Perplexity CEO：指数级增长、不懈改进与 AI 的未来

所以如果你今天在美国 18 岁，必须从头开始，你还会读本科和博士吗？

So if you were 18 today in the US and had to start all over again, would you still do bachelor's PhD?

现实地说，这条路已经没了。

Realistically, it's gone.

这位是 Aravind，Perplexity 的 CEO 兼创始人，一家价值 200 亿美元的公司，挑战着 OpenAI、谷歌和 Meta 等巨头。他们刚刚发布了 Comet，第一款能为你思考并行动的 AI 浏览器。它还能理解你、你的需求、你的预算，一切。它有记忆。我亲自试过，哇，它能替我预订会议、回复邮件、研究产品，甚至在线购物。亚马逊用 AI 做推荐排名，谷歌用 AI 做搜索排名。所以他们可以影响甚至操纵你买东西。但第一次，你手里有了一个 AI。它实际上在保护你，给你对抗大科技的力量。但 AI 在赋予你前所未有的力量的同时，也在以超乎预期的速度重塑工作世界。看起来那些入门级工作正在消失，对吧？你认为会发生什么？

This is Aravind, CEO and founder of Perplexity, a 20 billion company challenging giants like OpenAI, Google, and Meta. They just released Comet, the first AI browser that thinks and acts for you. It also understands you, your needs, your budget, everything. It has memory. I tried it myself and wow it can book meetings, reply to emails, research products, even shop online all without me. Amazon has AI do the recommendation ranking. Google has AI to do the search ranking. So they can influence things for and manipulate you to buying stuff. But for the first time you have an AI in your hands. It actually protects you. It actually gives you power against the big tech. But while AI gives you unprecedented power, it's also reshaping the world of work faster than anyone expected. And it looks like those entry-level jobs are disappearing, right? What do you think is going to happen?

我当然这么认为，人们需要更进一步。

I certainly think so like people need to push further.

那么机会还在吗，还是该重新思考一切？大家好，欢迎来到 Silicon Valley Girl。这是我第一次有重复嘉宾，你的进步太惊人了。我们在 2023 年 11 月采访过 Perplexity 创始人 Aravind。当时公司估值 5 亿美元，现在大约

So is there still opportunity or is it time to rethink everything? Hello everyone. Welcome to Silicon Valley Girl. It is the first time that I have a repeated guest on my podcast and your progress has been tremendous. We did a podcast with Aravind, the founder of Perplexity in November 2023. The company's valuation back then was 500 million and now it's around

实际上我们当时刚完成一轮融资，所以更像是 1.5 亿，然后我们即将以 5.2 亿估值完成融资。

actually we were just closing our series around then so it was more like 150 and we're just about to close around at 520.

好的。现在呢？

Okay. And now it's

200 亿。

20 billion.

大概吧。好的。你能用一个词描述这段时间吗？

Yeah roughly. Okay. Can you describe this period of time in one word?

指数级。

Exponential.

是的。你认为是什么在指数级增长中发挥了巨大作用？

Yeah. What do you think has played a tremendous role in this exponential growth?

我认为是我们不断追求快速交付同时保持质量的冲动，我们的不懈，我喜欢“不懈”这个词。事实上，我想买 relentless.com 这个域名，但发现它归杰夫·贝佐斯所有。

I think our constant urge for like shipping fast yet maintaining the quality, our relentlessness, like I love the word relentless. In fact, I wanted to buy the domain relentless.com except I figured it was owned by Jeff Bezos.

真的吗？

Really?

所以，如果你在浏览器里输入 relentless.com，它会重定向到 Amazon.com。他是我们公司的投资者。这就是公司的精神。我们从不放弃。我们非常好奇。我们不断问自己如何改进产品。我醒来第一件事就是阅读所有用户在不同社交平台上的反馈。我确保大多数 bug 立即得到处理。你可能会想，每天醒来就是修 bug、分类、找改进点，这到底能带来什么？但我们坚信一个信条：1.01 的 365 次方等于 37.78。

So, if you actually type relentless.com on your browser, it will redirect to Amazon.com. He's an investor in our company. That's like the spirit of the company. We don't ever give up. We are very curious. We constantly keep asking questions on how to improve the product. The first thing I do when I wake up is read everything people all the users saying on different social platforms. I make sure most of the bugs are immediately attended to. And you may think like oh what really comes out of this process of waking up every day and just bug fixing and triaging and like trying to identify places for improvement. But we really believe in the mantra of like 1.01 to the 365 is 37.78.

解释一下。

Explain.

好的。如果你每天改进 1%，年底你会改进多少？你会改进 3700%。不是 1%乘以 365。这就是指数级的概念。

Okay. So if you do a 1% improvement every day, how much do you improve at the end of the year? You improve 3,700%. You don't improve like 1% multiplied by 365. That's the concept of the exponential.

是的。

Yeah.

它会复利。所以每天看起来只是小步，但随着变化传播到更大规模的用户，他们会通过口碑带来更多用户，然后这些用户体验到你所做的所有改进。人们会感觉很好：上周用的产品，几周后再用，感觉大不相同，更快、更干净、更准确。于是人们开始信任。我们通过准确性和改进速度赢得用户信任。

It compounds. And so it may look like small steps every day but as the changes propagate to a larger scale of users they bring in more users through word of mouth and then those users are experiencing all the improvements you're making. People feel great using a product that you used last week and then use it a couple of weeks later and it feels much more different, much faster, much cleaner, much more accurate. So then people start trusting. We earn the users trust through our accuracy, our speed of improvement.

还有你的营销，F1，你有没有看到 F1 之后或者节目播出后的峰值？

And your marketing formula one have you seen the spikes after like formula 1 and when the episode went live?

很难归因于这类品牌合作。很容易归因的是你在 Instagram 或 YouTube 上投广告，有明确的行动号召按钮，引导去应用商店安装，然后看转化了多少查询或用户是否订阅。这很容易追踪和优化漏斗。但优化起来难得多。比如，如果刘易斯·汉密尔顿头盔上有 Perplexity，一场 F1 比赛后有多少人会去安装应用？

It's hard to attribute to these kind of brand partnerships. It's very easy to attribute you run some ads on Instagram or YouTube and like there's a clear call to action button in terms of going to the app store play store and getting an install and then looking at how many queries that translated to or whether that user ended up subscribing to the app. It's very easy to track this and optimize the funnels. It's much harder to optimize. Okay, like if Lewis Hamilton has perplexity on his helmet, how many people are actually going and installing the app after a Formula 1 race?

是的。

Yeah.

这很难追踪。最好的追踪方式是品牌知名度：比如英国人或 F1 群体现在是否更了解 Perplexity？我认为应该是这样。我们做过一些基础研究。但核心想法其实是把我们的品牌与一些有史以来最伟大的人联系起来。

It's pretty difficult to track that. Best way to track that is like brand awareness like are people in the UK or people who typically are in the Formula 1 crowd are they more aware of Perplexity now? And I think that should be the case. We have run some basic studies. But the core idea there is honestly to associate our brand with some of the greatest people of all time.

你知道史蒂夫·乔布斯搞过著名的“Think Different”广告，对吧？他从耐克得到灵感，因为耐克颂扬伟大运动员。所以广告里你会看到爱因斯坦、甘地、披头士，所有真正激励史蒂夫的人。所以我们在品牌关联上一直很有意图，至少体育方面。

You know like Steve Jobs ran this famous think different campaign, right? He got the inspiration from Nike because Nike celebrated great athletes. So that's why in the ad you see Einstein, you see Gandhi, you see all these like the Beatles, you see all the people who really inspired Steve. And so that's one thing we've been pretty intentional about like who we want to associate with in terms of the brand. Like at least in sports.

刘易斯·汉密尔顿是有史以来最伟大的车手之一，被认为是 F1 史上最伟大的。所以我们做了这个品牌合作。我们还会在 F1 领域做更多。这是一项非常需要脑力的运动，每个人都着迷于接待、赛车，肾上腺素飙升，但根本上这项运动很大程度上是工程导向的：谁有最好的引擎，谁授权谁的引擎。我认为全是制造商在竞争谁有最好的车，而技术仍然重要，条件重要，轮胎重要，一切都很重要。所以我们看到很多人在应用上问这类问题。

Lewis Hamilton is one of the greatest of all time. It's considered the greatest of all time in Formula 1. And so that's why we did that brand partnership. And we're also going to be doing more particularly in F1. It's a pretty cerebral sport like everyone's attracted by all the hospitality and the race cars and it's very high adrenaline thing but fundamentally the sport is largely engineering oriented like who has the best engine, who's licensing whose engines. I think it's all about constructors competing for who has the best cars and skill still matters, conditions matters, tires matters, like everything matters. So we do see a lot of people asking these kind of questions on the app.

两年前我第一次采访 Aravind 时，Perplexity 估值约 1.5 亿美元。今天，它是一家 200 亿美元的公司，与谷歌竞争搜索查询。那么，如何像 Aravind 一样规模化？首先要确保每个客户接触点都和你一样努力。从电子邮件开始。因为当你从零开始建设时，你不仅是创始人，还是营销人员、支持团队和运营主管。在我们公司，21%的销售直接来自电子邮件营销。所以保持这个渠道强大是没得商量的。这就是为什么我推荐一个叫 Omnisend 的平台。它是一个集电子邮件、短信、弹窗、推送通知和细分于一体的平台。

2 years ago when I first interviewed Aravind, Perplexity was valued at around $150 million. Today, it's a $20 billion company that competes with Google for our search queries. So, how do you scale like Aravind? You start by making sure that every customer touch point work as hard as you do. And that starts with email. Because when you're building from the ground up, you're not just the founder, you're also the marketer, support team, and ops lead all in one. In our company, 21% of sales come directly from email marketing. So, keeping that channel strong is non-negotiable. That's why I can recommend a platform called Omnisend. It's an all-in-one platform for email, SMS, pop-ups, push notifications, and segmentation.

6 月份的查询量相比 5 月出现了巨大增长。你能归因于什么吗？发生了什么？

There was this huge spike in queries in June compared to May. Can you attribute that to something? What happened?

我不确定是 6 月还是 7 月，但过去几个月我们确实增长了很多。实际上，7 月份发生的主要事件，我认为从根本上推动了增长，就是我们发布了智能体式浏览器 Comet。这是第一次，你可以拥有一个能与你一起思考、代表你采取行动并实际去完成事情的浏览器。

I don't know if it happened in June or July, but definitely we grew a lot in the last couple of months. Actually, the main event that happened in July that I would say is fundamentally led to growth is our release of our agentic browser, Comet. For the first time, you could have a browser that can actually think with you and take actions on your behalf and actually go and do stuff.

我这里有 Comet。你能展示一个你认为每个人都应该开始使用的杀手级功能吗？

I have Comet right here. Can you show one killer feature that you think everyone should start using on Comet?

我喜欢这个功能，因为它很简单，而且几乎总是有效。你知道，如果你想找一个我记得看过但不确定具体是什么的视频，我可以用简单的自然语言搜索，并让它从那个确切时刻打开并播放。例如，有一个视频中黄仁勋说：“我宁愿把人折磨成伟人，也不愿解雇他们。”你能找到它并从那个时刻开始播放给我吗？关键不在于它能做视频搜索之类的，而在于它如何进行多步推理：找到视频，提取字幕，然后实际打开标签页并开始播放。

I love this thing because it's a very simple thing where it kind of always works. You know, if you want to go to some video I remember I watched before but I don't know exactly what it is, I can search for it with just simple natural language and ask it to just open and play from that exact moment. So, for example, there's a video where Jensen Huang says, 'I would rather torture people to greatness than fire them.' Can you find it and play it for me from that exact moment? The point here is not that it can do a video search or anything like that. It's how it does multi-step reasoning, finds the video, pulls the transcript, and actually opens a tab and starts playing.

哇。这正是我希望的那句话。

Wow. That was the phrase I was hoping.

然后你可以与这个视频聊天。我昨天看的时候做了一件事，比如这里有很多其他视频，这个有一个小时长。我可以说：“黄仁勋在这次采访中说了哪些不寻常的事情？”之类的。我不想看全部内容。也许他反复说 AI 会爆发等等，但我只想知道新东西。然后它会读取所有字幕。它特别提到了 GPU 使用时间机器——我从未听他说过 GPU 使用时间机器——当然还有规模和灵活性、科学到应用、Omniverse 的成本。所以我可以说：“你能把这个作为邮件分享吗？”我不知道你是否连接了你的……你能把这封邮件发送给创始人兼 CEO Aravind Srinivas 吗？你甚至可以编辑它。你可以添加“从 Comet 发送”之类的，如果你愿意还可以优化邮件，然后直接从那里发送。你不需要打开、复制粘贴，不需要打开这个标签页，也不需要从头撰写邮件。这只是一个例子。你还可以从这里安排会议。你可以说：“从亚马逊给我订购黄仁勋穿的那件皮夹克。”它会找到它。当然，它可能很贵，所以不会直接下单。它会做很多这类多步操作。

And then you can chat with this video. Something that I did yesterday when I was watching, for example, there's a lot of other videos here. This is like one hour long. I could literally say, 'What are some non-trivial things Jensen said in this interview?' Something like that, right? I don't want to watch everything. Maybe he's saying a lot of things he's repetitively saying that AI is going to explode, blah blah blah, but I only want to know new things. And so it's reading all the transcripts. It's particularly talking about how GPU uses time machines—I've never heard him say that GPUs use time machines—and of course scale and flexibility, science to application, Omniverse was cost. So I can actually say, 'Can you just share this as an email?' I don't know if you're connected to your... Can you send this as an email to Aravind Srinivas, the founder and CEO? You can actually even edit it. You can add 'send from Comet' or something like that, and you can even refine the email if you want, and you can just send it from there. You don't have to open, you don't have to copy-paste this. You don't have to open this tab. You don't have to compose email all over. This is just one example. You could also schedule meetings from here. You could say, 'Order the leather jacket Jensen's wearing for me from Amazon.' It's going to find it. Of course, it's going to be pretty expensive, so it's not going to place the order directly. So it's going to do a lot of these multi-step things.

那些靠付费广告生存的企业会怎么样？它会跳过付费广告吗？或者它如何研究购买什么产品的信息？

What's going to happen to businesses who survive on paid ads? Is it going to skip paid ads or how is it going to research information about what product to buy?

所以你可以指示你的 Comet 助手说：“我希望你跳过所有广告。”当你让 Comet 替你看 YouTube 时，你可以要求 Comet：“忽略所有广告，只告诉我内容的核心意思。”你可以让 Comet 说：“去看看 Aravind 最喜欢的 Instagram 网红，阅读他们所有的评论，然后根据你已经了解我的信息，告诉我什么最适合我，并替我购买。”

So you can instruct your Comet assistant to say, 'I want you to just skip all the ads.' When you're tasking Comet to go watch YouTube for you, you can ask Comet to say, 'Ignore all the ads. Just get me the core sense of what the content is about.' You could ask Comet to say, 'Go look at favorite Instagram influencers of Aravind and go read the reviews of all of them and then tell me what's best for me based on what you already know about me and go buy it for me.'

那么，企业如何优化以适应这种体验？

So, how can businesses optimize for that experience?

我认为企业真正应该做的就是专注于打造优秀的产品，并让很多人阅读评论，对吧？因为 Perplexity 会阅读评论、观看视频。所以尽可能多地获取关于你产品的内容。解决方案不应该是花钱请人写好评。但老实说，打造优秀的产品，传播口碑，然后人们谈论它、称赞它，Comet 会读取所有这些，并能够向你推荐。这可能是可行的方法。

I would say that what businesses should really do is just focus on building a great product and get a lot of people to read reviews, right? Because Perplexity reads reviews, watches videos. So get as much content about your product as possible, right? The solution shouldn't be to pay people to write great things about you. But honestly, building a great product and getting the word out, and then people talk about it, rave about it, and then Comet reads all that and is able to suggest that to you. That would be the way to go, maybe.

你认为我们多久会转向让 AI 助手替我们购物？我觉得现在已经可以了。但我的意思是，社会上仍然存在摩擦，我记得你谈到我们如何多付钱给房地产经纪人、财务顾问，但当我想到一个 60 多岁、已经习惯传统方式的人时，他们不会轻易转向让 AI 替他们做事。

How fast do you think we're going to switch to asking our AI assistants to buy things for us? I think you can do that already. But I mean, there's still friction in society, and I remember you were talking about how we're overpaying real estate people, financial advisors, but still when I think about a person who's in their 60s who's done it for a while, they're not going to just switch to AI to do things for them.

我给你讲个轶事，它让我相信为什么值得构建这些东西。有一次，我在以前住的 Noe Valley 理发，另一个人也在同一家店理发，他年纪稍大，他抱怨说他花了整整一天——直到理发前大概三四个小时——都在找一台新洗衣机，因为这是一笔相当昂贵的购买。

I'll tell you an anecdote which kind of convinced me why it's worth building these things. I was once having a haircut somewhere in Noe Valley when I used to live there, and another person was also having a haircut in the same place, and he was somewhat older and he was ranting about how he spent his entire day until that haircut—probably 3-4 hours—just looking for a new washing machine because it's a pretty expensive purchase.

所以，你会花很多时间决定买什么，浪费一整天看评论、看视频、刷评论，最后反而比开始时更困惑，对吧？

So, you're going to spend a lot of time deciding what to buy and you've wasted your entire day just reading reviews and watching videos and scrolling through reviews, and you end up getting even more confused than you were at the beginning, right?

这正是我们希望 Comet 能帮你的：快速做出好决策，把那些琐碎的认知劳动交给 Comet。你为什么要去读所有评论？很无聊。而且，有人花钱写好评。所以你怎么从噪音中提取信号？你得跨平台看评论，不能只看一个平台。我可以花钱在谷歌地图上写好评，但我不可能每个平台都花钱。我没那么周到。所以如果产品很差、过度宣传，肯定有人在 YouTube 或 Reddit 上说负面评价。

And this is the kind of thing we want Comet to help you with: make really good decisions really quickly and delegate all the mundane cognitive labor to something like Comet. Why do you want to go read all the reviews? It's boring. Anyway, people are paid to write good things. So how do you get the signal from the noise? You have to consume the reviews across different platforms, not just one platform. I could pay someone to write good reviews on Google Maps, but I couldn't necessarily pay someone to do it on every single platform. I wouldn't be that thoughtful. So definitely if your product is pretty bad, overhyped, then someone's going to say something negative on YouTube or Reddit is going to see.

对。

Yeah.

它会给你一个相当平衡的视角。比如，这里有五个不同的选项。想象一下，在几秒或几分钟内读完三四十个标签页，然后告诉你该怎么做，而且它还了解你、你的需求、你的预算，一切。它有记忆。Comet 可以回答关于你自己的问题，给你建议如何做得更好。所以我认为，AI 的意义不在于用户是老是少，是更习惯传统工具还是快速接受新事物，而在于实用性。

And it gives you a pretty balanced perspective. Okay, here are the five different things. Imagine reading like 30 or 40 tabs in a matter of few seconds or minutes, then telling you what to do, and it also understands you, your needs, your budget, everything, right? It has memory. Comet can answer questions about yourself and give you tips on how to do better. So I think that's the kind of way in which we feel AI is less about whether someone's old or young, whether someone is more used to traditional tools or quick to change and adopt new things. It's truly about the utility.

我再举个例子。我在为与摩托罗拉的合作伙伴关系做演讲。我让 Comet 去读 Jens 和史蒂夫·乔布斯的所有演讲，还有我公关人员写的当前草稿，然后让它稍微编辑一下，让内容更宏大，但保留我的说话风格。所以它看了我的采访，也看了那些人的采访，然后试着融合一下，我想说一些稍微更有启发性的话，但不想显得不真实。它做得非常好，好到我觉得我本来会花 1000 美元请人做这个。

But I'll give you another example. I was doing a stage talk for a partnership with Motorola. I asked Comet to go read all the talks given by Jens and Steve Jobs and everything, and also read the current draft that was written by my comms person, and then I want you to edit it a little bit to feel a little more grandiose, but I still want to preserve how I speak. So you watch my interviews too and you watch these guys' interviews and you try to blend a little where I want to say some things that slightly feel more inspiring but I don't want to appear inauthentic. And you did a great job actually. You did such a good job that I was like damn, I would have paid someone like $1,000 for that.

我们转到 2030 年。购物会是什么样子？每个人都有一个可以对话的 AI 智能体。所以亚马逊应用死了。沃尔玛应用……

Let's transition to 2030. What does shopping look like? Everybody has an AI agent that they talk to. So Amazon app is dead. Walmart app is...

我不会说它死了。从根本上说，亚马逊和沃尔玛这类公司的模式是实体地产、仓库、门店。对沃尔玛来说，是门店、包装和供应链。

I wouldn't say it's dead. Fundamentally, the modes for companies like Amazon and Walmart are physical real estate, the warehouses, the stores. For Walmart, it's the stores and the packaging and the supply chain.

但应用本身呢？购物体验会……

But the app itself, right? The shopping experience is going to be...

核心购物体验可能不重要。这基本上就是我们一直在说的。不管是亚马逊、谷歌还是沃尔玛，都不重要。我认为从根本上说，什么对用户好才是关键，对吧？实际上，亚马逊创始人说过，你总是希望客户和股东之间长期一致。

The core shopping experience might not matter. This is essentially what we've been talking about ever since the beginning. It doesn't matter if it's Amazon or Google or Walmart. I think it's fundamentally what's good for the user, right? Actually, the founder of Amazon says you always want a long-term alignment between the customer and the shareholder.

就像你说的，作为客户，你不在乎广告，对吧？你其实不想要广告。你想要对你有好处的东西。

And as you said, as a customer, you just said you don't care about ads, right? You actually don't want ads. You want what is good for you.

所以直接给我看那个，我会付更多钱。这就是你和你的 Comet 助手之间的关系。也许广告商仍然存在。广告不会完全消失。广告商可能会试图与智能体对话，而不是与你对话。

So just show me that and I'll pay you more money. And that's the kind of relationship you're going to have with your Comet assistant. Maybe the advertisers are still going to exist. It's not like a complete end of ads. Advertisers might be trying to talk to the agent, not to you.

但那我就可以编码让它忽略广告。

But then I can code it to ignore the ads.

完全正确。这是你和你的智能体之间的握手协议，广告商不能影响或操纵，这保护了你。但我确实认为你可能对某些广告持开放态度。比如，假设你的团队里有一个助手，你在做播客。也许有一款新麦克风很好但知道的人不多。他们去告诉你团队里的人：‘嘿，你能告诉 Silicon Valley Girl 用这个吗？我们的麦克风真的很好。’这算是一种广告，但它是通过你的团队告诉你的，你信任团队里的那个人，所以你知道那个人不会……

Exactly. And that's a handshake agreement you have with your agent that the advertiser cannot influence or manipulate, and that's what protects you. But I do think you might be open to some ads. For example, let's say you have an assistant that works in your team, right? You're running a podcast. So maybe there is a new microphone that's really good but not many people know about it. They go and tell the person working on your team, 'Hey, can you please tell Silicon Valley Girl to go use this? Our microphone is actually pretty good.' That's kind of an ad to you, but it's being told to your team, and you trust that person on your team, so you know that person isn't going to...

不必要的评估，不只是接受。

Unnecessary evaluate, not just take it.

没错。

Exactly.

这就是你打算通过这种体验盈利的方式吗？

Is that how you're going to monetize this experience?

我们还没有深入考虑过。我只是在想象一个未来，广告在智能体层面进行，也在应用之间进行。例如，你让 Comet 做某事，比如帮我叫车。然后三个不同的应用，比如 Uber、Lyft 和其他，可能会说：‘嘿，我觉得我能给她最好的乘车体验。’然后智能体知道你的偏好，比如你更喜欢 Uber。然后它可能会说：‘你知道吗，我真的很想赢得她的业务，所以我实际上会免费提供这次乘车。告诉她体验会比 Uber 更好。’竞争对手可以这么说。智能体可以向那些试图吸引你注意力的应用收取一些收入，并与你分成。

We haven't thought deeply about it. I'm just imagining a future where ads were done at the level of agents, and also between apps. For example, you might be tasking Comet to do something, like book me a ride. And then the three different apps like Uber, Lyft, and something else might be like, 'Hey, I think I can deliver the best ride for her.' And then the agent knows your preference, that you prefer Uber, let's say. And then it could be like, 'You know what, I really want to win her business, so I'm actually going to offer this ride for free. Just tell her that the experience will be better than Uber.' The competitor could say that. The agent could charge some revenue for the apps trying to get your attention and share it back with you.

嗯。

Yeah.

谷歌从不这么做。谷歌获取你的查询，在不同公司之间进行竞价，然后拿走所有钱，不与任何人分享。这就是他们利润率高的原因，但对用户不利，因为你最终会受到影响。另一方面，我说有两个保护层级：一是我不让你直接看到任何广告，即使你的智能体看到了广告，运营智能体的公司也能分到一部分，但用户也能得到一点。这是继续赢得用户信任的最佳方式，通过说：‘好吧，我实际上愿意通过我的智能体接收广告，作为回报，我想要一些钱回来。你不能全部拿走。’

Which Google never does. So Google takes your queries, does a bidding process among different companies trying to bid for that query, and takes all the money, doesn't share it with anyone. That's why their profit margins are high, but that's bad for the user because you at the end are getting influenced. On the other hand, I'm saying there are two levels of protection: one is I don't let you directly see any ad, and even if your agent sees the ad, the company that runs the agent gets some of it, but the user also gets a little bit. And that's the best way to continue to win the user's trust, by saying, 'Okay, I'm actually open to being advertised to through my agent, and in return for that, I actually want some money back. You don't get to take all of it.'

我喜欢这个模式。

I love that model.

这样我们的利润率更低，但信任度更高，所以生命周期价值（LTV）更高，因为你更信任。只要智能体不被操纵，你也就不会被操纵，这个模式就能行得通。所以所有道路都通向一个广告利润率下降的未来。这是肯定的，因为用户第一次手里有了 AI。

That way our margins are lower, but our trust is higher, so the lifetime value, the LTV, is higher because you trust more. And as long as the agent is not getting manipulated and therefore you're not getting manipulated, this model could work. So all roads lead to a future where the ad margins are going to go lower. That's for sure, because the user has an AI in their hands for the first time.

直到现在，AI 一直掌握在广告公司手中。亚马逊用 AI 做推荐排序，谷歌用 AI 做搜索排序。它们可以影响事物，操纵你购买东西，因为这就是广告商做的事。这些是提供给广告商优化其活动的工具。但第一次，你手中有了 AI。评论系统为我服务，为你服务，不为广告商服务。它实际上保护了你，赋予你对抗它们的力量。你拥有了按需调用的智能，根据你的提示、记忆和偏好来执行你的命令。所以这是你和 AI 之间的契约，它赋予你对抗大科技公司的力量。

Until now, AI existed in the hands of advertising companies. Amazon has AI for recommendation ranking, Google has AI for search ranking. They can influence things and manipulate you into buying stuff because that's what advertisers do. Those are the tools provided to advertisers to optimize their campaigns. But for the first time, you have an AI in your hands. The comment system works for me. It works for you. It doesn't work for the advertiser. It actually protects you. It gives you power against them. You have intelligence on demand to summon and do your bidding based on your prompt, your memory, and your preferences. So that's a contract between you and the AI, and that gives you power against big tech.

因此，部分资金可以回流到你手中。我喜欢这一点，我喜欢你思考问题的方式。我希望每个创建 AI 智能体的人都像你这样思考。

And therefore some of that money can flow back to you. I love that. I love the way you think about that. I hope everybody who's creating AI agents thinks the way you do.

是的。

Yeah.

其他工作呢？我让 Perplexity 分析我的投资组合。我问它是否应该解雇我的理财顾问，它说，是的，你应该立即这么做。这是我第一次看到 AI 给出如此明确的回答，因为它说，你知道，他正在使用的策略是购买非常昂贵的共同基金。这些职业——理财顾问、房地产经纪人——会怎么样？因为 Perplexity 现在就能做到，它为我制定了重新平衡的完整策略。

What about other jobs? So I asked Perplexity to analyze my portfolio. I asked it whether I should fire my financial advisor, and it was like, yes, you should do this immediately. That's the first time I saw something that definitive from AI because it said, you know, this is a strategy he's utilizing, he's buying very expensive mutual funds. What's going to happen to these professions: financial advisors, real estate brokers? Because Perplexity does it now and it created the whole strategy for me to rebalance and everything.

是的，我考虑过这个问题。实际上，我的想法是，房地产经纪人从 Redfin 或 Zillow 获取房源信息，借助评论或浏览器。你可以设置一个定期运行的任务。虽然现在还不能完全实现，但很快就会实现。你可以设置一个任务说：每天早上，查看 Redfin 上所有新房源。这是我的偏好。Redfin 已经有了一些类似功能，但非常硬编码，有很多下拉选择器和过滤器，效果并不理想。但你可以有这样的功能：每当有低于某个预算且位于这个街区的房子，就给我发送推送通知，并代表我提交申请，安排看房或参观。你可以访问我的日历，一切都能做到。

Yeah, I've thought about this. Actually, the way I think about it is like real estate agents getting the listings on Redfin or Zillow with a comment or browser. You could just have a recurring task that's running. It's not there yet, but it's going to work pretty soon. You could have a task that says, every morning, look at all the new listings on Redfin. These are my preferences. Redfin already has some of these features, but it's very hardcoded and there are a lot of selector drops and filters that don't really work properly. But you could have something like: every time there's a house below a certain budget and in this neighborhood, send me a push notification and submit an application on my behalf and schedule a meeting or a visit. You have access to my calendar, everything you can do.

所以，就坚持这个想法。

So, just stick with that.

所以，如果你的经纪人只给你看 Redfin 和 Zillow 上的房源，我认为他们毫无用处。但如果你的经纪人帮你做非公开市场交易，他们仍然很有用，因为那些东西在网上不存在。

So, if your agent only shows you listings on Redfin and Zillow, I think they're useless. But if your agent is doing off-market deals for you, they're still pretty useful because those things do not exist on the web.

好的。

Okay.

我觉得，拥有智能和算力在你手中，归根结底意味着你可以推动现有行业更努力地为你工作。私人财富管理或理财顾问也是如此。如果理财顾问的工作是告诉你买什么股票，那么我认为他们不会比 Perplexity 更聪明。你可以召唤 Perplexity，让它阅读所有股票的分析师报告和所有早间新闻，然后回来告诉你对你的投资组合的影响。如果沃伦·巴菲特拥有和你完全一样的投资组合，他会怎么做？它能回答这样的问题，而任何理财顾问都做不到。

I feel like having the power of intelligence and compute at your hands always comes back to that: you can push the existing industry to work harder for you. Same thing with private wealth management or financial advisors. If the job of the financial advisor is to tell you what stocks to buy, then I don't think they're going to be smarter than Perplexity, which you can summon and ask to read all the analyst reports on every stock and all the morning news, and it comes back to tell you the impact on your portfolio. What would Warren Buffett do if he had exactly your portfolio? It can answer questions like that that no financial advisor can.

但如果你的理财顾问为你做更多事情，比如让你获得某些基金，如私募股权或对冲基金，或者 Citadel 之类的基金，或者一些初创公司、风险投资，这些你无法通过公开市场获得，那么他们可能还是有用的，对吧？关键在于：你手中有了这种新力量。所以去问问那些拿你钱的人，让他们为你做更多工作，或者重新审视你和他们之间的契约。

But if your financial advisor does more stuff for you, like getting you access to certain funds such as private equity or hedge funds, or access to funds of Citadel or something like that, or some startups, venture capital which you cannot get through just the public markets, then maybe they're useful, right? It's all about: you have this new power in your hands. So go and ask the people who are taking your money to do more work for you, or the contract between you and them.

而且那些入门级工作似乎正在消失，对吧？因为他们是初级理财顾问，只能做管理股票之类的事情。你认为会发生什么？

And it looks like those entry-level jobs are disappearing, right? Because they are an entry-level financial adviser. You can only do like managing stocks. What do you think is going to happen?

是的，我当然这么认为。人们需要更进一步。如果你只是管理投资组合，那你做的事情并不多。我也有一些私人财富经理，我认为我使用他们的方式是帮我转移资金之类的事情，因为我没有时间自己管理这一切。他们用时间帮助我，也给我一些关于如何节税等方面的建议。所以你的附加值应该扩展到不仅仅是管理投资组合。

Yeah, I certainly think so. People need to push further. If you're just literally managing a portfolio, you're not doing much. I have some private wealth managers too, and I think the way I use them is helping me move money around those kinds of things because I don't have time to manage all this myself. They help me through their time and also give me some advice on how to save taxes, those kinds of things. So your value add should expand further than just managing a portfolio.

如果我们将这一点应用到任何市场，作为应届毕业生，你如何做到这一点？你认为人们仍然需要像你一样攻读博士学位，还是那个时代已经结束了？

If we apply this to any market, how can you do this as a recent graduate? And do you believe people still need to do their PhDs like you did, or is that era over?

你在博士阶段学到的基本东西是学会如何学习。我在博士期间做了很多研究，但今天已经不重要了。但我获得了学习新课题、深入挖掘、理解一切、收集信息、咨询最佳专家的能力——顺便说一句，这已经不再需要了，因为你有像 Perplexity 这样的工具。但可以说，你有了提出正确问题的能力，然后获得一定水平的学科专业知识，从而做出决策并拓展。这就是你获得的技能。

The fundamental thing you learn in a PhD is learning to learn. I did a lot of research in my PhD which doesn't matter today. But I got the ability to go learn a new topic, dig deep, understand everything, gather information, consult whoever is the best expert—which by the way is not needed anymore because you have tools like Perplexity. But let's say you have the ability to ask the right questions and then you acquire a certain level of subject expertise to go make decisions and branch out. That is what you've acquired, that skill.

值得花四年时间吗？

Is it worth four years?

这就是问题所在。我想说我决定读博士有很多原因。其中一个显然是我对这个课题本身的兴趣，但另一个原因是这是我来到美国的唯一途径。很多印度人来这里读硕士。但你必须意识到，要来美国读硕士，你需要自己支付学费。我没有那个能力。而读博士，大学为你付费。硕士是你付钱给大学。这是原因之一，但不是唯一原因。所以，显然，仅仅因为有人给你一张免费进入美国的门票，并不意味着你应该读博士，因为这是一段相当艰苦的旅程。我对这个课题有深厚的热情。我热爱 AI。我热爱深入挖掘事物的能力。我本质上对深入任何课题都充满好奇。所以，如果你认为自己具备这些特质，那么我认为这段旅程在你的人生中非常非常有用。我不认为你需要博士学位才能为 AI 这个课题做出积极贡献，无论是作为研究员、工程师还是企业家。

That's the thing. I would say I decided to do a PhD for many reasons. One of them obviously was my own interest in the topic, but the other was that it was my only way to get to America. A lot of Indians come here for masters. But you've got to realize that in order to come here for a masters, you need to be able to fund your own tuition fee. I didn't have that ability. So PhD, the university pays for you. Masters, you pay the university. That was one of my reasons, not the only reason. So obviously just because someone gives you a free entry ticket to the United States doesn't mean you should do a PhD, because it's a pretty grueling journey. I had a deep passion for the topic. I loved AI. I love the ability to dig deep into things. I'm fundamentally curious about learning deeply about any topic. So if those are attributes that you think you align with, I think that journey is very, very useful in your life. I don't believe that you need a PhD to be contributing positively to AI as a topic, either as a researcher, engineer, or entrepreneur.

但我觉得，如果你退一步看，超越物质目标，从做人的根本出发，你会变得更加脚踏实地，更愿意追求真理、不断质疑、用苏格拉底式的方法学习，以及学会如何学习。

But I think if you zoom out and think beyond materialistic goals, like just foundationally as a person, you become a lot more grounded in the willingness to seek truth, the relentless questioning, the Socratic method of learning, and learning to learn.

我觉得这三种品质正体现了那句谚语：‘授人以鱼，不如授人以渔。’这就是我对博士学位的感受。

And I feel like these three qualities are reflective of this proverb: 'Give a man a fish, you feed him for a day. Teach a man to fish, you feed him for a lifetime.' That's how I feel about a PhD.

所以，如果你真的想成为一个喜欢学习新事物的人，我仍然会建议你去读博。也许我可以停下现在做的事，去探索一个新的领域或课题，我仍然可以相信自己能学会并取得成功。

So that's why I would still recommend people to do it if you truly want to be someone who enjoys learning about new things. Maybe I can stop doing this thing and still find a new field or topic to learn, and I can still bet on myself to learn that and be successful at it.

那么，如果你现在 18 岁，在美国，必须从头开始，你还会……

So if you were 18 today, in the US, and had to start all over again, would you still...

我 22 岁时本科毕业，然后来到这里。

When I was 22, I finished my undergrad and came here.

但如果你在美国，必须从零开始你的职业生涯，你会怎么做？

But if you were in the US and had to start your career from scratch, what would you do?

听着，我当然会追求自己的兴趣，但 18 岁还太早，不知道自己真正感兴趣的是什么。很多人以为自己知道，但我觉得 18 岁还太早。所以我会随便选一件事，然后努力把它做好。我认为人们低估了内在自信的重要性——相信自己，无论做什么都能做好。

Look, I would definitely pursue my interests, but 18 is too early to know what you're interested in. A lot of people think they know, but I feel 18 is still pretty early. So I would just pick something and try to do it well. I think people underestimate the importance of having that inner confidence to bet on yourself that no matter what you do, you can do it well.

所以，除非你已经在生活中做过类似的事情，否则就去真正做一件事，全身心投入，持续一段时间，因为只做几个月很难精通。你需要一两年才能真正变得非常出色，达到顶尖水平。可以是编程、数学、AI，或者写应用。什么都行。要真正变得非常优秀，需要时间。

So unless you've done some things like that in your life already, try to actually go and do something, commit yourself to one thing and do it for a sustained period of time, because it's very hard to be good at something if you just do it for a couple of months. You need a year or two to actually get really good, top-notch at something. It could be programming, it could be math, it could be AI, it could be writing apps. Doesn't matter. It takes time to actually be really, really good.

所以我建议他们深入钻研一件事。可以是本科教育，也可以是在现有公司找一份软件工作。如果他们不给你全职，你可以从实习生做起，然后转正，投入大量时间。

So I would suggest them to go deep into something. It could be undergrad education. It could be finding a software job at one of the existing companies. You could start off as an intern if they don't hire you full-time, then convert and pour in the hours.

我后悔看 YouTube 视频——我知道你是 YouTuber，所以这么说有点奇怪，而且我们也在做 YouTube 节目，所以是的，你应该看 YouTube 并从中获取知识。但我记得我和桑达尔聊天时说过，YouTube 是个很棒的 App，但我希望它更专注于知识和学习，他同意我的看法，因为它不只是那样，对吧？

I regret watching YouTube videos—and I know you're a YouTuber, so it's kind of weird saying this, and we're doing a whole YouTube thing, so yeah, you should watch YouTube and get knowledge out of it. But I remember telling Sundar when I talked to him that YouTube is a great app, but I wish you focused more on knowledge and learning, and he agreed with me because it's not just that, right?

所以我后悔在 YouTube 上浪费了大量时间看没用的东西。当你很年轻的时候，好好利用时间，和那些能激励你变得更好的同伴在一起，多学习，成为一台学习机器。或者非常非常努力地工作。然后你就能获得那种能力，去做好任何事。你可以相信自己。

So I regret the amount of time I wasted on YouTube just watching useless stuff. When you're very young, just utilize the time, surround yourself with peers who would push you to be better, learn a lot, be a learning machine. Or work really, really hard. And then you acquire that ability to go and be good at anything. You can bet on yourself.

好的。比如你如何应对生活中的困难任务？你只有在有信心自己能坚持下去、事情不顺时也能看到隧道尽头的光明时，才能接受挑战。你一生中至少要有过一次这样的经历，才能为接下来的事情相信自己……

Okay. Like how do you take on hard tasks in life? You can only take it on if you have some confidence that you know you can keep persisting and things are not going your way, and you'll see the light at the end of the tunnel. You have to have done it at least once in your life to bet on yourself for the next set of...

你的自信就是来自这里吗？

Is that where your confidence comes from?

是的。我经历过……

Yeah. I've done...

你就像说‘我们要超越谷歌，我们要拿下这个。’这种自信从何而来？

You're like, 'We're going to take over Google. We're going to take over this.' Where does it come from?

你不需要超越；你可以用自己的方式成功。但当你有一个像谷歌这样的竞争对手时，这非常困难。这就是让大多数人退缩的原因。你怎么在一个领域里竞争？比如我们刚开始时，有其他初创公司也在做同样的事；它们现在都不在了。它们要么被收购，要么转型了。你的竞争对手不断消失，而你一直在生存、生存、生存。就像《鱿鱼游戏》一样。

You don't need to take over; you can just be successful in your own way. But when you have a competitor like Google, it's very hard. That's what turns most people off. How do you compete in a space where, let's say when we started, there were some other startups trying to do the same thing; they're all not around anymore. They either sold or pivoted. Your set of competitors keep going away, and you're surviving, surviving, surviving. It's like Squid Games.

最后你发现自己身处一个只有谷歌、OpenAI 和你自己的竞技场。从资金来看，我们筹集了 10 亿美元。OpenAI 筹集了近 700 亿或 800 亿美元。谷歌仅现金就有 1000 亿美元，而且每年收入 2000-3000 亿美元。所以你显然是在和巨兽竞争。因此，你必须相信自己，仍然要想办法，尽管全世界都在对你喊：你必须卖给苹果、卖给 Meta，游戏结束了，你几年内就会倒闭。我可以拿全部积蓄打赌这家公司会一文不值。而你必须看到这一切，仍然激励自己，也激励那些信任你的人，比如投资者或在这里工作的员工。

And then finally you end up in a scenario or arena where it's just Google, OpenAI, and yourself. Funding-wise, we raised a billion dollars. OpenAI has raised close to 80 billion or 70 billion. Google has like 100 billion in just cash, and they make 200-300 billion a year in revenue. So you are obviously competing with giant monsters. So you have to bet on yourself to still figure out ways despite the whole world screaming at you that you have to sell to Apple, sell to Meta, the game is over, you're going to die in a couple years. I can bet my entire savings that this company is going to go to nothing. And you have to be able to read all that and still motivate yourself and also motivate the people who have trusted you, like the investors or the employees working here.

当你读到这些评论时，你对自己说什么？

What do you tell yourself when you read those comments?

我对自己说，我会证明你错了。

I tell myself that I'll prove you wrong.

这让你继续前进。太棒了。你认为这是让你成为今天这样的你的主要特质吗？你的自信。

And it keeps you going. That's amazing. Do you think that's your main trait that made you who you are? Your confidence.

这是我的重要特质之一。我不会说这是唯一的。我会说最终的主要特质并不是原始的或别的什么。我通常就是喜欢学习任何话题，享受其中的智力层面。那是我最重要的特质。我认为人们有多种特质，我也有。而生存本能绝对是对我非常有帮助的东西。

One of my important traits. I wouldn't say that's the singular one. I would say the single main trait at the end is not primal or anything. I generally just like to learn a lot about any topic, and I enjoy the intellectual aspects of it. That's my most important trait. I would say people have multiple traits, and so do I. And the survival instinct is definitely something that has served me very well.

我喜欢这个说法。我的意思是，移民都是幸存者。

I love it. I mean, immigrants are all survivors.

是的。

We are.

嗯。

Yeah.

你现在还是全天候工作吗？

Do you still work 24/7?

没有人能全天候工作。

No one can work 24/7.

好吧，醒着的时候都在工作。

Well, all the time you're awake.

是的。严格来说不是。我确实花时间锻炼，也花时间陪家人。所以我几乎每个醒着的时刻都在工作或思考工作。对我来说，很难不看手机、不查看通知、不在脑子里想某个问题。有时别人跟我说话时我会走神，努力保持当下专注。这些都是困难的事。而这些正是我愿意牺牲的，以便把我正在做的事情做得非常好。人们喜欢想，‘哦，工作就是工作。’

Yeah. Technically no. I do spend time working out. I do spend time with my family. So I do work or think about work almost every waking hour. So it's very hard for me to not look at my phone or not check notifications, not be thinking about some problem in my head. Zoning out sometimes when someone's talking to me, trying to stay present in the moment. These are all hard things. And so these are the things I would sacrifice in order to do whatever I'm doing really well. People like to think, 'Oh, work is work.'

你回家后，就关掉 Slack 和笔记本电脑，再也不打开，只是专注在当下。我做不到。即使散步时，我也在思考事情，还会接电话。

After you go back home, you just shut down Slack and your laptop, you never open it and you're just staying present in the moment at home. I cannot do that. And even on a walk, I'm thinking about something there, there's some phone calls.

所以，如果那算工作的话，是的。

So yeah, if that counts as work, yes.

我妈妈在我小时候总是说，任何不是高质量、不分心的时间都不算真正的工作。我一直相信这一点。是的，我可以在散步时回复 Slack，但那并不意味着我真的在工作。有些人喜欢把那算作工作时间。我知道埃隆说他每天工作 12 小时，我肯定他把这些都算进去了。但真正让我觉得高效的时间，是我专注于产品、思考如何改进，或者与内部人员讨论如何让应用增长、加入新想法、改进模型编排、降低成本、推出新功能的时候。那些时刻我处于不被打扰的状态，那才是我能做出最佳工作的时候。这通常发生在白天。除此之外，我大多只是在多任务处理。

I mean I've always, my mom like whenever I was young always said like anything that's not quality time, undistracted time, is not really work. And I've always believed that. Yes, I can be replying on Slack while I'm taking a walk. That doesn't mean I'm really working. Some people like to count that as hours. I know Elon says he works 12 hours a day. I'm sure he counts all that. But truly, the time I spend where I'm just looking at the product and thinking how to improve it, or talking to some of our people internally on what we can do to grow the app more, or what new ideas we can put into the app to make it even better, how can we improve the orchestration of the models even better, how can we reduce the cost, how can we ship new features, what new things we could be doing—those are the times when I think I'm in an uninterrupted state, and that's where I feel I can do my best work. That usually happens during the day. Other than that, I'm mostly just multitasking.

最后几个问题，给所有想在 AI 领域创业的人。你认为机会还存在吗？除了大家都在做的 GPT 包装器，你觉得还有机会吗，还是人们应该转向其他方向？他们应该关注哪些领域？

Couple last questions for everyone who's trying to build something in AI. Do you think opportunity still exists? Aside from like everyone's building a GPT wrapper, right? Do you still think there's still opportunity for that, or people should switch, and what are the areas that they should be looking at?

现实地说，难度大大增加了。是的，你想做个 GPT 包装器，因为自己建模型要难得多。

Realistically, it's gotten much harder. Yes, you would want to build a GPT wrapper because building your own model is way harder.

几乎不可能……

It's almost impossible for...

几乎不可能，我甚至认为人们不应该把建模型看作一次性的事。那是每个人都会犯的错误。在我们刚开始时，任何试图建模型的人，我都会问：‘启动要花多少钱？’我甚至有一个投资人，现在去了 Meta，他说：‘训练一个 GPT-3 模型要花 20 万美元，你为什么做不到？我给你一百万，你花 20 万在模型上。’但问题不在于那一个模型。而在于不断迭代模型、建设下一个集群、招聘人才来研究下一个模型应该是什么，并持续迭代，降低成本，提升能力，改进推理等等。这需要一个实验室、一个集群、人员、里程碑，以及将所有这些模型转化为产品、业务和收入的方法。然后你基本上会说：‘哦，你必须再建一个 OpenAI。’那你是什么？你为什么存在？人才套利在哪里？所以我认为做产品至少在统计上更可行一些。但即使在那里，你也必须与那些自己也在做产品的实验室竞争。

Almost impossible, and I don't even think people should be looking at it as building a model once. That's the mistake everybody made. Whoever tried to build models around the time we started, I was like, 'Yeah, how much does it cost to launch?' I even had an investor who now went to Meta and he's like, 'Oh, it's going to cost $200,000 to train a GPT-3 model, so why can't you do it? I'll give you a million and you spend 200k on the model.' It's not about that one model. It's about that relentless exercise of iterating the models, building the next cluster, hiring the talent to figure out what the next model should be, and continually iterating, bringing down the cost, improving the capabilities, improving the reasoning, all that stuff. That requires a lab and a cluster and the people and the milestones and some way to actually turn all those models into a product and a business and revenue. And then you end up basically saying, 'Oh, you have to build another OpenAI.' Then what are you? Why do you need to exist? Where's the talent arbitrage? So that's why I think product is actually at least a little more statistically probable. But even there, you have to compete with the likes of the labs themselves, which are building their own products.

还有大科技公司和像我们这样的新贵。没有人会说：‘哦，Perplexity 只做研究，我可以去做电商。’

And the big tech companies and upstarts like us. And no one's going to be like, 'Oh, Perplexity is just going to do research. I can go do the commerce stuff.'

不，没人只做电商。

No, no one is in commerce.

完全正确。每个人都在做横向产品，这也适用于我们。‘哦，ChatGPT 永远不会做购物，永远不会做旅行，不会做金融。’这些都不对。每个人都会做任何有效的事情。

Exactly right. Like everyone's building horizontal products, and that applies to us too. 'Oh, ChatGPT will never do shopping. ChatGPT will never do travel. It won't do finance.' All that is not true. Everyone's going to do anything that works.

所以我想说，你可以下的赌注就是做你真正痴迷的事情，因为从根本上说，这是对你自己的赌注。

So I would just say the bet you can make is do what you truly are obsessed about, because fundamentally it's a bet on yourself.

这不是对市场的赌注，也不是对生态系统的赌注，比如竞争对手会做什么或不做什么。不要试图成为那种白板战略大师，这完全没意义。当你的想法成功并获得 1 亿或 10 亿收入时，永远要预期现有玩家会跟进，因为每个人都在寻找 AI 领域的增量收入，因为资本支出太高了。所以证明这一切的唯一方法就是将其转化为商业利润。

It's not a bet on the market. It's not a bet on the ecosystem, like what competitors will do or will not do. Don't try to be this whiteboard strategy master; it's completely pointless. When your idea works and gets like 100 million or a billion in revenue, always expect existing people to go after it because everyone's looking for that incremental revenue in AI because the capex is so high. So the only way to justify all this is to actually turn that into business profits.

然后他们会追着你。所以你能赌的唯一一件事就是你是否对一个主题如此痴迷，以至于无论有多少困难你都会去做，然后你会证明世界错了，因为你深入其中，没有人比你更关心这个问题。对我们来说，这就是为什么——你问我最重要的特质是什么——我热爱学习，而这是最好的学习应用之一。它是为我们设计的，为我们的创始人设计的；我们为自己而建。关键赌注是每个人都喜欢有一个可以召唤来为他们做事的 AI。每个人都想在生活中有更多时间。所以我认为为自己而建，希望这是世界上很多人想要的，然后你可以把它变成一个可扩展的产品和可扩展的公司。

And then they'll go after you. So the only thing you can bet on is whether you are so obsessed about a topic that you will do it anyway regardless of all the odds stacked against you, and then you'll prove the world wrong because you go so far deep into that and no one cared about the problem more than you did. For us, for me, that's why the whole—you asked me what my most important trait is—I love learning, and this is one of the best apps for learning. It's designed for us. It's designed for our founders; we built it for ourselves. The key bet was everybody in the world loves having an AI that they can summon to go do stuff for them. Everybody wants more time in their lives. So I think build for yourself, hopefully that's a thing that a lot of people in the world want, and therefore you can turn it into a scalable product and a scalable company.

然后在某个阶段被收购，我猜。谢谢，回答得很好。最后一个问题，除了 Perplexity，你还用其他 AI 应用吗？你最喜欢的前三个是什么？

And get acquired at some stage, I guess. Thank you. That's a great answer. My last question, do you use anything aside from Perplexity in terms of AI apps, and what are your top three favorites then?

作为工作的一部分，我需要使用其他应用，看看它们在哪些方面比我们做得更好，因为没有人能在所有方面都最好。所以我手机上有所有 AI 应用，比如 ChatGPT、Grok、Gemini、Claude——几乎都有。我没有 Copilot 之类的，主要是因为它们在消费者分发方面不那么主流。但在自然使用方面，我经常去尝试。我想说 ChatGPT 我测试得最多，但主要还是为了测试。

Well, as part of my job, I am required to use other apps to see what they are doing better than us, because no one's going to be best at everything. So I have all the AI apps on my phone, but you know, like ChatGPT, Grok, Gemini, Claude—almost all of them I have. I don't have Copilot and maybe some others, but mainly because they're not that mainstream in terms of consumer distribution. But in terms of organic usage, I do go pretty often and try. I would say I've tested ChatGPT the most, but again, it mostly comes down to testing.

所以你不用那些针对营养或图像编辑的包装器应用？如果你需要什么，就直接问 Perplexity？

So you don't use any wrapper apps that are specific for, I don't know, nutrition or image editing? So if you need something, you will just ask Perplexity?

是的，如果不行，我会问 ChatGPT。

Yeah, if it doesn't work, I would ask it in ChatGPT.

如果那里也不行，比如我需要编辑，我就直接自己处理。如果需要图像编辑，我也做过。

And if it doesn't work there also, for example if I need to edit, I'll just do my usage right. If I need to do image editing, I've done that.

如果 ChatGPT 和 Perplexity 都不行，那我就自己写代码来做。

And if it doesn't work on both ChatGPT and Perplexity, then I would just write code myself and do it.

我喜欢这样。

I love that.

但那只是我，那是你。

But that's just me, that's you.

是的，搜索也一样。比如我们有一个集成的网球课程，我喜欢网球，喜欢看实时比分，所以我会去谷歌查比分。

Yeah, same thing with search. Like if it does, for example we have an integrated tennis course and I love tennis, I like following live scores, so I go to Google and get the score.

我们并没有一直做货币换算，还在改进中。所以我用谷歌的一些小工具，有时像大海捞针一样找链接，可能还不是最好的产品。

We haven't all done currency conversions all the time. We are still working on that. So I use Google for some of the widgets they have, and sometimes finding these needles in a haystack like links, maybe not the best product on that yet.

你说的链接是什么意思？

What do you mean by links?

比如有些博客文章里的老链接，我小时候读过，记得很清楚。Perplexity 有时能帮我找到，但链接太旧了。所以我仍然习惯用谷歌的各种宏指令，比如我知道怎么加日期后缀之类的。那是因为我们就是这样长大的，对吧？那时候，谷歌搜索其实是一门技能。

Like there are some really old links on some blog post that I definitely remember. I read it when I was a kid, and Perplexity sometimes helps me find that, but it's very old links. So I still tend to use Google with all the macros, like I know how to suffix the dates and all that stuff. That's because we grew up like that, right? Back then, googling was actually a skill.

所以如果你知道怎么用那些特定短语搜索链接……

So if you knew how to search for links with those specific phrases...

对，对。加上‘或’包含和那些日期，你就能找到任何想要的东西。这也给了我改进产品的想法。也许你可以教 AI 自己用这些宏指令，然后针对你正在构建的索引，训练它用这些关键词参数来检索。所以，我的生活里，主要用 Perplexity、谷歌、ChatGPT 这三个应用。偶尔用 Midjourney，因为它的审美质量很好。嗯，差不多就这些了。

Yeah. Yeah. And 'or' contains and those dates, you can get whatever you want. And that also gives me ideas on how to improve the product too. Maybe you teach the AI to use some of these macros yourself, and then against your index you're building, train it to retrieve these things with these keyword arguments. So yeah, my life, I think I mostly use Perplexity, Google, ChatGPT. These are the three main apps I use. Midjourney sometimes because it has a pretty good aesthetic quality. And yeah, I feel like that's it.

明白了，谢谢。

Got it. Thank you.

我觉得这答案有点无聊。

I mean, it's a kind of a boring answer, I guess.

但那就是你，对吧？

But that's you, right?

对，那就是我。

Yeah, that's me.

那就是你。

That's you.

关键是，Comet 现在能让我做任何事。我甚至不需要打开这些独立应用。Comet 作为浏览器，搜索效果甚至比 Perplexity 还好，因为……

The thing is, Comet lets me do anything right now. I don't even have to go to these separate apps. Comet as a browser, it actually does better search than even Perplexity because...

你会让所有人都免费使用所有功能吗？

Will you make it free for everyone with all the features?

不是所有功能，我们总得赚点钱。但几乎所有功能，我们认为都会对所有人开放。

Yeah, not every feature. We have to make some money here. But almost all the features we think will be available to everybody.

它也会在手机上用，对吧？还是说因为我只在桌面端用？

And it's going to be on mobile too, right? Or is it because I'm only using it on desktop?

没错。我们今天刚宣布，在 Play 商店可以预注册获取。App Store 的预注册很快也会公布。到今年年底，我想应该会在 iOS 和 Android 上可用，同时在 Windows 和 Mac 上全面推出。

That's right. So, we just announced today that on the Play Store you can pre-register to get it. App Store pre-registering will be announced shortly. By end of the year, I think we should have it available on iOS and Android. And also have it be generally available on both Windows and Mac.

我喜欢。确实如此。

I love it. It generally is.

顺便说一句，这就是 Comet 的特点。你可以召唤它去用 Money、用 Photoshop，它直接帮你搞定。你甚至不需要显式打开这些应用。所以它是一个超级应用，一个全能应用。

So that's the thing with Comet, by the way. You can summon it to go use Money, go use Photoshop, and it just does it for you. You don't even have to have these as apps you explicitly open. So it's a super app. It's an everything app.

是的，太迷人了。非常感谢。这太鼓舞人心了。我感觉我们现在就在触碰未来。谢谢，非常感谢。

It is. It is fascinating. Thank you so much. That is so inspiring. I feel like we're tapping into the future right now. Thank you. Thanks a lot.