚讨论了技术的重大转变：从搜索到社交，从桌面到移动，现在则是全部转向人工智能。

So it's interesting because we've just talked about these big shifts in technology, search to social, desktop to mobile, and now all of it to AI.

有些人认为，人工智能是最后一个发明，此后不会再有其他重大转变了。

There's some people that say AI is the last invention that there won't be another shift after this.

你相信这一点吗？

Do you believe that?

完全不是这样。

Not at all.

真的吗？

Really?

我总喜欢做这种思维练习。

I I always I like to do the thought exercise.

美国成立于1776年。

So The United States was found in 1776.

当时，我不知道我们经济中农业占比多少，但我猜全国大约95%的人都是农民。

At the time, I don't know what percentage of our economy was agrarian, but I'm guessing it was like 95% of the country were farmers.

我总是想象，不知道哪位开国元勋最懂技术，也许是本杰明·富兰克林吧。

And I always imagined taking I don't know who the most tech savvy founding father was, but Benjamin Franklin, maybe.

我不确定。

I don't know.

试着想象如果他突然传送过来坐在我们中间，首先那会有点酷。

And try if he were teleported right here in between us, first of that would be a little cool.

但其次，他要花多长时间才能弄清楚食物都是从哪儿来的呢？

But secondly, like, how long would it take him just to understand where all the food comes from?

可能需要很长时间，因为我们长期以来一直在自动化农业和分销体系，现在却对此习以为常。

Probably a really long time because we've been automating essentially agriculture, distribution for so long, we just take it for granted right now.

我认为，我们常常看到新技术取代了我们目前所做的工作。

I think, you know, we often see new technologies and see what it displaces that we currently do.

但我们是围绕技术构建了经济和文化，而不是反过来。

But we create an economy and a culture around our technology, not the other way around.

所以在过去大约三百年里，我们让能源变得相对充裕。

And so in those past, you know, three hundred years or something, we've made power relatively abundant.

我们让食物变得充裕。

We've made food abundant.

我们让交通变得相对便利。

We've made transportation relatively abundant.

想想看，我们俩都是二十四小时前乘飞机来到这里的，现在却在通过互联网做播客。

And, you know, just think about the fact that both of us flew here on airplanes, you know, twenty four hours ago and we're podcasting over the Internet.

这太令人震惊了。

It's mind blowing.

对吧？

Right?

我们并不觉得这令人震惊，但它确实令人震惊。

We don't think of it as mind blowing, but it's mind blowing.

我想我们只是缺乏想象力。

I think we just lack the imagination.

我觉得这很美妙。

I think that's delightful.

我感到兴奋的是，在我有生之年，希望再过二十五年，会出现一些今天我无法理解的事物、我无法理解的工作，以及希望由人工智能发现的新技术。

I'm excited that in my lifetime, hopefully twenty five years from now, there will be things that today I wouldn't understand and jobs I don't understand, technologies that are hopefully discovered by AI.

很多人看到人工智能能做我们能做的事，就觉得它削弱了我们的独特性，但我不同意，就像汽车并没有取代马一样。

I think a lot of people look at AI because it can do things that we can and it somehow thinks it takes away from our humanity and I just disagree just like the car didn't take away from the horse.

这只是不同而已，我对它持乐观态度。

It's just different and I think I'm just optimistic about it.

我不是说不会出现有意义的颠覆。

I don't mean there won't be meaningful disruptions.

技术很难，我们刚刚提到，我自认为是软件工程师，就在过去四个月里，这份工作已经发生了巨大变化。

Technology is hard and we just talked about I self identify as a software engineer and even over the past four months that job has changed dramatically.

曾经造就我的那些技能，现在比四年前的价值低多了。

The skills that made me what I am are less valuable now than they were, you know, four years ago.

但我对这种变化感到兴奋，因为我对人类进步充满期待。

But I'm excited for that because I'm excited for the progress of Means for Humanity.

好的。

Alright.

我们来聊聊Sierra吧。

Let's talk about Sierra.

你现在已经运营了七个季度，这很惊人，因为你的年经常性收入已经达到十亿美元了。

So you're seven quarters in, which is wild because you're already doing a 100,000,000 in annual recurring revenue.

你50%的客户年收入超过十亿美元，20%的客户年收入超过一百亿美元。

Have 50% of your customers have more than 1,000,000,000 in revenue and 20% of your customers have more than 10,000,000,000 in revenue.

所以你是在销售AI客户互动解决方案。

So you are selling in AI customer engagement.

你面向的是非常大的公司，并且能拿到非常大的订单。

You're selling into very big companies and you're getting very big deals.

你的某些客户，比如SiriusXM，甚至会让您的平台自动采取行动。

Some of your customers like SiriusXM will actually have your your platform take action.

对吧？

Right?

这可不是聊天机器人。

So this is not chatbot.

我觉得这一点很重要，必须说明。

I think this is important to say.

这可不是聊天机器人。

This is not chatbot.

这是智能代理。

It's agent.

当我的卫星收音机无法工作时，人工智能会为我重置它，让我能继续收听，而这通常需要

It's when my satellite radio isn't working, the AI will reset it for me, and I'll be able to listen again where that typically would have taken

顺便说一下，这个人造智能代理正在向太空中的卫星发送信号，卫星再与你的汽车通信。

And by the way, that that AI agent's sending the signal up to a satellite in space, which is communicating with your car.

那个人是谁？

Person Who is that?

没有人类参与。

No person involved.

人工智能代理与卫星对话，这在2026年已经是再平常不过的事了。

AI agents talking to satellites, like, and that's just a normal thing that we do now in 2026.

这太棒了。

It's pretty awesome.

所以我的问题是，这确实是个大事件。

And so my question, because that is that's a big deal.

对吧？

Right?

这需要Sirius说：好吧，布雷特，继续让你的概率性AI技术发挥作用。

That that requires Sirius to say, alright, Brett, go ahead and let your AI technology, which is probabilistic.

对吧？

Right?

发送信号到我们的卫星，我们会相信一切都没问题。

Go ahead and send a signal to our satellites, and we're gonna trust everything is good.

这对我来说是个大问题，关于Sierra，因为关于AI的推出，已经有很多讨论，哪些地方做对了、哪些做错了，以及AI项目是如何从试点进入生产的。而你现在已经在生产中与这些大公司合作，让他们将重要的操作托付给你。

And that to me is a big question that I have about Sierra which is there's been so much discussion about this AI rollout and what's, you know, gone right and wrong and and what's taken the AI projects from pilot into production And you've got it in production with these big companies that are trusting you with big actions.

你是怎么让他们对这些事情产生信任的？

How have you gotten them to trust you with this stuff?

我认为这一切都始于客户和消费者。

I think it all starts with the customer and the consumer.

是的。

Yeah.

你刚才提到了一些AI的风险，我会谈谈这些。

You were implying some of the risks of AI, which I'll talk about.

但如果你调查你的听众以及他们对等待通话的感受，我会很惊讶如果能找到一个人对这种体验有正面评价。

But if you survey your listeners and survey their sentiment about waiting on hold, it would surprise me if you could find one person who had a positive experience with that.

我觉得我们一生中花了好几年时间在等待通话上。

I think we spend like years of our life waiting on hold.

而原因在某种程度上是商业和科技因素的混合。

And the reason for it is somewhat like a mix of sort of business and technical.

直到最近，电话线路还是模拟的。

Until recently, the phone line was analog.

因此在那里无法构建数字体验。

So there was no way to build a digital experience there.

所以如果有人打电话给你，你就必须配备呼叫中心来接听电话。

So if someone did call you on the phone, you had to staff a call center to answer it.

而如果你想避免等待，就必须多雇一些人，但他们大部分时间都闲着，这会带来巨大的成本。

And if you wanna not wait on hold, had to overstaff that call center by with people would be sitting around most of the time doing nothing, which just cost a lot of money.

如果你考虑一个消费品牌，假设每位客户每月的平均收入是10美元，那可能还低于一次电话呼叫的成本。

And if you think of a consumer brand and let's say your average revenue per customer every month is $10, that might be less than the cost of a single phone call.

因此，对大多数企业来说，实际上根本负担不起与客户进行电话沟通的成本。

So you actually can't afford for most businesses, literally can't afford to have a phone conversation with your customers.

但现在你可以了，因为你可以让一个AI代理来接电话。

But now you can because you can actually have an AI agent pick up the phone.

你不需要再等待通话。

You don't need to wait on hold.

它可以支持多种语言。

It could be multilingual.

它可以完美访问你的系统。

It can have perfect access to your systems.

它可以帮你找到英国当地的零售商。

It can help you find if you know the retailer next in The UK.

它可以帮你查找你的订单。

It can help, you know, find your order.

如果你在伦敦点餐配送，它可以帮助你，无论是司机还是消费者。

If you if you order food in London and delivery, it can help you whether you're a driver or a a consumer of that.

Cigna健康保险，我们与他们合作，在不到两个月内就上线了Sierra系统，它能帮助你理解你的保险福利并处理理赔。

Cigna Health Insurance, we went live with them in less than two months with Sierra and it can help you understand your benefits, process claims.

所有这些功能都非常实用。

All of these things are just really useful.

而且我认为，因为消费者不想等待，他们希望立即得到答复。

And and I think because consumers don't wanna wait and they would like an answer now.

而且由于这项技术能够实现——因为我们已经将最后一个模拟渠道——电话，实现了数字化——这同时也解锁了新的商业模式。

And because you can do things with this technology because we've essentially digitized the last analog channel which is the telephone, it's unlocking new business models as well.

我最喜欢的一个例子是Rocket Mortgage，对于那些不住在美国的人，我想它是美国最大的消费者抵押贷款发起机构。

One of my favorite examples of this is Rocket Mortgage which for those of you not from The States is I think the largest consumer mortgage originator in The States.

如果你访问redfin.com，你可以使用基于Sierra构建的AI代理来寻找房屋。

If you go to redfin.com, you can use an AI agent built on Sierra to find a home.

然后你可以前往rocket.com为这套房子融资并申请抵押贷款。

You can go then rocket.com and finance that home and get a mortgage.

接着，通过他们新推出的Mr.

And then with their new Mr.

通过AI代理提供Porter贷款收购服务。

Porter acquisition service that loan, all with AI agents.

太棒了。

It's amazing.

对。

Right.

对。

Right.

而这正是你要解决的痛点，对吧？

And this is this is the pain, right?

你正在解决这个问题。

That you're solving.

但我还想多听听你和他们讨论的内容。

But I want to hear a little bit more about the discussions that you have with them.

你告诉他们要信任我们的技术，这种说服力。

The fact that and the persuasion where you've said, trust our technology.

他们已经看过这项技术了。

They've looked at the technology.

这项技术现在足够好了吗？

Is the technology good enough now?

因为之前有人怀疑它能否处理如此复杂的事情。

Because there were some doubts that it could handle such complex things.

所以谈谈这方面的情况。

So talk a little bit about about that.

你们是如何让他们信任西耶拉来执行这些操作的？

How have you gotten them to trust Sierra with these actions?

因为我们大家都同意，这个痛点是真实存在的。

Because we we all agree that the pain is real.

是的。

Yeah.

好吧，我先从一件重要的事情说起，不夸张地说，那就是人类也经常犯错。

Well, so I'll start with actually an important thing to to ground ourselves, no pun intended, and which is where humans make a lot of mistakes too.

如果你的员工在与客户沟通，每个员工都能完美完成所有任务的可能性几乎为零。

So if you have an associate talking to your customers, the odds that every associate is gonna do everything perfectly is essentially zero.

我认为我们对计算机的期望更高，这可以理解。

And I think we just have a higher expectation of computers, which is understandable.

但这些AI代理所做的事情，并不一定是在取代之前已经完美无缺的流程。

But, you know, what these AI agents are doing aren't necessarily doing something that was previously perfect.

我认为很多公司都明白这一点。

And I think a lot of companies understand that.

如果你有庞大的销售团队或客户服务团队，你一定深有体会，因为你肯定接到过客户打来的抱怨电话，事情没办好。

And if you if for anyone who has a large sales team or customer service team, know exactly what I'm talking about because you've gotten the phone calls from a client where it didn't go well.

所以，首先我想反直觉地指出，AI代理实际上比它们所取代的大多数系统更可靠。

So first, I would actually argue counterintuitively, AI agents are actually more reliable than most of the systems that they replace.

顺便说一句，这并不意味着它们是完美的。

It doesn't mean they're perfect, by the way.

它们只是比之前那些极易出错的人工操作体系更可靠一些。

They're just more perfect than the very fallible human operation systems that that preceded them.

第二点是我们在CIRA所做的事，本质上是为这些原本非确定性的系统增加更多鲁棒性和确定性。

And the second thing is essentially what we do at CIRA is try to essentially put more robustness and determinism around these inherently nondeterministic deterministic systems.

我们实现这一点的最佳方式之一是使用一种叫做模拟的技术。

One of the best ways we do that is through a technology called simulations.

我们的所有客户都拥有一个数据库，里面通常有数百甚至数千场模拟对话，他们在每次发布新版本代理前都会运行这些对话，模拟从愤怒的客户到异常情况再到背景噪音等各种场景。

So all of our clients have a database of usually hundreds or sometimes even thousands of simulated conversations that they run before every release of their agent that simulates everything from an angry customer to an unusual case to background noise.

是AI在和AI对话吗？

Is it AIs talking to AIs?

没错。

That's right.

它还能模拟不同语言、口音和背景噪音。

And it can simulate languages, accents, background noise.

这意味着在你的代理上线之前，发现所有缺陷的不是你的客户，而是你的模拟系统。

And it means that essentially before your agent goes live, your customers aren't finding all the flaws, your simulations are.

对于在听的工程师们来说，你可以把它看作是你代理的回归测试套件。

For those engineers listening, you can think of it as almost a regression test suite for your agent.

我们还使用AI监控工具。

We also use AI monitors.

那么，你如何利用AI来监控AI，以发现诸如幻觉之类的问题呢？

So how can you use AI to monitor the AI to look for things like hallucinations?

它是否说了那些并未在提供给它的数据中出现的内容？

Was it saying something that wasn't in the data presented to it?

或者更细微的问题，比如AI是否让客户感到沮丧？

Or even subtle things like, was the AI being frustrating to your customer?

它是否重复啰嗦？

Was it being repetitive?

这使你能做到，我们办公室里常说的一个笑话是：每个问题的解决方案都是更多的AI。

And it enables you to, you know, the the joke that we say in the office is the solution to every problem and AI is more AI.

这可能有其局限性，但我们真的相信这一点。

And there's probably a limit to that, but we really believe it.

而有趣的是，有了AI监控工具，你可以监控所有对话。

And what's really fun about that is, you know, with AI monitors, you can monitor all conversations.

因此，你可以远远超越运营团队的规模。

So you can go well beyond the scale of your operations teams.

即使在你确实需要人工查看对话的情况下，你也可以把问题线索放在 haystack 的顶部。

And even where you do want people looking at the conversations, you can put the needles at the top of the haystack.

这样，你的运营团队就只需要关注那些可能存在潜在问题的对话。

So your operations teams are just looking at the conversations that have potential issues.

从而真正形成一个测试、监控和人工审查的良性循环，我们希望这能成为一个良性循环。

And so really creating a virtuous cycle of, you know, testing, monitoring, and human review to create what we hope is a virtuous cycle.

因此，你的智能代理每天在线运行，都会比前一天更稳健。

So every day that your agent is live, it's more robust than the day before.

这并不意味着它完美无缺，但确实体现了——让我回到我职业生涯中软件演进的方式。

It doesn't mean it's perfect, but it's really you know, attention, I'll just go back to, like, the way software's evolved over my career.

大约在2000年代初，有一个Outlook蠕虫病毒席卷了几乎每个人的桌面，而那时大多数公司还没有CSO这个职位。

In roughly, I think it was, like, the early two thousands, there's that Outlook worm that took over, like, everyone's desktop and it was before there were CSOs in most companies.

从那以后，我们逐步建立了CISO这个角色。

Since then, we've developed the role of the CISO.

我们已经将软件开发生命周期正规化，这本质上是一种使软件更稳健的方法论。

We've formalized the software development life cycle, which is essentially a methodology to make software robust.

在AI代理开发中，我们同样提出了代理开发生命周期这一概念。

We've essentially in AI agent development have this idea of an agent development life cycle.

同样的理念是：不要期待完美，但通过这种方法论以及我们构建的产品和平台，你可以打造一个稳健的代理，使其可靠，并随着时间推移变得更加可信。

The same idea which is don't expect perfection, but with this methodology and the product and the platform we built, you can make a robust agent and you can make it reliable and you can make it something that becomes more trustworthy over time.

那么你的意思是，这些AI代理和自动驾驶汽车不一样？

Is is what you're saying then that with these AI agents, it's not like self driving cars.

对吧？

Right?

比如自动驾驶，公司很快就达到了95%的水平，但要从95%提升到100%却非常困难，即使看起来取得了进展，我们至今仍未全面部署。

So self driving for instance was it was, I don't want say trivial, but the company's got it to 95% pretty fast and getting from 95 to a 100 has been very difficult and we still don't have it rolled out even though it seems like there's been progress that's been made.

但如果你在自动驾驶中犯一个错误，后果就是灾难性的。

But if you make one mistake there, you know, it's it's catastrophic.

而你认为，对于AI代理和你所提供的解决方案，人们是否更愿意接受一些错误？因为你的基本前提是这些有缺陷的人类工作者，如果你的AI能少犯几个错误，你实际上已经做得更好了？

Whereas do you think that there's much more of a willingness to allow some errors with AI agents and the type of solution that you provide because your your baseline is these fallible human workers and if you're if you make a couple fewer mistakes with AI, you're you're still doing a better job?

我不确定这个问题有没有非黑即白的答案。

I'm not sure there's a black and white answer to that.

我的意思是，Sierra 能成为有史以来增长最快的企业软件公司之一，是因为这项技术已经适用于许多公司和许多应用场景。

I mean, the reason why Sierra has become one of the fastest growing enterprise software companies of all time is because the technology is ready for many companies and many applications.

但这并不意味着它适用于所有情况。

But it doesn't mean it's ready for all.

自动驾驶汽车的挑战在于，出错的后果可能是人员受伤。

The challenge with self driving cars is that the consequences of being wrong can be human injury.

因此，它被要求达到非常非常高的标准，这是合理的。

It's rightfully sort of held to a very, very high standard.

我可能会说，它的标准甚至高于人类驾驶员，因为

I might argue held to a higher standard than human drivers which

这确实是这样

And it's for sure it

是的。

is.

如果我们能让监管机构允许它更早地推广，我们可能会更安全，但这是另一个完全不同的议题。

It's I I we might be safer if if regulators allowed it to roll out sooner, but it's a whole separate discussion.

我们在Sierra开发的技术有趣之处在于，你不必将它用于所有应用场景；但如果你想想找回密码、查找订单，甚至找到你健康保险网络中的医疗提供者，AI代理现在就能做到这些。

What's interesting about the technology that we make at Sera is it's not, you don't have to use it for all applications, but if you think about recovering your password or finding your order or even finding a healthcare provider in the network of your health insurance company, AI agents can do that now.

这就是为什么我们正被全球最大的医疗、金融服务、电信和消费类公司所青睐。

And that's what our, you know, that's why we're being pulled by the largest healthcare financial services, telecommunications and consumer companies in the world.

这并不意味着它现在已准备好应对所有情况，但这也正是我们存在的原因。

It doesn't mean it's ready for everything yet, but that's why we exist.

如果你想象一下三年或四年后的我们，我会告诉你，越来越多的关键任务应用将变得适合使用这项技术。

You know, if we were sitting here three or four years from now, I'll tell you the more and more mission critical applications that will be ready for it.

回到我们对话的起点——氛围编程，现在用氛围编程写你的博客可能非常安全。

And just take going back to where we started our conversation of vibe coding, it's probably very safe for you to vibe code your blog right now.

你现在应该用氛围编程来编写银行的登录系统吗？

Should you vibe code the login system for a bank right now?

我可能不会建议你这么做。

I probably wouldn't recommend it.

除非

Not if

你还想继续经营下去。

you wanna stay in business.

是的，没错。

Yeah, exactly.

我的意思是，所以它并不——你能问问氛围编程是否已准备好投入生产了吗？

I mean, it's that's so it doesn't and could you ask is vibe coding ready for production?

什么生产应用？

What production application?

我认为我们现在正处于技术尚不成熟的创新阶段。

And I think that's we're at that stage of innovation right now where the technology is immature.

没有人声称它是完美的。

No one's claiming it's perfect.

但认为应该等到它完美无缺才采取行动的想法，对那些持有这种心态的公司来说无异于死刑判决。

But the idea that you should wait until it's perfect to do anything, think is a death sentence for companies who have that that mentality.

我认为更好的问题是，我公司中哪些流程已经准备好适应当前的技术？

I think the better question is what processes at my company are ready for the current set of technologies?

顺便说一句，重要的是要认识到，问题不在于它是否会出错，而在于它出错时，你如何发现？

Recognizing, by the way, that it's not a question of if it will make a mistake, but when it does, how do you detect it?

你如何减轻影响？

How do you mitigate it?

你所建立的所有控制措施。

All the controls that you have in place.

我会主张，实际上对于大多数流程和大多数公司来说，这项技术如今已经准备就绪。

And I would make the argument that actually for most processes and most companies, the technology is ready today.

换一种说法，如果我们暂停在基础模型层的创新，仅凭当前技术，我认为经济中仍存在数万亿美元的价值。

I put another way, if we paused innovation at the foundation model layer, we would still have, I think, trillions of dollars of value in the economy with current technology.

好的。

Okay.

我还有太多问题想问。

I want I have so many more questions to ask.

我们还剩二十分钟，所以先休息一下，休息完马上回来。

We have twenty minutes left, so let's take a break and and we're gonna come back right after this.

精明理财、金融天才、货币魔术师。

Fiscally responsible, financial geniuses, monetary magicians.

这些是人们在谈到那些把车险转投Progressive并节省数百美元的司机时说的话。

These are things people say about drivers who switch their car insurance to Progressive and save hundreds.

因为Progressive对一次性付清保费、拥有房产等提供折扣。

Because Progressive offers discounts for paying in full, owning a home, and more.

此外，当您需要帮助时，可以信赖他们出色的客户服务，让您的每一分钱都花得更值。

Plus, you can count on their great customer service to help when you need it, so your dollar goes a long way.

访问progressive.com，看看您能否在车险上省钱。

Visit progressive.com to see if you could save on car insurance.

Progressive意外保险公司及其关联公司，潜在节省金额因情况而异，并非在所有州或情况下都适用。

Progressive Casualty Insurance Company and affiliates, potential savings will vary, not available in all states or situations.

我们现在与布雷特·泰勒在一起。

And we're here with Brett Taylor.

他是Sierra的首席执行官，也是OpenAI的董事长。

He's the CEO of Sierra, chair of OpenAI.

布雷特，在休息前你提到一件有趣的事，我觉得Cigna在短短几个月内就和你们合作上线了。

Interesting thing you said before the break, Brett, that Cigna, I think, rolled out in what, a couple of months with you guys.

有一整套咨询行业专门负责系统集成，基本上是在连接业务流程，通常需要一年时间，而且效率往往只有50%左右。

There is a whole industry of consultants that integrate systems and are, you know, basically connecting business processes and they take a year and they usually do it like 50% effective.

在这个世界里，咨询师会面临怎样的命运？

What's gonna happen to consultants in this world?

咨询师是一个广泛的类别。

Consultants is a broad category.

你有战略管理咨询师，有系统集成商，还有那些替你开发软件的外包公司。

You know, you have strategic management consultants, you have systems integrators, you have, you know, outsourcing firms, you know, that essentially can like build software on your behalf.

我认为这些不同类型的咨询机构都会受到这项技术的不同影响。

I think all of them will have different impacts on this technology.

我从第一性原理出发的观点是，软件工程代理正在降低软件开发的成本。

The first principles view I have is software engineering agents are bringing down the cost of developing software.

不会降到零，但会从极其昂贵变得相对便宜。

It's not gonna go to zero, but it's, you know, it's going from extremely expensive to relatively inexpensive.

这是一个巨大的变化。

And that's a huge change.

因此，过去公司外包软件开发是为了降低成本，而AI代理的成本可能比外包资源还低——虽然‘资源’这个词有点去人性化，但业界常这么用——这将颠覆这一部分行业。

And so where you had a dynamic where a company was outsourcing software development to save costs, The cost of the AI agent might be less than the cost of the outsourced, I'll say resource even so it's somewhat a dehumanizing term, but often the term used in the industry, and that will disrupt that part of the industry.

但每一个技术项目背后都是一场业务转型。

But behind every technology project is a business transformation.

如果你正在数字化业务的某个部分，软件只是实现目标的手段，而不是目标本身。

You know, if you're digitizing part of your business, the software is a means to the end, not the end to itself.

我认为大多数顾问都会说，他们参与的变革管理、关于如何实施和保持竞争力的实际建议，一直以来都是他们工作中更有价值的部分。

And I think most consultants would say the change management that they participate in, the actual advice on how to do that and how to be competitive has always been the more valuable part of what they do.

所以我认为，许多咨询公司可能需要从按小时计费的动手编码工作，转向变革管理和战略咨询。

So I think a lot of these consulting firms will probably need to transition away from the billable hours of the hands on the keyboard and more towards change management, strategic consulting.

我认为这将具有某种内在价值。

And I think that will have some innate value.

我认为从外部视角来看具有固有的价值。

I think the having an outside in perspective is inherently valuable.

对于管理团队来说，我认为这种视角不会消失。

Think for a management team, I don't think it's gonna go away.

但关键是，你为他们购买的是什么，以及他们如何提供这些服务，很可能会发生很大变化。

But, you know, what is it that you're buying for them and how they provide it, think probably will change a lot.

顺便说一下，正如我们一开始提到的，软件行业也在发生巨大变化。

And by the way, as we also started, the software industry is also changing a lot.

因此，目前还有很多未知数。

And so there's a lot of unknowns right now.

但我认为，随着软件生产因人工智能而日益商品化，咨询行业将更倾向于高杠杆型的服务。

But I think it's gonna skew towards higher and higher leverage types of consulting, given the commoditization of producing software, is obviously happening with AI.

好的。

Okay.

所以你是OpenAI的主席。

So you're the chair of OpenAI.

我想聊聊OpenAI以及其背后的基础技术。

I wanna talk a little bit about OpenAI and also like the broader foundational technology underneath.

首先，他们刚刚决定要开始投放广告。

First of all, they just decided that they're gonna start showing ads.

我觉得看到这个消息后最有趣的一条推文是：AGI还远得很，因为如果OpenAI都觉得必须靠给你放广告来赚钱，而不是重新定义世界运作方式，那还得等很久。

I think the funniest tweet that I saw in response to this was like AGI is nowhere close because if OpenAI thinks that they have to show you ads as opposed to like, you know, redevelop the way the world works, then it's gonna be a while.

你对此有什么看法？

What's your response to that?

如果你看一下OpenAI的商业模式，大致有三个支柱。

So if you look at OpenAI's business model, kind of three pillars.

第一个是ChatGPT。

The first is ChatGPT.

目前通过订阅来盈利，这正是Fiji宣布广告原则的地方，计划在这一块引入一些盈利方式。

Monetize through subscriptions today and this is where Fiji announced the ads principles where the plans are to introduce some monetization component to that.

然后是API，企业用它来基于这些模型构建应用程序。

Then you have the API, which companies are using to essentially build applications on top of these models.

然后是智能体，其中最有意义的可能是Codex，这是一个软件工程智能体。

And then you have agents, which a lot probably the most meaningful of that is Codex, which is a software engineering agent.

我对发展所有这些商业模式感到兴奋，因为正如许多人所提到的，训练这些模型和推理的成本非常高。

I am excited about developing all of those business models because, I mean, as has been covered by many people, the cost of training these models and the cost of inference is incredibly high.

因此，为了真正实现OpenAI的使命——确保人工通用智能惠及人类，我们需要一个可持续的商业模式。

And so to really make OpenAI sustainable towards our mission to ensure that artificial artificial general intelligence benefits humanity, we need a sustainable business.

这实际上意味着，我们要对我们的核心智能资产——这些模型——进行商业化。

And that really means, you know, monetizing the incredibly intelligent asset that we're producing, which is these models.

我也曾在多家公司工作过，包括谷歌和Facebook，它们都成功开发了广告平台。

I also having been at a number of companies, both Google and Facebook that effectively developed ads platforms.

我认为这与这些免费服务是相辅相成的。

I think I think it's quite complimentary to these free offerings.

是的。

I yeah.

我在谷歌时参与了AdWords的推出，互联网在 monetization 之前和之后的对比非常明显：在广告介入之后，情况明显更好了，因为广告与你的搜索内容相辅相成，实际上为用户带来了大量价值。

I was at Google when we launched AdWords and the before and after of the monetization for the Internet was it was strictly better on the other side of it because the ads were complemented what you search for and it actually, you know, provided actually a lot of value to consumers.

所以我认为这始终是一门艺术，因为你无法简单地用公式来衡量信任。

So I think there's always an art form because you just can't reduce trust.

你需要一个聊天GBGDB代理代表你行动。

You need the chat GBGDB agent acting on your behalf.

我相信团队能够以一种补充而非竞争的方式来提升这种体验。

And I'm confident the team can do it in a way that, you know, complements that experience rather than it competes with it.

我对这个机会感到兴奋。

And I'm excited for the the opportunity.

但有人认为，如果彩虹的另一端真的存在巨大的经济价值，那么广告就变得不必要了？

But the argument that if there were real great economic value on the other side of the rainbow that their ads would be unnecessary?

我不认同这种观点，因为这并不清晰，你知道，OpenAI开发了这项技术，并不意味着我们就能垄断所有给经济带来的好处。

I don't buy that just because it's unclear like, you know, does just because OpenAI is making this technology doesn't mean we're like monetizing all the benefits to the economy either.

我刚刚读到，不知道这是否属实，但我在一篇报道中看到，一位数学家使用GPT-5.2证明了一个长期未解的猜想。

And so, you know, I was just reading, I don't know if this is true, but I just was reading on acts that a mathematician proved one of the unproven conjectures using g p t 5.2.

我希望这是真的。

I hope it's true.

我本人还没有核实过。

I haven't personally verified it.

这太棒了。

That's that's amazing.

这是人类的进步，但并不意味着你们就该给我们佣金。

That is progress for humanity, but it's not like, you know, send us the commission.

我认为这件事并没有什么佣金。

I don't think there was a commission on it.

所以我们的使命没有改变。

And so I the mission hasn't changed.

我认为这只是一个方式，让你知道，推动OpenAI业务增长，以便我们能继续资助我认为是世界上最好的研究实验室。

And I think this is just a way of, you know, growing the OpenAI business so we can continue to finance what I think is the greatest research lab in the world.

好的。

Okay.

让我们谈谈这项研究。

And let's talk about that research.

所以OpenAI花了很多钱。

So OpenAI spending a lot of money.

我们就不讨论投资回报率了。

We won't get into the ROI discussion.

我觉得这个话题我们在这档节目里已经聊过很多次了。

I feel like that's been, we've had that on this show a lot.

但我还想聊聊，为什么有人认为这项技术可能会开始放缓。

But I want to talk a little bit about what the what the argument would be for this technology starting to slow down.

一年前，人们还在讨论模型是否会遇到瓶颈。

A year ago, was a discussion about whether the models were gonna hit a wall.

显然它们并没有遇到瓶颈，但你可以说它们是通过一些技巧在进步。

Clearly they haven't hit a wall, but you could argue that they're getting better through through tricks.

例如，模型可以通过上网搜索来学习某些东西，但这些知识并不会被整合到模型的核心中。

For instance, a model can learn something by going to the Internet and searching for it, but that doesn't get baked into the core of the model.

所以它们是借助这些技巧，或者有些人称之为 scaffolding（脚手架）或编排，来实现大部分改进，而底层技术并没有变得快很多。

So it's using these tricks or some people call it scaffolding orchestration to get most of these improvements and it's not like the underlying technology is getting that much faster.

这就是论点。

That's the argument.

所以，也许这些技巧最终会用尽。

And so then maybe eventually the tricks will run out.

你怎么看？

What do you think about that?

首先，我认为如果你从事科学工作并希望取得进展，那么所有的批评都值得倾听。

First of all, I think all criticism is important to listen to if you're in science and you want to make progress.

我的看法是，首先，过去一年我们所看到的进展中，有一个微妙的方面是模型变得更加聪明了。

My take, first of all, I think there's a one of the nuanced aspects of the progress we've seen particularly this past year is the models have gotten a lot more intelligent.

它们可能早在一年前就已经足够智能，足以应对大多数消费类应用，比如我计划去哥本哈根的旅行。

They may have been sufficiently intelligent like a year ago for most consumer applications like my trip planning to Copenhagen.

我不确定GPT-5.2的推理能力会对那次旅行规划带来多少提升。

I'm not sure how much the reasoning capabilities of GPT five two would have benefited that trip planning.

我们当时已经达到了旅行代理的巅峰。

We were at peak travel agent.

我这里有点开玩笑的意思。

I'm being a little facetious here.

如果你是用某个使用场景的话。

If you're using use case.

是的。

Yeah.

但如果你用编码器来帮你写软件，GPT-5和GPT-5.2之间的区别就巨大了。

Well, but if you're using codecs to write software for you, the difference between GPT five and GPT 5.2 was huge.

你可以在网上看到这一点，所有开发者都在使用它，因为推理能力对于编写复杂软件来说极其宝贵。

And you can see this online just all the developers using it because reasoning capabilities for authoring complex software is incredibly valuable.

所以我认为，一个正在显现的动态是，特别是对于那些使用ChatGPT和Gemini进行日常休闲使用的人——我们大多数人目前都是如此——许多模型的改进对于这类应用来说并不明显，但如果你用它来开发软件或证明一个未被证实的数学猜想，这些改进就极其显著。

So I think one of the dynamics that's sort of playing out particularly for people using ChatGPT and Gemini for I'll say casual everyday use which most of us are at this point is a lot of the model improvements are are not necessarily visible for those class of applications, but incredibly visible if you're using it to say develop software or to prove an unproven math conjecture.

所以这是一个有趣的动态，我认为在未来几年里这种趋势会加剧：对于特定任务而言，去年的模型已经足够，而新模型则需要用来实现某种形式的超级智能或通用人工智能。

So that's an interesting dynamic, which is and I think it's gonna probably over the next few years amplify, which is for a given task, the models from last year are sufficient and the new models will be necessary to achieve some semblance of super intelligence or artificial general intelligence.

但我们实际上已经达到了许多不同应用所需的足够智能水平。

But we're actually at sufficient intelligence for a lot of different applications.

我认为，这正是我们在评估进展、成本等各种因素时，必须从应用的角度来审视的原因。

And I think that's why it's really important when we're judging progress and cost and all these things that we'll actually probably have to start looking at through the lens of applications.

如果你在进行药物疗法发现，你可能会非常重视推理能力；但如果你只是计划稍微放松一点，那可能就没那么重要了。

If you're doing pharmaceutical, like therapy discovery, you probably care a lot about the reasoning capabilities if you're planning to trip a little less so.

关于使用工具，你的观点，我实际上认为这完全是积极的。

To your point on accessing tools, I actually think this is strictly a positive.

我认为，AI代理的一个重大突破将是长期任务，而编写代码、搜索互联网，我认为这非常好。

I think that one of the big breakthroughs with AI agents will be long running tasks and and I think that using writing code, searching the Internet, I think is great.

我认为，无论你的训练过程如何，互联网都是每秒都在变化的。

I think it enables whenever you whether or not whatever your training process is, the Internet changes on a second by second basis.

因此，拥有一个能够应对这种变化的AI代理非常重要。

So having an AI agent that can respond to that is very important.

但更重要的是，它可以访问私有数据库，可以访问，你知道的，结构化的系统，比如正在分析特定DNA链的系统，无论你想做什么。

But more importantly, it can access a private database, it can access, you know, a structured, you know, can access a system that's looking at a particular strand of DNA, whatever you might wanna do.

所以，Twilio至关重要。

So, Twilio is incredibly important.

我认为，在人工智能领域，要实现真正的通用人工智能，可能需要基于这个模型所做观察的强化学习，而大多数主流模型并没有这样做，尽管我并不是一名人工智能研究人员。

Probably the the one thing that I think in AI circles and and I'm not an AI researcher though is to achieve true AGI, will we need reinforcement learning from the observations that this model makes which most of the mainstream models don't do?

我对这一点不太确定，关于这个问题存在着非常健康的争论。

I'm not sure about that and there's a really healthy debate about that.

但我并不把这种工具使用视为一种权宜之计。

But I I am I I don't characterize this tool use as like a hack.

我认为这实际上是长期运行智能体的一种结构性输入，而这种智能体可能是我们实现接近通用人工智能所必需的。

I think it's actually a structural input to long running agents, which will probably be what we need to to to create something close to AGI.

好的。

Okay.

所以我们已经聊了一点你的个人经历。

So we've talked a little bit about your personal history.

我想最后用一个快速问答环节来谈谈你曾合作过的科技界一些领袖人物。

I just want to end here with a bit of a lightning round where we can talk about some of the leaders in the tech world that you've worked with.

他们都是我们的观众会非常熟悉的大人物。

They're gonna be big names that a lot of our audience will know.

从每个人身上，给我们讲一点你学到的东西。

Just give us one thing that you've learned from each.

我想先从马克·贝尼奥夫开始。

And I wanna start with Marc Benioff.

马克在围绕公司打造生态系统和社区方面非常出色。

Marc was amazing at creating an ecosystem and a community around a company.

我第一次参加Dreamforce时深受启发，看到那么多人到场，我意识到拥有客户和拥有社区是不同的，这一点我深有体会。

I was really inspired by my first Dreamforce and just seeing all the people that showed up there and realizing there's a difference between having customers and having a community and some that stuff is something I took away

从他身上学到的。

from him.

好的。

Alright.

马克·扎克伯格。

Mark Zuckerberg.

马克·扎克伯格可能是我合作过的时间最长、最具长远眼光的人。

Mark Zuckerberg was probably the longest longest term thinker I ever worked with.

我们过去经常在帕洛阿尔托散步，长时间讨论战略。

We used to go on long walks around Palo Alto talking about strategy.

每次我以为自己在做长期思考时，他想的却是我的两倍长远。

And every time I thought I was thinking long term, he was thinking about two times longer than me.

我现在也试着效仿这一点。

And I've tried to model that now.

我会想，我是否足够长远地思考了？

It's like, how am I thinking long term enough?

我想你从过去十年Facebook和Meta的表现中就能看出来。

And I think you can see it in just Facebook and Meta's performance over the past decade.

他总是在眺望地平线之外的地平线，这让我深深钦佩。

Like he's always looking at the horizon beyond the horizon, which I deeply admire.

说到扎克伯格的长期思维，你认为他们现在对人工智能的这个重大押注怎么样？

Speaking of Zuckerberg's long term thinking, do you think this big bet that they're making on artificial intelligence, right?

我的意思是，他们已经从你担任董事会主席的公司挖走了许多顶尖工程师。

I mean, they've poached many of the top engineers from the company that you're the board chair of.

你认为这是他的信念吗？即我们会停止与人类朋友在线交流，转而开始与数字朋友交流？

Do you think that's his belief that we are gonna stop communicating with our human friends online and we'll start communicating with our digital friends instead?

我认为我们会很长时间继续与朋友交流。

I think we're gonna be communicating with our friends for a long time.

我认为这只是说明马克很有信念，愿意言行一致，我认为这对一位首席执行官来说是非常值得钦佩的品质。

I think this is just mean, Mark has a lot of conviction and is willing to put his money where his mouth is and I think that's a real admirable trait for a CEO.

所以我认为这仅仅就是如此。

So I think that's all it is.

我不认为这是对人际关系的指责。

I don't think it's a indictment on human relationships.

哦，好吧。

Oh, okay.

我会持相反的观点。

I will take the other side of that.

不。

No.

真的吗？

Really?

我觉得

I think

我们很多人其实都有这种感觉，我的意思是，这不是很有趣吗？

we are we are many of us are I mean, we have this like, isn't it interesting?

我们正面临孤独危机，而这些机器人正进入这片虚空，告诉你你有多棒。

We have this loneliness crisis and into the void is coming these bots that will tell you how great you are.

记住关于你的一切。

Remember everything about you.

为你最好的利益着想。

Look out for your best interests.

我确实担心这种阿谀奉承的人工智能。

I actually am worried about sycophantic AI.

我觉得有些科学研究虽然部分被推翻了，但我很认同‘焦虑的一代’这个说法，它让我深刻感受到智能手机和社交媒体，尤其是对年轻人的负面影响。

I think it is some of the science, I guess, has been partially debunked, but I I I like the anxious generation and it resonated with me just the negative impact of smartphones and social media, particularly on young people.

我认为，对于任何这些新技术，都存在成瘾的风险，尤其是某些人工智能代理那种阿谀奉承的特性。

And I think with any of these new technologies, there's a risk of addictiveness and and, you know, particularly the sort of sycophantic nature of some of these agents.

但我是个乐观主义者。

But I'm an optimist.

我认为，总体而言，技术推动了社会进步，能让我们摆脱重复性劳动和琐碎任务，帮助我们实现自我价值。

I think technology broadly moves society forward and can unburden us from, you know, repetitive and meal tasks and enable us to be more self actualized.

所以我认为关注这个问题很重要，但认为它是生存性风险的想法是错误的。

So I do think it's important to worry about, but I also think it's wrong to think it's an existential risk.

我认为我们应该关注它，减轻这些风险，明智地引导我们的孩子，特别是初中和高中阶段的孩子，控制他们接触技术的时间。

And I think we should worry about it, mitigate those risks, be smart with our kids and, you know, particularly kids in middle school and secondary school about when they get access to technology.

但我对这些技术带来的好处非常乐观。

But I'm really optimistic for the benefits.

萨姆·阿尔特曼。

Sam Altman.

你作为OpenAI的董事会主席，曾与他共事过一段时间。

You've spent some time with him as the board chair of OpenAI.

他是怎么运作的？

How how does he operate?

你从他身上学到了什么？

What have you learned from him?

萨姆可能是我合作过的所有创始人中愿景最宏大的，他的超能力是让众人认同并追随他的愿景。

Sam probably has the most ambitious vision of any founder that I've worked with, and his superpower is aligning people to that vision.

我有幸没有真正了解过史蒂夫·乔布斯，但人们总提到他的现实扭曲力场，为什么那么多优秀的工程师会去为他工作，创造出像Macintosh或iPod这样的伟大产品。

I was didn't have the good fortune of really knowing Steve Jobs, but they always talked about the reality distortion field and why so many great engineers went and did great things like creating the Macintosh or creating the iPod.

我在萨姆身上看到了很多类似的特质。

And I see a lot of that in Sam.

他对OpenAI的宏伟计划是什么？

And what is what is his grand plan for OpenAI?

那里的长期潜力在哪里？

What is the long term potential there?

构建通用人工智能，并确保它造福人类。

To build AGI and ensure that it benefits humanity.

这一直是基金会的使命，至今仍是。

And it's always been the mission of the foundation and it still is.

但这句话掩盖了实现这一目标所面临的诸多挑战。

And But in that sentence obfuscates a lot of the challenges of doing so.

仅看看实现这一目标所需的资本投入，这一点在基金会成立时几乎没人预料到。

Just look at the even the capital requirements to do so which I don't think anyone knew when it was founded.

所以我认为，萨姆现在面临的难题是，将这一愿景转化为一个为期十年的实施计划，我认为这是人类历史上最了不起的技术成就之一，能参与其中令人振奋。

So I think the hard part Sam's doing now is taking that vision and mapping out a decade long plan to get there, which is I think one of the most remarkable technical achievements in human history and it's exciting to be a part of it.

那个业务和萨姆所面临的挑战中，有一个有趣的动态是，似乎每个人都能迅速赶上。

One interesting dynamic of that business and Sam's challenges, it does seem that everybody just catches up real quick.

你说得完全正确。

You know, it's absolutely right.

我的意思是，我现在看到的这一切，某种程度上正让我回到大学毕业时的起点。

I mean, this is what I see going on right now sort of maybe ending where we started when I graduated from university.

我在斯坦福读书时经历了互联网泡沫的兴起，随后在本科期间泡沫破裂了。

I was in Stanford when the dot com bubble happened and then it burst in the middle of my undergraduate education.

如果你回顾互联网的那个时期，大多数人都知道互联网将产生巨大影响，甚至大多数人也知道像电子商务和搜索这样的关键应用将会带来重大影响。

If you look at that period of the Internet, most people knew that the Internet was going to be impactful and most people even knew the key applications like ecommerce and search that would be impactful.

而这是一场残酷的竞争，决定谁将在这些市场中胜出。

And it was this cutthroat competition to decide who would win in those markets.

如果你还记得AltaVista，它曾经是排名第一的

If you remember AltaVista, which sort of the number one

那是搜索领域的一大用户。

That was search a big user.

是的。

Yeah.

对。

Yeah.

然后你有，比如，雅虎和Excite作为门户网站，还有

And then you had, know, Yahoo and Excite as the portals and you had on

Lycos。

lycos.

Buy.com，Lycos。

Buy.com, lycos.

你有Buy.com、亚马逊，每个国家都有自己的平台。

You had buy.com and Amazon and every country had their own.

我认为我们现在正处于类似的状态。

I think we're just in a similar state right now.

你不需要拥有人工智能博士学位，就能意识到这将对经济和社会产生巨大影响。

You don't need to be a have a PhD in artificial intelligence to think and say, wow, this is gonna have a big impact on the economy and society.

因此，全球所有的资本和聪明人都聚焦在同一件事上。

So you have all of the capital, all the smart people in the world all focused on the same thing.

因此，这种程度的竞争，在我看来是完全可预期的。

So this degree of competition is completely, in my opinion, expected.

这意味着对我们这些身处其中的人来说，压力巨大。

It means it's super stressful for those of us in the middle of it.

但这对世界来说是好事。

It's great for the world though.

我的意思是，这推动了创新，降低了成本。

I mean, drives innovation, it lowers costs.

所以我觉得这是一件非常好的事。

So I think it's just a great thing.

对于我们这些身处其中的人来说，睡眠很少，因为每天醒来都会遇到新的竞争对手。

For those of us in the middle, we don't get a lot of sleep because every day you wake up and there's new competition.

但自由市场科技经济的另一面正是如此，我认为我们将会拥有一些造福人类的惊人工具，我很高兴能参与其中。

But that's what's great about the like free market technology economy is like the other side of this is I think we're gonna have some amazing tools that benefit humanity and I'm excited to be a part of it.

但许多早期的先驱者都失败了。

But a lot of those early pioneers fell off.

百分之百。

A 100%.

这就是运作的方式。

That's just the way it works.

我认为会有一段整合期。

I think there'll be a period of consolidation.

嗯，这也是我的笑话。

Well, that's also my joke.

如果你全身心投入Buy.com而不是亚马逊，那么你对互联网泡沫的看法会大不相同。所以，我认为一些公司会倒闭，更可能的是被兼并。

Your perspective on the .com bubble is very different if you went all in on buy.com versus amazon And dot so you're gonna have some companies go out of business, I think more likely sort of be consolidated.

我觉得可能没有太多资本，但这些资本被盲目地投向了各个领域。

I think there's probably I don't see too much capital, but it's sort of been applied blindly to categories.

这很泡沫化。

It's bubbly.

是的。

Yeah.

这绝对是泡沫，毫无疑问。

It's definitely oh, it's absolutely a bubble.

但我认为，与此同时，确实有具有代际意义的公司正在诞生。

But I think it doesn't mean that there's not truly generational companies being created at the same time.

我认为这两者同时成立，也正因如此，风险投资不是给胆小者准备的。

And I think both are true at the same time and, you know, that's why, you know, venture capitalism for the weak of heart.

你会看到有些人表现得很差，而有些人则会写下他们的传奇故事，讲述自己做出的那些重大押注——这就是我成长的世界，我觉得这很令人兴奋，尽管有时会有点压力。

And you're gonna see some people do very poorly and some people writing their proverbial book about the the great bets they made and, you know, that's the world I grew up in and I think it's exciting though, mildly stressful at times.

是的。

Yeah.

那玛丽莎·梅耶尔呢？

What about Marissa Meyer?

玛丽莎是我的第一位老板。

Marissa was my first boss.

我大概明白了招聘时人才的重要性。

I probably learned the importance of people in hiring.

我是通过谷歌的一个项目加入的，叫做助理产品经理项目，她招聘的是来自技术背景的应届毕业生，这些人希望成为产品经理。

I came in through a program made at Google called the associate product manager program where she hired new grads out of technical degrees and who wanted to become product managers.

她说，与其去读MBA，不如在谷歌读MBA。

And she said, rather than getting an MBA, get an MBA at Google.

嗯。

Mhmm.

她招募的很多年轻人后来都负责了谷歌的重要部门，现在甚至经营着自己的公司。

And a lot of the young people she recruited there ended up running big parts of Google, running companies now.

但她花了大量时间来甄选谷歌的人才。

But she spent just a lot of her time curating the people at Google.

我认为这是她对这家公司最伟大的贡献之一。

Think it was one of her greatest contributions to that company.

我总是提醒自己，这就像吃蔬菜一样。

And I always remind myself, you know, it's like eating your vegetables.

如果你想让公司两年后变得卓越，那么现在专注于引进的人才，可能是最明智的做法。

If you want your company to be great two years from now, focusing on the people you're bringing in now is probably the smartest thing.

每当我这么做时，我总会想到她。

I always think about her when I do that.

谢丽尔·桑德伯格？

Sheryl Sandberg?

谢丽尔，这听起来可能有点奇怪。

Sheryl, this is gonna sound funny.

谢丽尔总是给我最严厉但也最好的反馈。

Sheryl always gave me the, like, harshest best feedback.

比如，他当时就说：‘哇哦。’

Like, he was like, woah.

是的。

Yeah.

嗯，我们这里并没有直奔主题，对吧？

Well, we're not cut we're not, like, cutting to the chase here, are we?

我意识到，我的职业生涯中很少真正得到过反馈。

And I realized that so much of my career hadn't been really given feedback.

如果有人真的关心你，他们愿意告诉你真相，而不是你爱听的话，这是一种馈赠。

And if someone actually cares about you, their willingness to actually tell you what you need to hear, not what you wanna hear is a gift.

所以我觉得给予反馈很难，因为你总想着它会让对方有什么感受，而不是想着你希望他们未来在事业上变得更好。

And so I think it's very hard to give feedback because you think about the way it will make someone feel rather than thinking about you want them to be better at their career in the future.

我从她身上学到了这一点，她太棒了。

So I learned that from her and she's amazing.

我认为她之所以成为硅谷一半人的导师，是有原因的。

I think she's a mentor to like half of Silicon Valley for a reason.

我们把拉里和谢尔盖放在一起吗？

Do we put Larry and Sergei together?

我觉得人们总是说‘拉里·谢尔盖’。

I feel like people just say Larry Sergey.

我们会把拉里和谢尔盖放在一起。

We'll put Larry and Sergey together.

对我来说，拉里始终以一种非凡的方式关注长期的技术方向。

Larry for me always focused on long term technology direction in a way that was remarkable.

当我刚到谷歌时，我完全不明白为什么我们要自己建造数据中心，后来才发现这竟是谷歌服务成本中极其重要的一部分。

When I got to Google, I had no idea why we're building our own data centers, and it was it turned out to be an incredibly important part of the cost to serve of Google.

当我们开发从谷歌地图到App Engine，再到谷歌云的过程中，他专注于从架构上打造规模上的不公平优势，这非常了不起。

When we were making everything from Google Maps to App Engine, became Google Cloud, his focus on, like, setting it up architecturally to have sort of unfair advantages of scale and and it was remarkable.

这是一种极其长远的技术视角。

Just incredibly long term technical view.

我把它写下来了。

I have this written down.

我只是想问你一下。

I'm just gonna ask you.

你和他互动之后，对埃隆·马斯克有什么看法？

What's your opinion of Elon Musk after your interactions with him?

可能是我们这个时代最伟大的企业家。

Probably the greatest entrepreneur of our time.

我的意思是，他本身就像一家公司。

I mean, he's a company.

他创造了从SpaceX到Starlink的巨大影响，更不用说特斯拉和X了。

He's created everything from SpaceX and look at the impact of Starlink, let alone Tesla and X.

虽然我们之间有些复杂的互动，但在这方面他无疑是无可争议的领袖。

So, you know, had some complicated interactions, but sort of the undisputed leader in that respect.

你做了个

Did a you

买下X是个不错的选择吗？

good choice buying X?

我和它没有太密切的联系。

I'm not very close to it.

自从交易完成以来，我就没有太关注它了。

So I haven't really followed it as much since since the transaction went through.

我没有强烈的意见，好吧。

I don't I don't have a strong opinion on Okay.

我学会了不问你们这样的人预测未来五年的事，但你能告诉我们未来一年会发生什么吗？

I've learned not to ask people in your position to predict like the next five years, but can you tell us what's gonna happen over the next year?

我认为我们会看到一些像数学猜想被证明这样的事情，让我的社交圈之外的社会开始意识到人工智能对科学的影响。

I think we will see a set of things like that math conjecture that's proven where society outside of the realm of my social circle starts to acknowledge the impact that AI is gonna have on science.

我认为这会以一种积极的方式改变人们对人工智能的正面看法，当我们意识到人工智能或许能帮助发现尚未治愈的疾病疗法、在物理学、清洁能源、电池存储方面取得突破时，我们将摆脱将人工智能仅仅视为聊天机器人的讨论，转而认识到人工智能是推动社会前进的力量。

And I think that will, in a good way, change the positive perception of AI when we realize this can help perhaps over time discover cures to uncured diseases, make breakthroughs in physics, clean energy, battery storage, where we'll get out of the discussion of AI as a chatbot and the discussion of AI is something that's gonna move society forward.

我非常期待那一天的到来。

I'm really looking forward to that.

很好，布雷特，谢谢你专程过来。

Well, Brett, it's great that you came down here.

这是我们第一次对话。

This is our first conversation.

我期待这一天很久了，希望这不会是我们最后一次交流。

I've been looking forward to it for a long time, and I hope it's not our last.

再次感谢你。

So thanks again.

谢谢你的邀请。

Thanks for having me.

好了，各位。

Alright, everybody.

感谢大家收听和观看，也感谢高通公司让我们在达沃斯的场地举办这次活动。

Thank you for listening and watching, and thank you to Qualcomm for having us here at your space at Davos.

我们下次再见，欢迎收听《大科技》播客。

We'll see you next time on Big Technology Podcast.

太棒了。

Awesome.

谢谢。

Thank you.

非常感谢。

So much.

太棒了。

Great stuff.

谢谢大家。

Thanks, everybody.