BG2 with Bill Gurley, Brad Gerstner & Aaron Levie | Software Valuations, Earnings, AI, Immigration & More | E02

你知道的。

You know?

但上周，Meta那边的情况，我觉得特别重要。

But last week, what we saw at a meta, and I think this is really significant.

你知道的。

You know?

很明显，他们大幅上调了业绩预期。

Obviously, there's a big beaten raise.

今年到目前为止，股价涨了超过30%，他们确实受益匪浅。

Stocks up over 30% year to date, they're clearly benefiting, right?

他们正处于人工智能的最前沿。

They are at the tip of the AI spear.

他们的视频互动量同比上涨了25%，这简直让人难以想象。

Their video engagement was up 25% year over year, which is just like hard to get your head around.

能做到这一点，只有一种可能。

There's only one way to do that.

这都是靠人工智能实现的。

That's through the application of AI.

他们的变现能力也大幅提升，你知道吧？但真正让我觉得厉害的是这一点。

We're seeing their monetization also up a lot, you know, but here's the thing that that I think is so impressive.

对吧？

Right?

他们的人均利润几乎翻了三倍。

They they nearly tripled their earnings per employee right there.

2022年他们的员工总数是八万七千人。

Their head count in '22 was 87,000.

这一季度，他们公布的数字是六万七千人。

This quarter, they reported 67,000.

他们一边这么做，一边坚信扁平化才能更快。

So they're doing this while they're like, really believe that flatter is faster.

现在很多人裁员都是减个9%、12%之类的，但那根本不是对商业模式的根本性调整。

And I think there are a lot of people who are cutting 9% or 12% or this or that, you know, and it's not a fundamental change to the ethos of the business.

这不过是多做点自保罢了，你知道的。我觉得马克身上体现的，是真正想把业务转型的决心，这在业绩上已经看得出来了。我之前还发过一条推文，说现实实验室还在亏，大概每年亏两百亿美金，算下来每股大概亏六块，这都是现实实验室拖累公司造成的。

It's just a little bit more CYA, you know, and I think that, you know, what you see with Mark is a real commitment to transform that business and it's showing up in, you know, in the results and, you know, I I think I tweeted something about, you know, reality labs is still losing, I think, $20,000,000,000 and if you look at that on a per share number, okay, and you that's about $6 a share, you know, in losses for the company due to reality labs.

如果你用二十倍的市盈率来算，那这部分就相当于给Meta的估值带来了每股负一百二十块的影响，原因就是他们对现实实验室的投入。

And if you apply a 20 x multiple to that, that's about negative a $120 a share that's being attributed to the valuation of meta as a result of Hold their fake on.

我不同意。

I object.

只有当你在分析里完全忽略这个部门未来的自由现金流时，这种说法才成立。

That that that's only that's only true if you put zero future free cash flow from that unit in your analysis.

是的。

Yeah.

不是这样的。

No.

那就是说

Which

也许你正是这么做的。

maybe you're doing.

我觉得大家应该都做了这样的分析，比尔，但你我都清楚，大多数卖方分析师只是给这家公司的盈利直接套了个20倍市盈率。

Well, I would like to think that everybody was doing that analysis, Bill, but you and I both know that most sell side analysts are slapping a 20x multiple on earnings for this company.

你知道的，这如今就是研究报告的常态了。

And you know, that that's what suffices for research these days.

从整体来看，我觉得这种做法简直太离谱了。

And I would just say in a consolidated view, that is pretty crazy to me.

扎克伯格在电话会议中也提到了这一点。

And Zuck even mentioned on the call

但另一方面，比尔，这也不算太离谱，毕竟你该怎么估算现实实验室真正的盈利潜力呢？

that- But also, I mean, alternatively, Bill, like it's not like it's also not crazy because how should you figure out like what the actual profit potential is of reality?

目前我们根本看不到任何迹象表明它会是什么样子。

Like, we have no evidence at the moment of what that looks like.

所以任何其他数字都只是凭空捏造的。

So everything else would just be a made up number.

所以我也说不准

So I don't know what

我不确定。

I don't know.

但不管怎样，马克在电话会议中提到，那款与雷朋合作的Meta AI眼镜，销量达到了预期的两倍。

But any non zero, you know, one of the things that Mark said during the call is that the meta AI glasses with Ray Ban, they sold twice as many as they expected to sell.

是的。

Okay.

我觉得AI设备马上就要迎来爆发了。

I think we're gonna have this explosion of AI devices.

这个我们待会儿再细说。

We're gonna talk about a little bit later.

但他指出，这算是现实实验室的成果与公司其他业务融合的第一个明确例子。

But he said that was the first clear example of some of this convergence between the work that they're doing in reality labs.

对吧？

Right?

还有他们其他业务板块的结合。

And what they're doing in the rest of the business.

所以，作为投资者，我觉得马克·扎克伯格每年投入200亿美元在AI和增强现实这类领域，哪怕只沾点边，每股价值也远超负120美元。

And so, you know, again, as an investor, I assign greater than negative $120 a share value to Mark Zuckerberg investing $20,000,000,000 a year on something that's in the zip code of AI, you know, an augmented reality.

这周另一件事，比尔，谷歌，你知道吧，财报其实不错，但今年股价却一直没涨，甚至有点跌。

The second thing we saw this week, Bill, Google, you know, I mean, it was a solid quarter, but the stock is flat to down now on the year.

我觉得大家对谷歌核心业务的担忧在增加，就是咱们之前聊过的那些威胁，比如ChatGPT、Perplexity这些。

I think you're seeing increasing concern about the core business, you know, from threats you and I've talked about this chat GPT perplexity.

你随便列一列就知道了。

Just go through the list.

你走进任何一个房间，都能感觉到。

You go into any room.

很明显，问答型引擎正变得越来越重要。

Clearly, answer engines are becoming more relevant.

你之前说，ChatGPT都能做出那个图表了。

Well, you know, when you said, you know, chat GPT could have made the chart.

你可没说Bard或者谷歌能做出那个图表。

You didn't say bard could have made the chart or Google could have made the chart.

所以我觉得谷歌云的表现相当平淡。

So I would say glue Google Cloud was pretty uninspired.

我觉得他们在效率和业务聚焦上还没真正下功夫。

And I think I don't think they've gotten real on the efficiency and the fitness.

我这儿有个幻灯片。

You know, I have a slide here.

我让团队整理了这个图表，看看他们在人员规模上的变化，以及过去六个季度里，人均利润是怎么变化的。

I asked my team to pull together this slide, plots, you know, what they've done on headcount and what that translates into in terms of earnings per employee over the course of those, you know, same six quarters, right?

Facebook 大幅削减了人员，人均利润却飙升了；而谷歌呢，虽然还算稳定，但根本没怎么调整。

And so Facebook is actually reduced headcount significantly and their earnings per employee has exploded and Google kind of, you know, they're fine, but they really just haven't made many changes.

如果只看短期的人均利润，那你就能直接看出两家股价表现的差距了。

Now, if this was just about earnings per employee in the short term, then you see your differential in terms of share price performance right there.

但这里更大的问题，其实是这对未来产品开发意味着什么。

I think a much, much bigger issue here is the implications this has for future product development.

这背后反映的是企业能否真正聚焦，我觉得‘简洁胜于繁复’这句话一点不假。

The implications it has for, you know, real focus out of the business and I just think this flatter is faster is very real.

我觉得Meta非常专注，而Google还在摸索该怎么在这些事情上集中精力。

I think that meta is very focused and I think that Google is still trying to sort out, you know, you know how to get focused around some of this stuff.

最后说说云软件，经过一年的紧缩，上周我们从亚马逊和Azure听到的是，核心业务的增速已经回到以前的水平了。

And then finally, cloud software, you know, after a year of belt tightening, you know, what we really heard out of Amazon and Azure last week is that kind of core workloads are back to their prior growth rates, right?

也就是说， CFO们基本已经完成了对核心业务的瘦身。

So that CFOs are largely through the, you know, the belt tightening on the core.

我们确实看到AI工作负载开始在亚马逊和Azure那边发力了。

And we saw real evidence, right, of AI workloads beginning to kick in both at Amazon as well as at Azure.

我觉得Azure这边，我这儿有几张图表，是团队整理出来的。

I think Azure, I've got a couple of charts here that the team threw together.

第一张图你能看到，右边明显出现了回升，这是三家云服务商加在一起的数据。

The first is just you can see the reacceleration on the right hand side in terms of this is all three of the cloud providers put together.

这是它们整体的增长率。

So this is their growth aggregated growth rate.

在右边，你终于能看到这个回升的拐点了。

And on the right hand side, you can see you finally saw that turn that reacceleration.

不过要注意，AWS 这些业务是从一千亿美元的年收入基础上重新加速的，对吧？

Now mind you, these things are reaccelerating in the case of AWS from a $100,000,000,000 run rate, right?

所以你看，他们是在已经很高的基础上继续提速。

So like the, you know, you're reaccelerating.

左边这些数字，是在巨大规模上实现的惊人增长。

These are massive growth rates at massive scale on the left side.

他们上个季度在三大平台上新增的年度经常性收入达到了惊人的150亿美元，创下了新纪录。

What you see is they actually set a record quarter 15,000,000,000 in net new ARR across these three platforms.

我觉得，这才是从各大云服务商那里真正透露出来的关键信息。

So I think, you know, that to me is is, you know, kind of the real story that came out of, you know, out of the big cloud providers.

而我们现在必须回答的一个问题是，这种趋势会不会蔓延到软件行业的其他公司？

And, you know, one of the questions I think that we now have to answer, you know, is does that read through apply to everybody else in software?

对吧？

Right?

这到底是只有大云服务商才有的特殊现象，还是更普遍的趋势？

Like, you know, is this something that's unique to the big cloud providers?

然后我脑子里还想着第二个问题：这种增长到底什么时候会碰到天花板？

And then the second question that I had in my mind, you know, was like, when do we just hit the wall on this?

对吧？

Right?

前几天有个朋友跟我说，他觉得三年后云服务商的收入反而会比现在还低，你说这可能吗？

I, you know, I had a friend to say to me the other day, they thought that three years from now that the cloud providers were the revenues would actually be smaller than they are today, right?

我真没法理解这一点，因为我刚和负责AWS的亚当·萨利茨基聊过，他说布拉德，这可是个上万亿美元的市场，而且还没算上人工智能呢。

Which I which I can't get my head around when I look at this, you know, I talked to Adam salipsky runs AWS and he said, Brad, this is a multi trillion dollar market and that does not include AI.

好吧。

Okay.

所以没错，一千亿美元听起来已经够吓人了，但其实这还只是个上万亿美元市场的零头，而且还没算AI的部分。

So yeah, a $100,000,000,000 is a lot, but it's a multi trillion dollar market and it doesn't include AI.

所以艾伦，我先不打断你了。

So Aaron, I can't get back over to you.

这周公开市场的动态我就先说这么多，尤其是软件行业这边——你对这些数字感到意外吗？你觉得这种趋势是云服务商独有的，还是你已经开始听到其他软件公司也在谈论类似的变化了？

That's, you know, that's my brain dump on what happened in the public markets this week, but on the software side in particular, you know, were you surprised at these numbers and how do you think about, is this unique to what's going on with the cloud providers, you know, or are you starting, you know, to, hear other software companies talking about how about similar trends?

是的，首先呢，

Yeah, so first of all,

你真该做空你朋友的投资组合，我从来没听过这种说法——三年后市场规模反而会变小。

you might want to short your friend's portfolio do the I've never heard that metric of in three years from now, we're going to be smaller.

不过这个观点还挺有意思的。

But that's an interesting take.

我真的懒得再去猜这个增长什么时候会停了。

I honestly am so done with trying to guess when this thing runs out.

真的，就是这么回事。

Like it is.

我觉得我们得坦诚一点，这个领域的潜在市场根本没上限，也完全没法估算。

I think we just have to honest, like, you basically are just it is an uncapped, unknowable total amount of addressable market in this space.

超大规模云服务商和其他软件公司之间确实有明显区别，因为企业可以把核心数据中心从原来的托管机房和服务器供应商直接迁移到云上。

There are clear differences between the hyperscalers and the rest of software simply because a company can just move their core data center infrastructure from their current colo and server providers to the cloud.

在这个转型过程中，一些软件公司能赚到钱，但绝大部分收益其实都归了云服务商。

And, you know, some software providers will make money in that transition, but that's basically 90% hyperscaler benefit.

所以这些并不会以同样的方式反映在其他软件和SaaS领域。

And so that doesn't sort of all show up in the rest of software and SaaS in the same way.

这就像是两种截然不同的情况：一边是这些大型基础设施迁移，云服务商正面临无限增长的市场机会。

So there's some kind of tale of two cities of these massive infrastructure migration where the hyperscalers are just, you know, again, uncapped market opportunity for these guys.

与此同时，还有很多周边服务——比如你想象的，Snowflake、Confluence这些工具，也会从中受益，因为整个过程中需要大量数据服务支持。

And then there's a lot of surrounding services like you would imagine that as this goes up, snowflakes and confluence whatnot also see some benefit because there's lots of data services that you need around that.

但确实，我们没看到任何放缓的迹象，我们接触的公司都在持续向云端迁移。

But, yeah, we're not seeing any slowdown of, you know, companies that we talk to just continuing to be moving to the cloud.

我每个月都能跟几十位CIO聊天，这让我挺幸运的。

I, you know, get the fortune of talking to, I don't know, dozens of CIOs a month.

没有一个人说他们的上云之旅已经结束了。

And there's nobody that is sort of like done with their cloud journey.

甚至连90%进度的都找不到。

There's nobody even at 90%.

这在整个经济各个行业都是如此。

And this is across every sector of the economy.

所以我们其实还处在相当早期的阶段。

So we're still weirdly relatively early.

而且正如亚当提到的，那还是在人工智能出现之前的事。

And then to Adam's point, like, yeah, like, that's even pre AI.

目前在企业环境中，人工智能基本上还处于试验阶段。

AI is just in the experimentation phase, for all intents and purposes in, in the enterprise right now.

从软件基础设施的角度来看，我们甚至还没搞清楚人工智能的潜在市场规模。

And so we we don't even have the AI TAM, you know, quite understood, from a a software infrastructure standpoint yet.

布拉德说，他通过多方了解发现，CIO们已经走出了收缩和重新评估的阶段。

Brad made the comment that in his view from from checking around that CIOs had moved past their kind of shrink and reassessment.

你同意这个看法吗？

Do you do you agree with that?

我不希望别人拿我这些个人观察当投资依据。

I I wouldn't want anybody to, like, trade on my on my, like, qualitative anecdotes.

但确实如此。

But Yes.

你说得对，亚伦和比尔。

You're right, Aaron and Bill.

提醒大家一下，这些只是我们的个人观点，不是投资建议。

As a reminder to everybody, just our opinions, not investment advice.

你知道，其实这里面有一套很清晰的联动系统——你听完了鲍威尔的讲话，回去在管理委员会里讨论，到底该不该加大预算？

But, you know, there's there's like, there's very clear, like, this interrelated, you know, system of like, like, you are like listening to Powell's commentary and then decide and then deciding at your management committee meeting, you know, should we should we lean into budget, you know?

十八个月前，我们听到的全是坏消息：通胀失控，利率即将飙升，科技公司大裁员。

And so so, you know, if we're only hearing doom and gloom eighteen months ago, inflation is exploding, interest rates are about to go crazy, massive layoffs happening in tech.

在这种环境下，谁会愿意投资呢？不管是软件公司，还是明尼阿波利斯的保险公司，大家的想法都差不多。

That's like a scary environment to then be investing, you know, and leaning into whether it's a software company or an insurance provider in Minneapolis, like we're all kind of thinking about this the same way.

但看看现在，通胀明显回落，利率要么见顶，要么已经开始下降。

Fast forward to today, you know, okay, inflation coming down clearly, interest rates either at the peak or starting to come down at some point.

我觉得，现在可以稍微多投一点了。比如之前想上线的那个需要大量数据支持的应用，工程师们之前忙着优化，现在可以稍微调一调，多给点精力到创新上。

Like I feel like, okay, we could invest a bit more, which means that that that incremental application we wanted to launch that had some data consumption as a part of it, you know, the extra focus of the engineers working on optimization versus innovation, you can kind of tune that knob a little bit.

我觉得，这些实际发生的变化，是经济模型里捕捉不到的，但正是这些细微的动向，让我对CIO们未来会更积极地推动增长项目，多了一点信心。

And so I think that's the that's the sort of qualitative not captured in the, you know, econometrics, you know, kind of elements of this that are actually happening now in businesses, which gives me a little bit of of, you know, incremental confidence that that means CIOs are going to be putting on on more growth initiatives going forward.

你刚才说到点子上了。

Hey, you just hit on something so important.

说实话，对我和我这些年的导师来说，最大的超额收益来源之一，就是正负反馈这个概念。

You know, perhaps one of the greatest sources of alpha for me and certainly my mentors in this business over a long period of time is this idea of positive and negative reflexivity, right?

当整个市场一片悲观的时候，人们的行为就会变得特别保守，对吧？

That in fact, when there is all doom and gloom that causes people to behave in a certain way, right?

而当信心开始回升——就像今年年初那样，利率的变动趋势也开始放缓的时候——情况就不同了。

And when confidence begins, you know, like as we started entering this year as confidence is turning and the second derivative on on interest rates is down, etcetera.

这时候就会出现正面的反馈效应。

It leads to the positive reflexivity on the other side.

这会推动加速发展，因为大家觉得安心了，而你得在之前就提前布局。

So it actually leads to acceleration because people feel comfortable in doing that and you need to invest ahead of that.

对吧？

Right?

很多时候，要想真正领先一步，就得在别人恐慌抛售、市场一片惨淡的时候出手，那时候人人都说情况不可能好转。

And and oftentimes to really get ahead of it, you have to be buying when there is, you know, the proverbial blood in the streets, when there is that doom and gloom and people say it can't ever, you know, it can't ever be different.

是啊。

Well, yeah.

对。

Yeah.

没错。

Yes.

我的意思是，我猜这其中肯定有 Timing 的问题，因为你得知道什么时候真正触底了。

I mean, I can only imagine that there is some timing element to that because you have to know when you've reached the bottom of the doing course.

不过撇开这个不谈——我真不想扯到政治上去，但就做个标记吧，我特别想不通，为什么我们现在对改变政府这么热情？

But that aside, which I don't actually want to like veer this politically at all, but like, as just like a bookmark, this is I get very confused why we are so passionate about changing the government right now.

人家明明干得挺好的。

Like they're like, they're nailing it.

为什么呢？

Like, why?

到底为什么？

Like, like, why?

干嘛好端端地突然抽走一块积木，看会引发什么变化？

Like, why pull out a Jenga piece out of nowhere and see what what changes?

不过呢，不同播客有不同的风格。

But again, you know, different podcasts.

我们其实特别幸运，居然就这么碰上了这么好的局面。

But like, like, are so lucky right now that we somehow landed this thing.

当然，边上还有一些小问题在处理，但说实话，我们真不该把这一切看得太理所当然。

And sure, there's, you know, a bunch of incremental issues on the margin that we're dealing with, but like, wow, we should like be, you know, not taking this for granted too much.

这算是一个软件视角。

So that's a software take.

我知道你对另一个话题也有看法。

I know you have opinions on this other one as well.

对吧？

Right?

现在网上正吵得厉害，关于谷歌和那十个蓝色链接，还有接下来会怎样。

There's a, you know, there's a spicy battle out there about, you know, Google and 10 blue links and, you know, and and what happens next.

上周末我发了条推文，这根本不是知不知道的问题。

And, you know, I tweeted over the weekend, you know, this isn't a question about like knowing, right?

拉里和谢尔盖早就明白，那十个蓝色链接只是通往答案路上的一个中转站。

Larry and Sergei always knew that 10 blue links was like a waypoint in route to answers, right?

任何搞信息检索的人，或者一直走在人工智能前沿的人，都清楚这一点。

Like anybody who is in the business of information retrieval or they've been around artificial intelligences and and and out in front of everybody else.

所以关键不在于愿不愿意。

And so the idea here is that, you know, I think it's not a matter of will.

也不是知不知道的问题。

I don't think it's a matter of knowing.

对吧？

Right?

我觉得这其实就是创新者的窘境。

I just think it's this this innovators dilemma, right?

谷歌面临的困境是，它99%甚至超过100%的利润都来自基于十个蓝色链接的广告模式，而这跟Perplexity、ChatGPT这些‘答案时代’的做法，本质上是冲突的。

The Google faces, which is that 99% of their profits or a 110% of their profits comes from an advertising model around 10 blue links that is just at its core seems somewhat at odds, with what perplexity is doing with what chat gbt is doing the age of answers.

所以我想咱们先把这个话题聊聊清楚。

And so I would just love to, you know, kick it around for a second.

如果你现在是谷歌的负责人，你会怎么做？

How if you're running Google today, right?

面对这样的创新者困境，你到底该往哪个方向走？

And you're faced with this innovator's dilemma around this, like what should they be doing here?

这已经超出了我的能力范围。

Well above my pay grade.

幸运的是，我只是个普通的企业软件从业者。

Fortunately, I am just a lowly enterprise software.

不过提醒一下，创新者困境本质上讲的是商业模式的问题，而不是技术问题。

But, know, I so so as a reminder, obviously, everybody, like Innovator's Dilemma is about business model, you know, difficulties.

我们总以为这是技术上的难题。

And we always think it's a technology problem.

其实根本在于，新的东西对你现有的商业模式没有吸引力。

It's always about your business model doesn't make the new thing attractive to you.

所以你一直回避去做，直到被颠覆为止。

And so you avoid doing it until you get disrupted.

说到这个，我可能想得还不够深入，但我觉得谷歌转向人工智能，并不完全是经典的创新者困境。

You know, I might take after, you know, thinking about this a little bit, not maybe not enough, is I don't think there's a totally 100% classic innovator dilemma issue with the move to AI with Google.

相反，从商业模型的角度看，它甚至可能更好。

I think on the other end of this, the business model could be as good if not better.

布拉德，你经常思考这个问题——如果谷歌直接告诉你该买什么，还促成交易，其实商业模式也差不多。

And, Brad, you think about this a lot of, like, of, like, if you if you actually had Google literally tell you the thing to buy and and they got a transaction on that, it would actually be a similar business model.

它们只是更快地把你带到答案，你依然会完成交易，依然进行消费，而它们依然是那个分发渠道。

They would just get you to that answer even faster, and you'd still transact and you still do commerce and they are still the distribution engine

没错。

Correct.

对所有需要做广告的人来说都是这样。

For all of the people that need to advertise somewhere.

每个人都要找到客户，所以你依然需要一个界面，把你的产品送到客户手里。

Everybody has to find a customer, And so you still need an interface to get my product into that customer's hands.

如果谷歌的AI比传统的十个蓝色链接做得更好，那当然再好不过了。

And so if Google's AI is doing that better than the 10 blue link model, that's great.

我觉得更复杂的问题其实更多是产品管理、用户行为和体验方面的挑战——该怎么引导用户适应这种新模式？

I actually think that the more complicated thing is probably more of a product management user behavior experience issue, which is how do I navigate my customer base?

怎么在引导用户的过程中，别让他们因为这段转变而彻底崩溃？

How do I navigate my user base to this model in a way where in the process I don't get totally hammered by that journey?

我知道该怎么处理这个问题。

And I know how to do that.

比如BART，明显就是他们在尝试解决这个问题的方式。

Like, BART is clearly their way of experimenting to do that.

但你知道，他们最大的问题在于：你打开搜索框，输入一个关键词，他们给你一堆结果。

But like, you know, their big problem is like, go to a search box, you type in a search, they give you a lot of results.

而Perplexity和ChatGPT的界面完全不一样。

And Perplexity and ChatGPT are just totally different product interfaces.

所以他们没法直接把谷歌首页改成那样。

And so they can't clearly make the Google homepage just do that.

所以他们得想办法

So they have to kind

听起来是不是有点自相矛盾？

of sound doesn't that doesn't that rhyme in your ear?

我的意思是，1999年的时候，谷歌的界面跟当时那些门户网站完全不一样。

I mean, 1999, Google was just a different interface relative to the portals that existed at the time.

你可能会问，雅虎明明有成熟的商业模式，靠卖网页广告位赚钱，为什么就不学谷歌呢？

And you could say, why the hell didn't Yahoo just copy Google because they had an existing business model that required them to sell real estate on this on this big port.

你大概有90%是对的。

There was probably probably probably you're 90% more right.

但当时谷歌在技术上确实也有独特优势。

But but there was also a literal tech tech differentiation that Google had.

而雅虎没有PageRank算法。

And Yahoo did not have PageRank.

所以谷歌给用户的搜索结果就是更准。

And so Google just literally gave you better results.

于是用户就慢慢迁移过去了。

And so the user base migrated.

我的意思是，也许除了某些内部的政治因素，比如谷歌的技术可能根本不是这个问题的关键。

Mean, maybe short of maybe some internal, you know, kind of political issues like Google's technology is probably not in the way of this issue.

所以现在作为用户，我会留下一个很大的变量——也许限制我们的，其实是我们的大脑本身。

And so now it's actually like as a user now, I will leave like one big X factor, is just like maybe our brains are the limiter here.

ChatGPT已经占据了我们大脑里那个‘提问—得答案’的小角落。

ChatGPT just owns that little slot in our brain of you ask a question, get an answer.

现在我们整个认知模式都变了，谷歌得面对这个全新的挑战。

And like now we have like a whole like brain rewiring problem that Google is going to have to contend with.

我不确定。

I don't know.

我的意思是，这已经超出了负责谷歌搜索首页的产品经理能掌控的范围。

I mean, like that's above the you know, that's above what a PM on the Google search homepage can do.

所以这个问题，真的挺复杂的。

So this one is, you know, pretty complicated.

比尔，你刚才举手了吗？

Bill, were you raising your hand?

我不想打断你。

I didn't wanna interrupt.

所以对于大学比赛日的粉丝们，我先慢点说。

So so for the college game day fans, I'll do a lead course, so not so fast.

我来谈谈另一面。

I'll take the other side of this one.

好的。

Okay.

我觉得这是个严重的问题。

I think it's a horrific problem.

哦，天哪。

Oh, wow.

但是

But

因为用户界面很容易被打乱。

because one is, is the user interface is subject to disruption.

我说要给你扔十来个链接，其实还不止十来个，对吧？

The whole idea that I'm gonna throw you 10 links, It's really more than 10, right?

上面四个广告，下面四个广告，你得翻来翻去才能找到你想要的东西。

Because there's four ads on top and four ads on the bottom and you gotta search through and find what you want.

我是个体育迷。

I'm a big sports fan.

我经常搜球队的阵容名单。

I often search for the roster of a team.

那根本排不到前面去。

That's not even at the top.

对。

No.

得往下翻五个链接才看得到。

It's like five links down.

还得绕开那些广告去找。

Have to find it and get around the ads.

所以这就是第一个问题。

So it's like, so that's one.

第二个，你已经提到过商业模式了。

Two, you already mentioned the business model.

他们的做法就是把用户扔进角斗场，让他们互相厮杀到死。

Their business model is to throw their customers in a cage match and let them compete with one other to the death.

这确实让收入暴涨，因为制造了大量愿意付高价的买家。

And that drove revenue sky high because it created prey to like super optimal payment.

没人喜欢看到四五个广告。

And no one wants to see four or five ads.

用交易模式很难达到广告模式那样的单次访问收入，你之前不是提过客户获取成本和终身价值吗？

You're not going to get to the same revenue per visit with a transaction model that you do with the ad model, because you already mentioned CAC TV.

用营销数据来看，用户在交易模式下愿意为首次购买花40%、50%甚至60%，但他们只愿付5%。

People will pay using marketing math, forty, fifty, 60% of first purchase with a transactional integration, they want to pay 5%.

如果你选择交易促成合作，而不是营销获客，成本能降低到十分之一。

So you have a 10X reduction if you were to partner on a transaction enablement than you do on a marketing lead.

所以客户根本不想在屏幕上看到这两种东西。

And so the customer wants neither of those on the screen.

我得稍微反驳一下。

I'm going to push back a little bit.

我之前说的交易，不一定是转成按效果付费的模式，但也可能是。

And when I said transaction, it didn't necessarily mean that you you turn it into a CPA model, but but maybe maybe you do.

我明白这里面的经济变化。

And and I understand the the the changes in economics there.

我想说的是，当我问ChatGPT一个问题，比如：

I I guess what I'm saying is that when I do a ChatGBT question of, hey.

给我这个东西的食谱。

Give me a recipe for this thing.

对。

Right.

实际上，他们却不让我自己去购买这些服务，这就是个问题。

It's actually a gap that they don't then let me go buy the things

没错。

that Right.

如果你真要买，肯定不会通过营销模式去买。

If you were to go buy them, you wouldn't buy them from a marketing model.

你会通过交易模式直接购买。

You'd buy them from a transactional model.

否则，你就会说：我想订一张飞往芝加哥的机票，再订个酒店房间。

Otherwise, you're gonna say, I'd like to book a flight to Chicago and get a hotel room.

但谷歌AI却不会直接帮你办，反而问你：要不要看看 Hotels.com、Expedia 和 Booking.com 的优惠？

And and and instead of give of doing it, the the Google AI is gonna say, would you like to hear deals from hotels.com and Expedia and booking.com?

我能不能把他们提供的优惠内容念给你听？

Can I read you through the deals they're offering?

嗯，

Well,

布拉德，布拉德，布拉德，这方面你最在行了。

Brad Brad Brad, you're the expert on this.

为什么谷歌不能从这趟航班赚40美元或30美元，而不是靠跳转到Expedia拿那2美元？

Why can't Google get $40 for that flight or $30 for that flight instead of the $2 for clicking through to Expedia?

我觉得比尔说得对，如果看看传统的佣金模式，旅行行业最典型的例子就是酒店，通常佣金是10%。

Well, I think to Bill's point, if you look at the traditional commission models, and I think the iconic example is in travel, the traditional commission model, for example, in hotels is 10%.

假设平均每晚房价是100美元，那他们愿意为这种佣金模式付10美元。

Let's just say the average daily rate is a $100, so they're willing to spend on a commission model $10.

但在这种边际广告模式下，他们为了争取下一个客户、推动增长，会把竞价一路推高到几乎耗尽利润，我觉得就是这样。

But in a marginal advertising model where they're bidding up all the way to, you know, their gross profit on, you know, in order to get that next customer to drive that growth rate, which I think which yeah.

如果你是个初创公司，那肯定更舍得砸钱。

And and and if you're a start up, you certainly go above it.

为了单笔交易，你可能花两倍、三倍的成本，对吧？

You may spend two or three x, right, on that individual purchase.

这种模式确实榨取了更多利润，但我觉得另一个关键是，如果我们把视角拉得再远一点，其实我们正在朝‘直接给出答案和行动’的方向走，对吧？

It certainly has extracted more rents, but I think the other piece here is that if we really just telescope way out here, like we are heading to answers and actions, right?

就连发明那个该死的蓝色链接模式的人，也不觉得这事儿就到此为止了。

And even the people who invented the damn blue link model don't believe that that is where this thing ends.

我只是想说，谷歌想复制一个99%的垄断地位，把搜索领域赚的全部利润都带到这个新世界里，哪怕现在‘ChatGPT’都快成动词了——这想法简直太离谱了。

So like And and all I'm saying is the idea that Google can replicate a 99% monopoly and take all of that pool of profits with it into this new world after letting chat GPT become the verb, at least at the start for what this new world is.

对我来说，这根本不可能信，但大家还是普遍觉得，谷歌会像统治搜索那样统治这个新领域。

It just, you know, that to me is almost impossible to believe, but the consensus view has continued to be that, you know, they're going to dominate this new thing the way they dominated the world of search.

如果他们真能做到，我希望自己能当个股东，一起沾点光。

And I just think if they pull that off, I hope I'm a shareholder along the way.

那将是资本主义历史上最厉害的一次借力打力。

That will be one of the greatest jujitsus in the history of capitalism.

我会非常佩服。

I'll be I'll be remarkably impressed.

最后再说一件事。

One last thing.

还有一个问题。

There's one last problem.

除了我刚才提到的所有问题，他们的核心商业模式就是把客户扔进角斗场，根本不可能培养出友好合作、互利共赢的文化。

In addition to everything I brought up, because their core business model is to throw their customers in a cage match, they don't they don't have the culture internally to partner in a friendly win win way.

我接触过的每一家和谷歌谈合作的公司，最后都收到过一份

Every company I've ever had that sat down to do a deal with Google has been shown

合同

a contract

那种合同，你根本不可能会接受。

that you would never ever, ever, ever consider.

如果我们真要走到这一步——我直接告诉谷歌我最常用的旅行网站是哪个，等等——那他们必须拿出一个双方都能接受的方案。

And if we're going to get to a place where I just tell Google my favorite travel site is this, my whatever, they're going to have to work out a deal that's reasonable on both sides.

但我认为他们做不到。

And I don't think they're capable of it.

他们根本没法坐下来心平气和地谈，这已经不只是商业问题，而是文化问题了。因为你每次跟外部合作伙伴开会，都摆出一副‘我说了算’的姿态，

I don't think they're capable of sitting down and it becomes a cultural problem because if every meeting you have with an external third party, you bring this kind of I'm in charge attitude, getting

这种态度是改不了的

that out

要让一家公司做到这一点，真的非常非常难。

of a company is very, very, very hard.

是的。

Yeah.

我的意思是，你其实没必要

I I mean, I I You don't have to

同意这个观点。

agree with that.

不同意。

No.

不同意。

No.

是的。

Yeah.

我知道科技圈里有很多这样的例子，但说到底，全世界都在给谷歌投广告。

I I I I'm I know that there's great anecdotes of of of that of that problem in in tech, but, like, at the end of the day, you know, the entire world is advertising on Google.

所以他们显然已经掌握了如何与全球每家公司的首席营销官合作的方法。

So they clearly have made they clearly figured out how to partner with the CMO of every company on the planet.

如果我们的首席营销官接到一个电话，说‘嘿，我是谷歌的，能帮你当用户搜索内容管理软件时获得更多客户’，我们还是会跟着渠道走，毕竟最终决定一切的是渠道在哪里。

So if my if our CMO got a call that said, hey, from Google, we can get you more customers when they ask a question about content management software, we're still going to say, we're going to follow where the distribution is at the end of the day.

所以这就是为什么我觉得，关键还是他们能不能把用户从现有的体验引导到新的模式上去。

And so then that's why I still think it's all like the name of the game is still can they navigate their customer base to a new user experience paradigm?

当然，这就像老生常谈的问题了：亚马逊会不会先成为一家影视公司，而不是等影视公司变成亚马逊或者苹果那样的企业。

Well, obviously, enough before you know, it's the classic can, can Amazon become a studio before the studios become Amazon or Apple or whatever.

说实话，现在最火的公司是Perplexity，连ChatGPT都排不上号。

I mean, the company with the buzz is perplexity, not not even ChatGPT.

所以没错。

And so Sure.

比如Bg2pod连前两名都进不去。

Like, Bg2pod's not in the top two.

是的。

Yeah.

然后

And then

但另一方面，他们其实可以直接把这些公司买下来。

but then you can do a different issue, which is like, obviously, they should just buy these things.

可利维和康并不允许这种情况发生。

And and then, you know, levie con doesn't let that happen.

所以，如果他们真收购了Perplexity，改名叫Google Assistant，我想我们根本不会这样讨论这个问题，但他们显然被某些限制绑住了手脚。

So Oh, that's so you know, how do you like like, if if they bought Perplexity, call it Google Assistant, be like, I don't think we'd be having this conversation the same way, but they obviously, you know, are are bound by yeah.

就像Instagram和WhatsApp那样。

Which which which is like the Instagram, WhatsApp.

我只是说，这就像Instagram和WhatsApp，还有Meta当年的做法。

I'm just saying that's like, what's a grant, that's like Instagram WhatsApp, what he's saying and what Meta did.

所以现在谷歌根本没法这么做，这

And so Google's like not allowed to do that right now, which

对，简直太荒谬了。

Yeah, it's totally crazy.

我的意思是，

Mean,

其实还有一点类似微软在1996到1997年的情况，那时候他们被罚了，不能做类似的事情。

Which actually, one last thing is similar to Microsoft in the '96, '97 timeframe where they were, you know, in the penalty box and couldn't do those similar things.

是的。

Yeah.

我觉得不只是政府不许，你得记住Instagram和WhatsApp的情况。

I would say it's not only the government's not allowing, but you got to remember with Instagram and WhatsApp.

他们当时根本不是面临创新者的窘境。

It was not they were not facing innovators dilemma.

那些收购并没有像Perplexity可能会对谷歌核心业务造成的那样，蚕食他们的主营业务。

They those deals did not cannibalize their core business in the way that perplexity would cannibalize at least in part the core business at Google.

不过让我换个角度说。

But let me shift this.

让我把话题往前推一推。

Let me shift this forward.

你之前在我们第一期节目里提到过一个话题，艾伦，我们聊过我的一位合伙人叫Mang做的那些有意思的投资。

You you brought up a subject on our episode one, Aaron, we talked about, you know, these interesting investments made by one of my partners had called Mang.

微软、亚马逊、英伟达和谷歌这些公司对大型模型业务的投资，比尔说，至少这些收入质量不高。

So investments by Microsoft, Amazon, Nvidia and Google into the big, model businesses, Bill suggested, you know, the at a very minimum that's low quality revenue.

但关于这种投资的奖励机制，我清楚记得，就在FTC发布声明的第二天，这事就发生了。

But you know, this credit for investment thing is we've seen historically in the past, wasn't lost on me that the very next day that the FTC came out.

我觉得我们之前发过一条推文提到过。

I think we have a tweet on this.

当时在Squawk上说，他们会调查这些开放AI公司和微软等企业之间的关系。

It was on squawk, you know, and said, hey, we're going to look into the nature of these relationships, you know, between, you know, the open AIs and Microsoft's, etcetera.

比尔，咱们在进入第三个话题前，先把这个说清楚，你觉得FTC这次调查最终会有什么结果？这跟你说的那些观点其实是一致的。

So Bill, just to close that one out before we jump into our third topic, you know, what do you think the outcome is of this inquiry that, you know, that the FTC announced, you know, which was right in line with some of the things that you were were talking about?

我担心的不是这几个公司本身，而是整个行业，这些交易扭曲了市场竞争。

My concern, which was really a concern for the industry more than just those players, is that those deals pervert and distort the market.

一旦市场被扭曲，人们就会做出一些意想不到的事。

And when that happens, people do things you wouldn't expect.

就像Zerp一样。

Just like Zerp.

突然间，竞争的样子就和你想象的不一样了，因为这些事都在发生。

All of a sudden competition doesn't look like you would think it would because those are happening.

再简单说一句，我觉得莉娜根本没提过会计之类的事，但既然现在这么受关注，或许能阻止未来再出现更多这类交易，我觉得这对行业整体更有利。

And and just real quick, I the one thing that I I don't think Lina was she never mentioned accounting or anything like that, but the bright lights, you know, being shined at it might prevent more of those deals in the future, which I think would be healthier for the environment.

我其实不太明白的是，如果我是大科技公司，我在这类投资中的动机和风投完全不一样，可定价权却掌握在他们手里。

The thing that I actually don't understand for you guys is, so if I'm a big tech company, I have an entirely different incentive structure for these for these rounds than obviously a venture capitalist, and yet they're pricing they're pricing these companies.

没错。

Totally.

这种扭曲影响的，要么是这些交易本身，要么就是整个行业里的其他人。

So that's the distortion effect for, you know, either those rounds directly or for everybody else in the space.

就好比，某家公司被一个完全不以经济逻辑行事的主体估值到两千亿美元，但你的收入或估值倍数跟这个根本没关系——因为这家公司的实际现金流或收购溢价，和那个只是签了笔交易的模型压根没有关联。

It's like, yes, X company got a $20,000,000,000 valuation from a non economic, you know, actor that does not make your your revenue or your multiple have any correlation to that because the acquisition premium or the ultimate cash flow of this company doesn't have any relationship to to the credit model that just, you know, got a deal done.

我觉得这更复杂，因为据我了解，在这个大模型竞争圈里，二级市场交易现在都成常态了，尤其是在招聘方面。

I think it's even trickier because based on what I've read, secondary transactions are kind of expected now in the hiring in that world, that LLM competition world.

如果你找不到财务投资者一起参与，因为这些都不是现金信用。

And if you can't get a financial investor to invest alongside, because these are non cash credits.

对。

Right.

那就没钱做二级市场交易，根本跟不上节奏。

Then you don't have the cash to do the secondaries you can't keep up.

这问题就大了。

And then that's tricky.

之前那期节目里，布拉德提到过，这些大公司可能另有动机，不是为了以合理价格获得股权。

Can you, if, and on our last show, Brad talked about why these big companies might have an incentive that's other than ownership at the right price.

就像你说的，估值根本不真实，这会带来各种问题，尤其是对公司运营方来说。

That, you know, as you're saying, the valuation's not real and that, you know, that create there's all kinds of problems this can create, especially if you're running the company.

是的。

Yeah.

可惜的是，这种局面迟早会崩，音乐总会有停下来的那一刻。

I think, unfortunately, there's like an inevitable there's gonna be an inevitable, you know, the the music will stop at some point on on on this particular dynamic.

OpenAI就不说了，毕竟他们确实有市场反响，收入也是实打实的。

OpenAI, you know, aside, just because they actually do have the traction, they they probably do have the real revenue.

对。

Right.

但我真不太看好这种趋势蔓延，从整个模式的健康性来看。

But I I don't love to see the broadening of of this just from a again, healthiness of this model.

我记得比尔提过，这有点像2021年的软银效应，资本被当成了摧毁经济的武器。

I mean, it's a Bill referred to it a little bit like the SoftBank effect from 2021 where capital was used as a weapon of economic destruction.

那时候资本成了决定成败的关键，而不是让产品和企业基本面自己说话。

Capital was the kingmaker and all of this rather than allowing the product and the fundamental performance of the business to go.

不过艾伦，我想稍微换个角度，聊聊更广一点的AI话题。

But I'm gonna just shift this a little bit Aaron, into perhaps a little bit broader discussion about AI.

你发过一条推文，我特别留意了，你说你见多识广，是备受尊敬的技术专家。

You had a tweet that caught my eye where you just said, you know, this is, you know, you've been around for a while, you're deeply respected, technologist.

你还说，现在世界上正发生着一系列技术交汇，这在很多方面堪称前所未有的特殊时刻。

You said, hey, we got this confluence of events going on in the world right now, confluence of technologies that make this in many ways the most extraordinary of times.

所以我们接下来深入聊聊AI，你能先说说那条推文的背景吗？然后我想听听你从内到外的见解。

And so as we dig in to AI a little bit with you, maybe a little context for that tweet and then I would love for you to go inside out.

你在自己的业务里，真正看到AI带来成效的地方在哪里？

Like where are you actually seeing the traction, for AI in your business?

你们团队目前用AI的三大重点是什么？

What are your guys' top three priorities in terms of leveraging AI?

你觉得这些重点和同行们关注的方向一致吗？

And how do you think those reflect the priorities of your peers?

当然可以。

Sure.

没错。

Yeah.

那条推文其实跟很多事有关，比如VisionPro之类的。

I mean, tweet specifically was related to a bunch of stuff including like VisionPro or whatnot.

我觉得现在真是个特别激动人心的年代，抛开经济因素不谈，单是做技术就让人热血沸腾。

Just think it's an incredibly exciting time, purely like, you know, economics aside, to just be building technology.

我们现在有这些平台可以搭建，它们正在做着十年前甚至二十年前根本不可能实现的事情。

We are literally given these platforms to build on that are doing incredible things that, you know, quite literally a decade or two ago was just not possible.

所以这更多是一种情感上的触动——真的，现在这个时代太令人兴奋了。

So that's that's like more of an emotional psychological thing of just like, hey, incredibly exciting times.

说到机会，尤其是AI方面，我们先简单说说Box，然后再展开讲。

As it relates to the opportunities and and maybe then AI specifically, we are we're so so maybe just, like, three seconds on Box, and then and then I'll broaden it.

我们在Box里存储了数千亿个文件。

So our what we're doing is we have hundreds of billions of files that are stored in Box.

每个文件里包含的价值，其实远超过客户目前能享受到的那部分。

And every single one of those files has more value inside of the file than what currently the customer is kind of getting, you know, benefit from.

因为以前你必须打开文件、查看、阅读，或者观看，才能真正从中获得商业价值。

Because you have to, like, open up the file, look at it, read it, you know, watch it to get actual, like, you know, real commercial economic benefit of that content.

有了AI，现在可以派一些小机器人自动处理这些文件，帮你挖掘更多价值。

With AI, you now have little bots that can run around and do things on that content to generate more value for you.

你能在业务中更快地做出决策。

You could get a decision in your business faster.

你可以把合同 summarise 一下，加快整个流程。

You could summarize a contract and accelerate a workflow.

能从非结构化数据里提取信息，自动化过去根本没法自动化的流程。

You could extract data from something that's unstructured to automate a process that was unautomatable before.

所以这对咱们来说特别令人兴奋。

So that's why it's incredibly exciting for us.

现在的情况是，虽然我们已经进入 ChatGPT 爆火的一年多，但说实话，我们还处在最开始的阶段。

Dynamic right now is I think we have even though we're, you know, a year and a quarter into Chatuchu Boutique phenomenon, we're still literally in the earliest of days.

我觉得大多数企业现在都还在试验阶段。

I think most enterprises right now are just in the experimental period.

他们正在摸索，该把 AI 用在哪个环节最合适。

They are trying to figure out where to plug AI into their overall stack.

目前还没人真正搞清楚，普通企业该怎么系统性地应用 AI。

I don't think anybody has sort of fully figured that out, you know, commonly across kind of normal corporations.

这意味着，AI 领域的胜负还远未定，谁都有机会。

And that means that it's anybody's game right now in terms of the winners and losers of AI.

我觉得，说实话，目前在软件领域，我更看好现有的大公司，因为拥有客户数据的环境必须是可信且安全的，这样就不需要来回传输数据了。

I think, you know, I probably prefer incumbents in software on the margin right now, because it's so important to have the customer data in an environment that is already trusted, is already secure, they don't have to move the information back and forth.

他们通常已经有一套成熟的业务流程了。

They often already have the workflow.

所以如果我要选谁能在CRM人工智能领域胜出，我觉得是Salesforce，而不是初创公司。

So if I'm thinking about who wins in CRM AI, it's to me Salesforce as opposed to a startup.

在IT服务管理AI领域，那就是ServiceNow胜过初创公司。

ITSM AI, it's ServiceNow versus a startup.

这道理很简单，因为在纯软件领域，创新者的窘境没那么明显。

That's like the quick thing because there's not as much innovator's dilemma in the kind of pure software categories.

事实上，谁拥有用户、数据和流程，谁就占了大便宜。

In fact, this is just really like an advantage for anybody, again, who has users, data, and workflows.

人工智能简直像梦想成真，现在我们能实实在在地为客户提供更多价值。

AI is like this dream come true because now we can just literally offer more value to our customers.

所以在传统的大公司领域，情况就是这样。

So that's sort of like in the classic incumbent categories.

我对初创公司极度看好。

I'm extremely bullish for startups.

它们可能不会去颠覆那些已知的软件类别。

They're just not going to probably be about disrupting, you know, known categories of software.

更可能是围绕经济中一些成熟的领域展开。

It's probably much more about known categories of like the economy.

过去二三十年，我们一直在把软件变成人类工作的上层工具，而现在，软件终于能直接替代人类去做那些事了。

And so I think if we spent like the past twenty to thirty years putting software, making it so, you know, software is a layer above what humans do, now we're at a point where software will just do what the human did.

这就开辟了全新的机会方向。

And that creates a whole new vector of opportunities.

我觉得，传统上看待这个问题的方式，就是找出哪些知识型工作可以被转化为可处理的模块。

And I think, know, the classic way to look at this is sort of like, you know, figure out which parts of knowledge work can convert into tokens.

那些地方就是全新的软件机会，以前根本不存在对应的软件。

And then those are the areas where there's new software opportunities that we did not have software for before.

而这些，就是新的商业机会。

And then those are the businesses.

我觉得人工智能领域会创造出数万亿美元的产值。

So I think there's you know, there'll be trillions of dollars made in AI.

但这些机会主要会集中在垂直领域。

But I think it's going to be in verticals.

更多是帮助提升人力在工作中的效率。

It's going to be, you know, kind of helping augment the people element of the work.

而是在围绕它的基础设施和支撑体系上。

It's going to be in the infrastructure and scaffolding around it.

我对纯AI模型提供商的商业前景不太乐观，也不是完全看空，只是没那么兴奋。

I'm a little bit bearish, not bearish, but just like more like on the margin, not as excited about the economics of the actual pure AI model providers.

你想想，扎克伯格随时可能把你们埋头苦干三五年的东西开源了。

I think it's tough to be in a space where Zuck, you know, at any day could just open source the thing that you've been working on for, you know, three or five years.

然后一瞬间，技术就来了个跨越式突破。

And then all of a sudden, you know, now, you know, there's just this leapfrog moment of technology.

我常想，很少有技术像这样，你辛辛苦苦打造的成果，作为知识产权几乎转眼就过时了。

And, you know, the the thing that I go back to is, like, there's there's rarely been technology I think we've seen where the actual, like like, the thing you're building, the asset you're building is almost perishable as an as a piece of IP.

这东西 literally 可能下一秒就过时了，还没法更新。

Like like, the thing literally, like, can just become obsolete, like, a second later, and you can't update it.

你没法把过时的东西稍微改一改，让它变好一点。

You don't like, I can't take the obsolete thing and make it a little better.

你只能重新训练一遍，再搞个新模型出来。

I have to, like, rerun the training run and then, like, do the new model.

所以我实际上已经把资本支出都押在这上面了。

And so I actually have, like, converted, you know, CapEx dollars into this like thing.

但随时可能冒出个更好的东西，把你这个甩在后面。

And at any moment, something else could be better than that thing.

而且我根本没法灵活调整，只能硬扛。

And I have no I can't pivot around that.

我只能困在这个容易过时的资产里。

Like, I'm stuck with this perishable asset.

所以你最后只能想，好吧，OpenAI、微软、谷歌、Facebook 这些人，大概就是该押注的那几个。

So then you kind of just say, okay, well, OpenAI, Microsoft, Google, you know, Facebook are are basically like where you you place the bets on who makes the models.

其他人就该专心做软件开发了。

And then everybody else should just be building software essentially.

那内部呢？

What about internally?

作为一家大型上市公司的负责人，我很想知道，你对程序员生产力的提升有什么期待？

I'm curious as a as a someone that runs a large public company, how do you feel about like, what's the right expectation for programmer productivity improvement?

你们有在衡量这个吗？

Are you measuring it?

你真的在意这个吗？

Do you care?

你们公司在哪些地方已经看到初步成效了？

Where in your org are you seeing impact early?

没错。

Yeah.

首先，编程方面，Copilot 已经带来了明显变化。

So, programming, obviously, in the first place with Copilot.

说实话，我们内部并没有做过专门的测量。

And I don't know, we have not done a specific measurement internally.

我平时会逛逛办公室，看看大家屏幕上在做什么。

You know, I'll walk through the office and see what's on people's screens.

AI确实被广泛用在ChatGPT上，帮人优化代码，比如怎么让SQL查询更高效，诸如此类的事情。

And, you know, AI is certainly actively being used for in ChatGPG to optimize a code, you know, the code that somebody is working on, you know, how do I change this, you know, SQL query to be more efficient, that kind of stuff.

Copilot当然主要用于写代码。

Copilot, obviously, for writing code.

关于效率提升的估算，我基本认同那些说法，大概在百分之三十到五十之间。

Obviously, the estimates are I would I would probably just agree with the estimates of, you know, thirty, forty, 50%.

我之前也听过类似的说法。

That's what I had heard.

德鲁克昨天说能提升七倍。

Druck said seven x yesterday.

算了，不扯这个了。

Anyway, move on.

德鲁克这么说的？

Druck said that?

我觉得他是这么说的。

I think he did.

是的。

Yeah.

好的。

Okay.

我还没有

I I have not

见过这个。

seen that.

我觉得这不太准确。

I don't think that was accurate.

好的。

Okay.

好的。

Okay.

明白了。

Got it.

好的。

Okay.

对。

Yeah.

我觉得30%这个数字挺不错的。

So I like the 30% number.

对。

Yeah.

我大概更倾向于30%到50%这个区间，就目前的情况来看，最好的情况可能就是这样。

So I I'm I'm probably more in the, you know, 30 to 50% camp in it, like, in the best case scenario right now with where we're at.

而且我觉得，编程仍然是最重要的应用场景。

And and and I think still coding is the the number one use case.

其实，我觉得这是思考人工智能时一个很重要的点。

And and there's actually I I think this actually is an important element of of thinking through AI.

比如，编程到现在还是比其他大多数应用场景表现得更好。

Like, coding is still working better than most other use cases.

这是因为编程基本上是在一个文本界面里进行的

And it's because you basically have a workflow where you're in a text interface

没错。

Yep.

整个流程是线性的，而且这个领域的大部分知识都是公开的、开源的，方便用来做训练。

That is just linear, where most of the the sort of knowledge of that of the of the field is all public and open source for the most part and available for training runs.

而代码的下一行预测，正好完美契合了这种预测模式。

And and the next line prediction, you again have the perfect interface for next line prediction.

但大多数知识型工作其实并不是这样的。

Most knowledge work actually doesn't look like that.

我们倒是希望它能变成那样。

Like, we we want it to.

我们愿意相信这一点。

We want to believe Yeah.

就连法律工作也是这样，你得跟客户沟通，然后再跟别人对比一些东西。

It's it's and so, you know, even even the legal work, I'm I'm I'm still, talking to the client and then and then, like, comparing something with somebody else.

所以，认为所有类型的知识工作都能像 GitHub Copilot 那样发挥作用，这个想法

And and so this idea that that you're get a get you're gonna get a GitHub Copilot effect for all forms of knowledge work

还太早

is not

至少在近期内，这么快下结论还为时过早。

At least in the near term, it's it's way too early to be kind of jumping to that conclusion.

编程比日常语言更精确。

Programming is like a more precise language than language.

没错。

Yeah.

而且，它还能立刻自动测试。

And then and then literally, it's automatically testable instantly.

没错。

Right.

但很多其他工作并不具备这些特点。

And so you just don't have those characteristics for a lot of other work.

所以我觉得，AI要想像GitHub Copilot那样深入普通知识工作者的日常，可能还得花更长时间。

So I think it's going to take longer than maybe we'd like for AI to sort of show up in the average knowledge workers sort of day at the level that GitHub Copilot did.

就算你能自动生成我的所有邮件，还写得更快，我依然得逐封阅读。

You know, if you could automatically write all of my emails, you know, faster, I still would have to read each one.

我还是要判断，我是否认同邮件里说的那些内容。

I still have to process like, do I agree with with that thing that that is being said?

我依然得理解信息的实质，以及发件人背后的潜在风险。

I still need to understand the substance of the content that from the risk from the sender.

AI不可能直接取代我。

I can't have AI just sort of jump replace me.

所以我觉得，就整体知识工作领域的变革而言，我们可能还是有点太早了。

So I think we're probably a little bit early in the general knowledge work sort of, you know, transformation.

我更感兴趣的是那些重复性很强的工作流程，人一遍遍做同样的事，我们可以用AI来替代，从而优化整个流程。

And the things I'm much more excited about are where can you take a process where a person does that kind of rote task over and over again, And we can swap that out with AI and improve that workflow.

这才是真正的机会所在，也是我们目前专注的方向。

And that's where the opportunity is and certainly what we're spending our time on.

这让我想起一件事。

I mean, it reminds me so much.

这周我看到一张图表，也许我们可以给摩根士丹利展示一下。

I saw this chart this week, maybe we can pull up for Morgan Stanley.

它显示了我们在一个合理的时间段内，总是系统性地低估超级周期的规模。

And it shows how we systematically underestimate the size of super cycles right over a reasonable period of time.

当我想到这里的情况时，他们回头做了些计算，发现超级周期在初期往往被低估了大约40%。

When I think about what's going on here of course they show they went back and did some math against initial estimates and said, you know, super cycles are about 40% underestimated at the start.

比如个人电脑或互联网刚出现时的预测，2000年预计用户是1.52亿，结果实际达到了3.61亿。

So, you know, the initial forecast for the PC or the initial forecast for Internet, 152,000,000 users in 2000 ended up being three sixty one million users.

当我思考未来会发生什么时，我完全相信在接下来的四个季度里，这种情况就会显现。

When I think about what's gonna happen here, I can absolutely see sometime over the next four quarters.

我们马上就要进入一个幻灭期，因为现在人人都一窝蜂地冲进来了。

We're gonna hit a zone of disillusionment because everybody's pig piled in here.

他们觉得所有事情现在都该发生了。

They think it's all happening now.

但实际情况会比预期慢一些，就像当年的互联网和云计算一样，真正走上正轨需要时间。

It's going to take a little bit longer just like the internet did just like cloud did, right to hit its stride.

等到那时候，那些一直反对这种超级周期的人，就会说这是炒作、是泡沫，诸如此类。

When that happens, those who were, you know, who are always against the super cycle, call it a fad, everything else.

他们会说：看吧，我早说过吧，根本没像你们想的那样发生，你们都押错注了，等等。

They're gonna say, see, I told you so, You know, it's not really happening the way you all thought it was, you know, you made bad bets, etcetera.

但我的感觉是，AI对经济生产力的影响，最终会超越互联网本身。

But my my sense here is, you know, and I've said the AI is gonna be bigger than the Internet itself in terms of impact on economic productivity.

不过我觉得，在接下来的几个季度里，幻灭期确实会慢慢到来。

But I do think sometime in the next several quarters we have this zone of disillusionment set in a little bit.

事情的发展不会像我们所有人当初想的那么快。

It doesn't happen quite as fast as we all thought.

但三年到五年后回头看，你发过一条推文，说这些模型的成本大幅下降了。

But then when we look back three to five years from now, you know, you had a tweet about, you know, the dramatic reduction in costs of these models.

你可以想象，ChatGPT五代或六代，成本比现在低了百分之九十。

So you can imagine chat GPT five or six and it's costing us 90, you know, percent less than today.

我们根本想象不到，这会怎样彻底改变一切。

I don't even think we can get our heads around how that's going to change everything.

不过，为了强调一下，大多数情况是，我们正深入探讨具体的应用场景，比如我有一份特别长的合同。

Well, well, so most of and and just to just to underscore this, most and, again, we're we're like we're deep in in in sort of the use cases of I have a really long contract.

我想读一下这份合同。

I wanna read the contract.

我想用AI自动提取合同里的数据。

I want to automatically present data from that contract with AI.

想想看，所有这些场景——发票、合同、演示文稿。

And so think about, like, just all of those things, you know, an invoice, a contract, a presentation.

客户向我们提出的需求，绝大多数关于内容处理的事情，现在完全都能实现。

Most of the things that customers are asking us for, what they want to do with their content is 100% possible.

关键在于，AI的成本得低到足以让这种应用场景从人工转向软件方式才值得去做。

And and just there's a curve, which is is AI cheap enough to to make that use case be worth transforming to a software based way of doing that versus human based way.

这正是市场想要的，因为你知道成本一定会继续下降。

And so that's like exactly what you want in a market because you know the costs are going to come down.

等到某个临界点一到，一切就水到渠成了。

And then you'll reach that convergence point.

所以我们现在越来越不受技术架构本身的限制了。

So like we're increasingly not limited by the architecture of the technology.

现在真正卡住我们的，是詹森能不能把成本再压得更低？

And now we're just limited by like, can Jensen make these things cheaper?

OpenAI的工程师们能不能设计出更高效的模型算法？

Can the, you know, can the engineers at OpenAI have more efficient model algorithms?

这太好了，因为我们可以直接顺着这条趋势往前走。

And that's great because then you can just ride that curve.

如今，企业想用AI做的应用场景多到数不清，远比成本和架构的扩展能力更紧迫。

Like, we are we are we are way more inundated with use cases that companies want to do with AI today than simply just like like the the scalability of these architectures, from a from a cost standpoint.

布拉德，我再给你一个理由，说明你对‘失望低谷’的看法可能没错。

Brad, I'll give you one other reason why I think you might be right about the trough of disillusionment.

我记得以前看过一个斯皮尔伯格的采访，他谈《大白鲨》的时候说过。

You know, I saw this inner great interview once with Spielberg when he was talking about Jaws.

他说，自己直到影片快到75%的时候才让鲨鱼露面。

And he said, you know, I didn't show the shark until like 75% of the way through this thing.

他解释说，因为人类的想象力，远比任何银幕上能呈现的东西都要宏大。

And the reason he said he did that was because humans imagination is much grander than anything I could possibly put on the screen.

其实，在大语言模型出现之前，我们早就有了人工智能。

And I, you know, we had AI before LLMs.

我记得尼申曾经提过这一点。

Like I heard Nikesh bring this up once.

但在我们今天的语境里，一提到人工智能，大家想到的就是ChatGPT和大语言模型带来的变化。

And so, but in our modern lexicon, we think of AI as chat GPT and what the LLM did.

正是这个变化，才真正让我们震惊。

That's what really shocked us.

我就是担心，这种‘一、二、三’的说法让人以为技术发展是线性推进的。

And I just worry that the one, two, three thing has caused people to think there's this linear extrapolation.

我觉得文本处理这块差不多已经解决了，该扫描的也都扫得差不多了。

I do think the text has kind of been solved and everything's kind of been scanned.

我觉得下一步的发展根本不是线性的，反而特别非线性。

And I think the next steps not linear are super linear.

我觉得接下来可能会有点像——

I think it's going to be a little bit of a Well,

我觉得云服务的发展曲线其实挺有启发性的，至少它大致反映了改变基础设施架构所需的那种变革管理节奏。

think the, I mean, the growth of cloud as just a shape of the curve is somewhat instructive because it at least closely approximates like the change management dynamic that is required to change your infrastructure architecture.

当然这也不是完美的类比，当年六七年我们看到AWS的时候，心里都咯噔一下：这玩意儿明显是未来。

So obviously not a perfect analogy, like anybody in 'six, 'seven, when we saw AWS, we're like, Oh, shit, like this is clearly the future.

我再也不想自己去管服务器了。

I would never want to go and manage servers again.

可现在都2024年了，从收入增长来看，云服务的增速还是最猛的。

But now here we are, it's 2024, and you have the biggest growth rates, you know, from a dollar standpoint still happening.

所以为什么

And so why it

拿什么说

take why

这件事怎么还要花十八年才能实现？

is it taking eighteen years for that still to occur?

因为明尼阿波利斯的一个数据中心里， still 有人得手动迁移数据。

It's because there's somebody in a data center in Minneapolis that still has to just move the data.

他们得把服务迁到云端。

They have to move the services into the cloud.

所以我们跟客户聊的时候，他们会说：我们给你们演示一下，AI 现在能立刻读取发票，还比人做得更便宜高效。

And so the equivalent of this is we'll talk to a customer and they say, we will you know, we'll we'll we'll show them the demo of of AI can now read that invoice and and do what the human was doing instantly and much more cost effectively.

但要把整个工作流程改过来，用 AI 替代原来的人工环节，还是得花一两年做变革管理。

But it's still gonna take a year or two to do the change management of the workflow itself to swap in AI for where the where the human was.

所以现在把这个情况放大

And so now multiply that