Perplexity的“答案引擎”威胁到谷歌了吗？

剑指史上数一数二的商业模式

Taking aim at one of the best business models of all times

阿拉温德·斯里尼瓦（Aravind Srinivas）被加州大学伯克利分校录取攻读博士学位时，他的母亲感到失望。和许多印度父母一样，她希望儿子能入读麻省理工学院。但这最终无碍斯里尼瓦走上康庄大道：在美国西海岸，他曾进入OpenAI和谷歌的DeepMind实习，两者后来都成为生成式AI的领导者。凭借这些经验，斯里尼瓦与他人联合创办了Perplexity，这家提供类似维基百科的快速搜索查询的生成式AI创业公司最近拿到了10亿美元的估值。在接受采访时，斯里尼瓦显得平易近人，但也雄心勃勃。他的“答案引擎”要挑战的是谷歌搜索这一史上数一数二的成功商业模式。想想当年马丁·路德是怎么挑战天主教会的。

WHEN ARAVIND SRINIVAS was accepted at the University of California, Berkeley, to do a PhD, his mother was disappointed. Like many Indian parents, she wanted him to go to the Massachusetts Institute of Technology. But things worked out after all; on the west coast he interned at OpenAI and Google’s DeepMind, both of which became leaders in generative artificial intelligence (AI). With that experience, he co-founded Perplexity, a generative-AI startup recently valued at $1bn that provides fast, Wikipedia-like responses to search queries. He is an unassuming interviewee, but an ambitious one. His “answer engine” is aimed at competing with Google search, one of the best business models of all time. Think Martin Luther taking on the Catholic church.

斯里尼瓦深受颠覆理念的熏陶。最近有一位播客主持人问他OpenAI和DeepMind在文化上有何异同，他解释道，前者由工程师主导的自由放任文化颠覆了后者执着于研究的“非常英国式”的等级文化（DeepMind在伦敦创立）。在讨论谷歌母公司Alphabet时，他也使用了颠覆理论。他没有解释Perplexity的商业模式将何以能攻击谷歌这一搜索巨头，而是援引了克莱顿·克里斯滕森（Clayton Christensen）在1997年出版的畅销管理书《创新者的窘境》（The Innovator’s Dilemma）中阐述的这个著名概念，以指出他认为Alphabet存在的致命弱点。这不是他的独家观点。之前也有人以“创新者的窘境”来解释谷歌为何受到OpenAI的ChatGPT以及You.com等其他生成式AI网站的威胁。这个说法很吸引人，但并不准确。

Mr Srinivas is a student of disruption. When a podcaster asked him recently to compare the cultures of OpenAI and DeepMind, he explained how the engineer-led, free-wheeling approach of the former disrupted what he called the research-obsessed “very British” hierarchy of the latter (which was founded in London). He resorts to disruption theory when discussing Alphabet, Google’s parent company. Rather than explaining how Perplexity’s business model will enable it to attack the search giant, he uses a celebrated concept outlined in “The Innovator’s Dilemma”, a management bestseller from 1997 by Clayton Christensen, to identify what he sees as Alphabet’s Achilles heel. He is not alone. The innovator’s dilemma has been invoked to explain why Google is threatened by OpenAI’s ChatGPT and by other generative-AI sites such as You.com. The argument is seductive. But it is off the mark.

正如克里斯滕森所述，“创新者的窘境”解释了为何新技术会摧毁大企业：如果它们选择与新兴公司竞争，就会危及它们自身的标准和品牌；如果不竞争，又可能被下一波创新浪潮淘汰。简而言之，该理论指出，老企业在取悦自己最优质的客户方面太过拿手，从不会想要向下进入低端市场。这就给了颠覆者机会。他们一开始以不够好的产品瞄准小众市场，通过不断的坚持不懈的改进，最终大获成功。这个理论也可以解释数码摄影如何最终把柯达推向破产，以及为什么苹果的iPhone颠覆的不是手机，而是笔记本电脑。

The dilemma, as presented by Christensen, explains why new technologies cause great companies to fail. If they compete with upstarts, they jeopardise their own standards and brand. If they don’t, they risk falling victim to the next wave of innovation. In a nutshell, the theory states that an incumbent is so good at pleasing its best customers that it would never dream of going downmarket. That gives insurgents an opportunity. They target a niche of the market with initially subpar products. Through relentless improvement, eventually they hit the big time. You can use it to understand how digital photography killed Kodak, and why Apple’s iPhone disrupted not mobile phones, but laptops.

斯里尼瓦再提这一理论是为说明谷歌的搜索业务何以可能由福变祸。用户点击链接几乎不耗费谷歌什么成本。但广告主需要就每次被点击竞价，这给Alphabet带来了巨额利润。生成式AI改变了这个模式。首先，给出响应的成本增加了，因为与AI相关的问答要比搜索查询消耗更多算力。其次，生成式AI提供的是答案，不是链接，因此广告主的广告粒度下降。简而言之，斯里尼瓦认为，如果Alphabet放弃搜索而转向类似Perplexity的产品，成本将上升，收入将大跌，利润率将受到影响，投资者就会弃它而去。没有利润可以损失的Perplexity正是瞄上了这一点。

Mr Srinivas brings up the theory to explain why Google’s search business could turn from a blessing to a curse. It costs Google almost nothing when users click on its links. But advertisers bid on the cost per click, providing Alphabet with whopping profit margins. Generative AI shifts the model. First, the results cost more, because AI-related Q&A uses more computing power than search queries. Second, they provide answers, not links, hence less granularity for advertisers. In short, if Alphabet were to abandon search for a Perplexity-like product, Mr Srinivas argues, costs would rise, revenues would plummet, margins would suffer and investors would head for the hills. That is where Perplexity, with no profits to jeopardise, sees something to aim at.

这在理论上说得通，但并非创新者窘境的实际案例。在克里斯滕森的表述中，在位企业忽视了颠覆者，因为后者一开始只是在市场边缘渗透，而不是正面交锋。但斯里尼瓦是堂而皇之地向谷歌发起挑战。新兴公司一般是要通过不太讲究的低成本技术抢走被忽视的客户。然而，Perplexity采用最终可能会包含广告的订阅模式，可能比谷歌更昂贵，而且答案应该会精细讲究得多（尽管并不总是准确）。

This is plausible in theory, but it is not an application of the innovator’s dilemma. In Christensen’s formulation, the incumbents overlook the insurgents because these start by nibbling at the fringes of a market, not by going head to head. Yet Mr Srinivas has openly thrown down the gauntlet to Google. Upstarts are supposed to win over underserved customers with cheap, scrappy technology. Yet Perplexity, with a subscription model that may eventually include ads, can be more expensive than Google and its answers tend to be far more polished (if not always accurate).

Perplexity看起来更像是克里斯滕森所说的“持续”创新（让好产品变得更好）的例子，而非颠覆创新。这没什么不好。但这个游戏Alphabet也能玩。它拥有研究人才和雄厚资金，可以不断改进生成式AI搜索。目前Alphabet正在试验一种名为“搜索生成体验”的AI工具，并表示此类查询的计算成本已经比刚推出时下降了80%。它有信心能利用AI更好地从广告中获取收入。与此同时，其搜索收入仍在增长，第一季度同比增长14%。这可不是“宗教改革”已至的样子。

Rather than being a disrupter, Perplexity looks more like an example of what Christensen called “sustaining” innovation—making good products better. There is nothing wrong with that. But it is a game that Alphabet can play, too. It has the researchers and deep pockets to keep improving generative-AI search. It is experimenting with an AI tool called “search generative experience”, and says the computing costs of such queries have fallen by 80% since they were first introduced. It is confident it will be able to use AI to better monetise ads. Meanwhile its search revenues continue to boom; they rose by 14% year on year in the first quarter. Not exactly the start of the Reformation.

简单来说，谷歌目前不像是面临窘境。它可以根据自身利益所在决定是否迎战Perplexity。斯里尼瓦对Perplexity优势的解释更为准确。通过从各种封闭和开源的大语言模型中收集答案，他的产品可以利用每个模型的分析优势及不同定价结构来提高性能、降低成本。Perplexity的对话能力很可能会越来越强。不难想象它与一个杀手级设备搭配后的效果——想想AI爱情电影《她》（Her）中的耳机。

In short, Google does not appear to face a dilemma at present. It can compete or not, depending on where its interests lie. Mr Srinivas does a better job explaining Perplexity’s strengths. By gleaning answers from a variety of large language models, both closed and open-source, his product can take advantage of each model’s analytical strengths, as well as their varying pricing structures, to improve performance and lower costs. It is likely to become increasingly conversational. It is not hard to imagine it pairing up with a killer device—think of the earpiece in the movie “Her”, an AI love story.

那会是一种什么设备？斯里尼瓦表示，这是个价值万亿美元的问题。但他认为，苹果的iPhone是个巨大障碍。“这是条护城河。”他拿起笔者的手机说。原因在于iPhone这个硬件与苹果的操作系统、应用商店和支付平台环环紧扣，他认为这使其几乎无法被攻破。

What such a device could be, Mr Srinivas says, is the trillion-dollar question. But he reckons there is a huge hurdle in the shape of Apple’s iPhone. “This is the moat,” he says, picking up your columnist’s device. That is because of the interplay between the hardware and Apple’s operating system, app store and payments platform, which he thinks makes it almost invincible.

他可能又错了。相比Alphabet，苹果可能更容易遭遇创新者窘境。苹果是全球最负盛誉的公司之一，专注深耕最优质客户（例如那些买得起3499美元的增强现实头显的人）。它绝不会冒损害品牌价值的风险提供制作粗糙的廉价产品。AI小设备（从吊坠到Rabbit R1想出的各种东西）制造商希望有朝一日能击败强大的iPhone，但目前仍有太多不足，苹果根本不屑反击。这看起来倒像是要出现颠覆的样子。

Again he may be wrong. Apple may be more exposed to the innovator’s dilemma than Alphabet. It is one of the world’s most reputable companies. It is laser-focused on its best customers (those, say, who can afford a $3,499 augmented-reality headset). It would never risk its brand by offering a cheap, shoddy product. Makers of AI gizmos, from pendants to whatever the Rabbit R1 thinks it is, one day hope to vanquish the mighty iPhone but remain far too flawed for Apple to bother responding to. Sounds like a recipe for disruption. ■

Generated by RSStT. The copyright belongs to the original author.

Source