As established search engine companies, most notably Google, enhance their platforms with GenAI technology, startup companies aspire to completely reimagine AI-powered search. Confronting rivals who possess a theoretical infinity of users may appear to be an insurmountable challenge. However, this emerging category of search engines is confident that it can carve out even a modest space by providing an exceptional user experience.
On January 4, one of the cohorts, Perplexity AI, revealed that it had raised $73.6 million in a capital round led by IVP. Additional investments came from NEA, Databricks Ventures, Elad Gil, who was once a vice president at Twitter, Tobi Lutke, who was CEO of Shopify, Nat Friedman, who was CEO of GitHub, and Guillermo Rauch, founder of Vercel. Nvidia and, most notably, Jeff Bezos were among the additional participants in the round.
After-funding insiders familiar with the situation informed that Perplexity is valued at $520 million. That is negligible in comparison to GenAI’s fledgling companies. However, given that Perplexity has only existed since August 2022, the ascent is still quite remarkable.
A team of engineers with expertise in artificial intelligence (AI), distributed systems (DS), databases (DB), and search engines (SE), including Johnny Ho (HO) and Andy Konwinski (AK) established Perplexity. Formerly employed at OpenAI, Srinivas, the CEO of Perplexity, researched GenAI and language models comparable to Stable Diffusion and DALL-E 3.
Perplexity presents an interface resembling that of a chatbot, enabling users to pose inquiries in natural language (e.g., “Does one burn calories while sleeping?” or “Which country receives the fewest visits?”). After receiving a summary with source citations (mainly websites and articles) from the platform’s AI, users may pose follow-up inquiries to delve more extensively into a specific topic.
“With Perplexity, users can get instant … answers to any question with full sources and citations included,” stated Srinivas. “Perplexity is for anyone and everyone who uses technology to search for information.”
The Perplexity platform is supported by a variety of GenAI models that have been both internally developed and externally licensed. Pro plan subscribers of Perplexity ($20 per month) can change models (Google’s Gemini, Mistra 7Bl, Anthropic’s Claude 2.1, and OpenAI’s GPT-4 are currently in rotation). This feature unlocks additional functionalities such as file uploads, which enable consumers to upload files including images, and let the models analyze them to generate answers (e.g., “Summarize pages 2”). Additionally, Pro plan subscribers gain unlimited access to Perplexity’s Copilot, which incorporates personal preferences into search queries.
One could argue that the experience is akin to that of Microsoft’s ChatGPT, Google’s Bard, or Bard. Even Perplexity’s chat-forward user interface resembles the most popular GenAI tools of the present day.
Other than the obvious rivals, you.com, a search engine startup, also provides tools for summarizing and citing sources that are powered by AI and optionally by GPT-4.
Perplexity, according to Srinivas, provides better search filtering and discovery capabilities than the majority of platforms. For instance, users can restrict their queries to academic papers or peruse popular search topics submitted by other platform users. It is not obvious that these entities are sufficiently unique to prevent their replication, or have not been replicated thus far. However, Perplexity’s goals exceed searching. It has initiated the provision of its proprietary GenAI models, which ostensibly benefit from enhanced performance by utilizing Perplexity’s search index and the public web, via an API accessible only to Pro customers.
The source bears doubts regarding the durability of GenAI search tools for several reasons, one of which is the high cost of operating AI models. At one point, OpenAI allocated an estimated daily expenditure of $700,000 to meet the burgeoning demand for ChatGPT. In the meantime, Microsoft’s AI code generator reportedly faces a monthly average loss of $20 per user.
The sources with knowledge of the situation informed that Perplexity’s current annual recurring revenue ranges from $5 million to $10 million. That appears reasonably sound… When the millions of dollars that frequently go into training GenAI models like the one at Perplexity are considered.
As expected, concerns regarding misuse and misinformation undoubtedly arise with GenAI search tools such as Perplexity. In the end, AI is not the most accurate summarizer; it occasionally neglects crucial information, misinterprets and exaggerates language, and fabricates facts in an extremely authoritative manner. Moreover, it is susceptible to spreading discrimination and toxicity, as recently illustrated by Perplexity’s models.
Copyright is yet another potential obstacle in the path to success for Perplexity. To generate essays, code, emails, and articles, among other things, GenAI models “learn” from examples. Many vendors, presumably Perplexity, sweep the web to feed millions to billions of these examples to include in their training datasets. Artists, authors, and other copyright holders are opposed by vendors’ claim that the fair use doctrine affords blanket protection for their web-scraping activities; they have filed litigation seeking compensation.
It is worth noting that although a growing number of GenAI vendors provide policies safeguarding their clients against IP claims, Perplexity does not offer such protection. Customers agree to “hold harmless” Perplexity from any claims, damages, or liabilities that may arise as a result of utilizing its services, as stated in the company’s terms of service. This provision absolves Perplexity of any responsibility about legal fees.
Certain plaintiffs, including The New York Times, have contended that GenAI search experiences engage in anticompetitive practices by diverting publishers’ content, consumers, and ad revenue. Whether “anticompetitive” or not, the technology is affecting traffic. 75% of the time, a search engine like Google could respond to a user’s query without necessitating a click-through to its website, according to a model from The Atlantic. (OpenAI is one of the few vendors that has signed contracts with specific news publishers; Perplexity is among the majority that has yet to do so.)
This is marketed by Srinivas as a feature, not a flaw.“[With Perplexity, there’s] no need to click on different links, compare answers, or endlessly dig for information,” he explained. “The era of sifting through SEO spam, sponsored links, and multiple sources will be replaced by a more efficient model of knowledge acquisition and sharing, propelling society into a new era of accelerated learning and research.”
Despite the numerous uncertainties surrounding Perplexity’s business model, GenAI, and consumer search in general, its investors appear unfazed. As of now, the startup has amassed more than $100 million in funding; a significant portion of that capital, according to Srinivas, is being used to develop new product functionality and expand its 39-person staff. The startup claims that it has 10 million active monthly users.
“Perplexity is intensely building a product capable of bringing the power of AI to billions,” IVP general partner Cack Wilhelm added via email. “Aravind possesses the unique ability to uphold a grand, long-term vision while shipping product relentlessly, requirements to tackle a problem as important and fundamental as search.”
