Perplexity, a startup search engine with an A.I.-enabled chatbot interface, has announced a host of new features aimed at staying ahead of the increasingly crowded field.
The San Francisco-based company, which was founded by two A.I. Researchers who had previously worked at Meta, OpenAI, Google, and Alphabet’s DeepMind, has garnered investment from some big names in A.I., including Meta’s chief scientist Yann LeCun and Google’s senior vice president and head of research Jeff Dean.
Perplexity competes with OpenAI’s ChatGPT, Microsoft’s new OpenAI-powered Bing chat, and Google’s Bard, as well as other chat-search hybrids from companies such as You.com and Duck Duck Go.
Among Perplexity’s new features is the ability for users to specify which kinds of sources they would like the chatbot to search when formulating responses. These new search verticals include news articles, academic journals and papers, Wikipedia, YouTube, and Reddit threads, as well as an option to do a general internet search.
Users will also now be able to save their dialogue threads with the search engine for later reference, and have the ability to edit and refine these threads. Perplexity will also now support longer, multi-line queries in its search bar. It will support requests for help with coding and code debugging. In addition, it will be able to provide responses in different formats, including arranging information in tables.
Among the biggest shifts is that Perplexity is asking users to register and create a sign in to use the new features. This will allow users to better customize their experience using the search engine, but it will also allow the company to gather more data about users, which may be important in the future if Perplexity decides to roll-out paid versions of its service. The company said it will continue to offer a simplified version of the search chatbot that will not require users to login.
“We are introducing sign ups and more advanced features for personalization,” Perplexity co-founder and CEO Aravind Srinivas told Coins2Day in an email. He described the new features and the creation of logins for users as important “first steps” in the startup’s “evolution as a consumer company.”
Just last week, Perplexity announced that a new $26 million Series A venture capital funding round lead by New Enterprise Associates, with participation from Databricks Ventures, the venture capital arm of the A.I. Company Databricks, as well as angel investments from LeCun and Dean, and former GitHub CEO Nat Friedman.
The company said that its search engine had 10 million monthly visits in February—a minuscule fraction of the approximately 90 billion people who use a standard Google search each month. The company launched an Apple iOS app last week. The company said it has had over 100,000 people sign up for the app in the six days since its release.
But Perplexity hopes to stand out from the other new chatbot search interfaces with the accuracy of its answers, Srinivas said. Perplexity has come up with better methods for what’s called “grounding”—making sure what its chatbot is saying is rooted in factual information—and its chatbot uses footnotes to cite the sources of every factual sentence it provides.
So far, the tendency of other chatbots such as ChatGPT, Bing, and Bard to invent information—a phenomenon A.I. Researchers call “hallucination”—or, in other cases, to rely on inaccurate sources, represents a big problem for the ability of these A.I.-powered chatbot interfaces to rival traditional search. Google currently dominates the search market, with a greater than 90% market share.
Srinivas co-founded Perplexity in 2022 along with Denis Yarats, now the company’s chief technology officer, as well as Johnny Ho and Andy Konwinski. Yarats and Ho had spent time working as A.I. Researchers or engineers at question-answering site Quora. Konwinksi was among the founding team at Databricks. Srinivas had interned at DeepMind and Google and later joined OpenAI as a researcher. Yarats had joined Meta’s A.I. Research team after his time at Quora.
