Discover more from Generational
Future Unicorn #229: Perplexity AI
Talk to your search engine
The Quild Future Unicorn is a weekly product-focused note highlighting early-stage startups with statistically significant signals of becoming unicorns.
This series is powered by Specter, a data intelligence provider for the world's leading investors like Accel and Bessemer. I have been working with data-driven tools for venture for a long time, and Specter's is the best one.
Perplexity.ai is developing a search engine using large language models. Today, it offers two services: Perplexity Ask, which allows users to ask questions and receive answers, powered by OpenAI GPT and Microsoft Bing. (2) BirdSQL, a structured search engine for Twitter that uses OpenAI Codex to translate natural language into SQL.
With news of Microsoft rolling out a ChatGPT + Bing search experience by March 2023, Perplexity would have to find a new angle or wedge. But the caliber of the founding team is still one to bet on.
Andy co-founded Databricks
Top company alumni
Denis worked as an engineer at Microsoft (2 yrs), Quora (3 yrs), Facebook (6 yrs)
Aravind worked as researcher at OpenAI (1 yr) and Google (1 yr)
Top university alumni
Denis graduated from NYU (PhD)
Aravind graduated from UC Berkeley (PhD)
Andy graduated from UC Berkeley (PhD)
The new Foundation Model Primer for Investors and Builders is out. Since its first publication last August 2022, the report has expanded: >2x in length, 3x in the number of companies include in the market map, and an outlook for next year. Check it out here: Foundation Model Report
Pain point and persona
Google is everyone's information gateway. If you have a question, whether medical, financial, or random, you Google it (or Bing it). Google does a great job surfacing the top 3 relevant links at the top of their search results. The challenge is that there is still a lot of information and clicks to go through. You would read the previews of the links, click into them, read through the website, take notes, and then synthesize all of the information.
It doesn't sound as bad as I made it to be. I grew up in the 90s where going to the library was the norm as a kid. Research back then meant going to the library, decoding the shelf indexes, curating a stack of books, looking through the page indexes, and hopefully finding the information you're looking for.
That aside, wouldn't it be nice if someone (or something) just answered the question for you concisely? OpenAI's ChatGPT opened up the possibility of a new way of searching. You can ask it any question and it'll answer concisely based on its stock knowledge of the world up until 2021.
But wouldn't it even be better if someone combined both ChatGPT and Google/Bing so the answers are up to date and attributable with website links? Its like an analyst or a librarian in a digital box. That is what Perplexity is building.
Perplexity's home page is just like any search engine's. A search bar with a section on trending searches.
If you ask ChatGPT what the latest news is, it'll say that it can't browse the internet.
But if you ask Perplexity, it'll give an answer with links.
The concise answer wasn't particularly impressive for that one. But Perplexity also has a neat feature to ask for a Detailed answer. I was able to verify that Perplexity's answers are the latest news since it cites its sources.
And I love their answer to what makes a unicorn. I didn't know it also meant a third-party in a relationship. Ha.
The team at Perplexity is able to do this by combining both OpenAI and Bing APIs. Each search query is sent to the Bing API. The top Bing result snippets are included in the prompt to OpenAI's GPT, which then produces a summary that includes citations. Its a brilliant combination of two powerful APIs into a single user interface that changes how we look for information on the web.