Will Perplexity.AI's Answer Engine Attract Five Percent of Google Users
January 10, 2024
By Ignatius Chithelen*
What is the weight of the moon, I asked Perplexity. AI today. The new search - or answer - engine’s reply in 1.28 seconds was: “The Moon’s mass is approximately 7.35 x 10^22 Kg, which is about 1.2% of Earth’s mass. However, the weight of the Moon is not directly measurable, as it is in orbit around Earth and therefore experiences no gravitational force.”
This was also the answer given by a scientist on NASA’s website.
The same question asked on Google came up with 351 million results in 0.44 seconds, with an answer from The Guardian newspaper on top.
At a time when artificial intelligence (AI) hallucinations cause concern, Perplexity is on a grand mission “to build the fastest and most accurate platform for answering any question,” Aravind Srinivas, co-founder and Chief Executive of the San Francisco based startup, stated in a blog post last week.
Since launching in December 2022, Perplexity has grown to more than 10 million monthly active users, including one million mobile users. It answered over half a billion queries in 2023, mainly from academics, students, knowledge workers and business to business vendors, who rely on frequent research for their day-to-day work.
Perplexity does not require perfectly phrased questions. It chats with a user: it asks, listens, and refines its search. Users get more relevant answers, including sources and citations, than from Google, Bing and other traditional search engines, AI chatbots, or research tools.
Perplexity was founded on the belief that searching for information should be “free from the influence of advertising-driven models,” Srinivas, 29-years-old, states in a blog post.
Perplexity’s answer engine is powered by artificial intelligence (AI) tools GPT-4, and Claude 2. GPT-4, the latest deep learning AI model from OpenAI, accepts image and text inputs and emits text outputs.
Claude 2 is an AI assistant created by Anthropic, a San Francisco based startup. In September 2023, Amazon announced it will invest up to $4 billion in Anthropic, which in turn will use Amazon’s web services (AWS) as its primary cloud services provider and also buy AWS’s chips.
Prior to co-founding Perplexity, from September 2021 to August 2022, Srinivas was a research scientist working on language and diffusion generative models at San Francisco based OpenAI. Earlier in 2018, he spent four months at the company as a research intern.
OpenAI is valued at $100 billion, according to Bloomberg. Its launch of ChatGPT and DALL-E, in late 2022, sparked a boom in the use of AI tools. GPT-4, used by Perplexity, can follow complex instructions in natural language and solve difficult problems with accuracy. DALL-E, which is integrated into GPT-4, is an AI system that can create realistic images and art from a description in natural language.
Users of Perplexity’s free plan are allowed five queries every four hours. Paid subscribers can search up to 600 times a day. Subscriptions cost $20 per month or $200 per year.
Paid subscribers can upload one or multiple files, including emails, articles, PDFs, and images, and explore their contents with questions. They can organize search results in a library by project or topic, invite contributors and switch privacy settings. They can also ask questions about their files and search the web, without leaving the platform.
Perplexity is yet to disclose the number of paid subscribers using its platform. It has annual revenues of between $5 million and $10 million, Srinivas told The Wall Street Journal.
Photo: CEO Satya Nadella announcing integration of GPT-4 with Microsoft products
Last week, Perplexity raised $74 million in a new funding round at a valuation of $520 million. Investors included Amazon founder Jeff Bezos and NVIDIA, whose computer chips power artificial intelligence (AI) tools and applications, including at Perplexity.
Perplexity “will become the go-to place for trusted information, which is why I am so excited that Perplexity is NVIDIA’s first consumer investment from our corporate arm." Jonathan Cohen, NVIDIA’s vice president of Applied Research stated in a Perplexity blog about the fund raise. Santa Clara, California, based NVIDIA has a market value of $1.3 trillion.
Perplexity launched its first answer engine in December 2022. A month after its founding in August that year, it raised $3.1 million from angel investors, including Ashish Vaswani, co-inventor of Transformer, a neural network that learns meaning by tracking relationships in sequential data like the words in a sentence. Other early investors include Abhay Parasnis, a former chief technology officer and chief product officer at Adobe, a San Jose, California, based design software vendor with a market value of $264 billion.
Aravind Srinivas co-founded Perplexity with Denis Yarats, Johnny Ho and Andy Konwinski, in part since he faced problems getting a U.S. work visa as a sole founder. Earlier, Yarats was an AI research scientist at Facebook parent Meta and a machine learning engineer at answer platform Quora. Ho worked as a quantitative trader on Wall Street and an engineer at Quora. Konwinski was among the founding team at Databricks, a data and AI company.
From May 2020 to April 2021, while working on his PhD, Srinivas was a research intern at Google’s head office in Mountain View, California. For five months in 2019, he was a research intern at DeepMind, the London based AI research group owned by Google. While at DeepMind, he was inspired to start a company by learning about the work done by Larry Page, co-founder of Google, Srinivas said at a First Mark Capital event in New York in November 2023..
Srinivas earned a PhD in computer science from the University of California, Berkeley, 2021, and dual B.Tech and M.Tech degrees in electrical engineering from the Indian Institute of Technology, Madras, 2017.
Srinivas is an angel investor mainly in AI related startups.
Photo: Sundar Pichai, CEO, Google and parent Alphabet
In May 2023, Google launched a "test" version of Search Generative Experience (SGE), an AI-based search engine which, like Perplexity, answers questions. When asked the weight of the moon, Google’s SGE gave a better answer than Perplexity: roughly 81 billion tons, if placed on the Earth’s surface. Earth weighs 81 times more than the moon.
So why has Google not yet integrated SGE into its search tab? Apparently, Google fears widespread adoption of answer engines, including its SGE, will lead to sharp declines in advertising revenues for its online search engine.
Engines like Google’s SGE and Perplexity answer search questions with one answer, unlike the 351 million results for a Google search for the weight of the moon. In most cases, users are unlikely to click on numerous ads that may appear with an answer on SGE - or Perplexity - since they have gotten the specific information they sought. This likely means far fewer advertisers will buy ads, to compete for a user's attention on an SGE answer, compared to the dozens of advertisers found on the first page of a typical Google search result.
I asked Perplexity, for instance, what is the best Indian restaurant in New York City? The answer “is subjective and can depend on individual preferences. However, some highly regarded Indian restaurants in NYC include” Dhamaka, Semma, Tamarind Tribeca and Junoon.
Google's SGE does not answer the question, instead stating: "New York City has many good Indian restaurants, including upscale restaurants, places for regional food, and other top restaurants." The search result also displays links to lists of top Indian restaurants from Yelp, Eater and other sources as well as links to the sites of more than a dozen Indian restaurants, apparently all of them ads.
Google, whose parent Alphabet has a $1.8 trillion market value, overwhelmingly dominates the traditional online search market in the U.S. with an 88% share. Even Bing, owned by Microsoft which has a $2.8 trillion market value, has been unable to get users to switch from Google. Bing's market share has stayed around 8%, despite Microsoft's Windows system controlling the operations of more than 120 million personal computers, laptops and tablets in the U.S.
In 2023, ad revenues from online search in the U.S. totaled $242 billion, with Google pocketing an estimated 90% share. Google has more than 35 billion product listings from advertisers.
Google’s apparent delay in offering SGE to all its search engine users, coupled with SGE apparently avoiding answering questions which may turn off some advertisers, provides an opportunity for Perplexity. But will users pay $200 a year for Perplexity when, for most searches, they can get free information from Google’s search engine?
Photo: Sridhar Ramaswamy, co-founder Neeva
In 2020, Sridhar Ramaswamy and Vivek Raghunathan, former senior executives at Google, co-founded Mountain View, California based Neeva. A data-driven search engine, it enabled users to control and shape their search results. Users could ask that search results, of political issues for instance, be from independent sources. Neeva, an ad-free search engine, charged users $4.95 per month.
It was reportedly using Microsoft’s Bing for basic search results. Neeva raised $78 million in funding and was valued at $300 million in early 2021. Greylock and Sequoia Capital, two leading Silicon Valley venture firms, were key investors. Greylock was an early investor in Airbnb and LinkedIn while Sequoia was an early investor in Google and Yahoo!
In March 2023, Neeva shut down, likely unable to quickly shift to using AI tools since it did not create and own its technology. It was acquired, for its technical talent, for an undisclosed amount. The buyer was Snowflake, a Bozeman, Montana, cloud-based data platform with a market value of $65 billion.
While developing and using some of its own technology, Perplexity uses tools from OpenAI, Claude and other vendors. So, Perplexity risks facing a similar fate like Neeva, especially if a new answer engine, using superior AI tools, is developed by a rival.
Srinivas says Perplexity is model agnostic. “Our goal is to develop our model to be optimal for our specific product…(and not be ) tied to any model provider,” he told Decyrpt.co. Perplexity’s success will depend on “strong execution, building teams and shipping products,” he said at the First Mark Capital event.
Microsoft, which has reportedly invested $13 billion in OpenAI, competes against Perplexity. Microsoft is integrating OpenAI’s tools in its Bing search engine as well as in a range of other products.
To survive, Perplexity needs to quickly attract millions of additional users, especially paid users, thereby enabling it to raise hundreds of millions of dollars in additional investor funding.
Unlike sales of traditional software products, Perplexity and other AI services enjoy few major cost benefits from scaling up their sales. Rising number of AI users means rising costs, which are mostly fixed, including for AI chips, OpenAI and other tools, cloud computing and storage.
To date Perplexity, which has less than 40 employees, has raised $100 million. It could build a $1 billion annual revenue business if it attracts five million paying customers, about two percent of Google’s 276 million monthly active users in the U.S. Srinivas told decrypt.co, “I'm happy with just executing on the boring task of being the most accurate answer engine in the world.”
*Ignatius Chithelen is author of Passage from India to America and Six Degrees of Education.
FOR MORE UNIQUE STORIES ON INDIANS AND INDIA:
For access to stories each week email: gitimescontact@gmail.com
or subscribe via Substack, follow via LINKEDIN or TWITTER or FACEBOOK
(c) All rights reserved. Copyright under United States Law
