This is part of our #LaunchWithAI series. Interested in sharing your story? Contact us here!
The rapid integration of generative AI tools like ChatGPT into our lives has been breathtaking. Every day, there are meaningful advancements in how large language models (LLMs) are assisting users with research, content creation, and more. Now the question is, what’s the next step in the evolution of AI chatbots?
Perplexity is working on the answer. Founded in 2022, the San Francisco-based startup built a conversational search engine using LLMs to deliver more accurate and intuitive results. Their technology is based on the GPT-3 language model from OpenAI, but as co-founder and CEO Aravind Srinivas explains, Perplexity is on a mission to use the potential of LLMs to make search more natural and conversational and deliver more relevant results. The idea is to evolve from “search engines” to “answer engines.”
I connected with Aravind to discuss how he hopes Perplexity can help users easily get questions answered, as well as how Microsoft gave his company an early competitive edge and what they’ve learned from the collaboration.
The fundamental human need for answers
“We want to be the world’s most knowledge-centric company,” Aravind says. “We want to amplify the productivity and knowledge of every person on the planet. Because at the end of the day all people want is answers, not to sift through a bunch of links. We built Perplexity as an answer-first platform.”
With even Microsoft CEO Satya Nadella dropping the term “answer engine” recently, Aravind stresses the difference between Perplexity and other search engines and chatbots. Perplexity, he says, reimagines the interaction of asking and answering questions by “picking the needle from the haystack” to give users a succinct response.
“We are catering to a fundamental human need for information and answers,” Aravind says. “Our intent is to help people learn quickly. With Perplexity, it’s almost like you’re talking to Wikipedia. We’re providing high-quality intelligence as a service, with our ability to parse web pages on the fly to make sense of what a user asks.”
When it came to building their intelligence-as-a-service model, Aravind says his team learned that Microsoft’s platform was their top option.
“We need to be serving compute-bound, super-smart AI models,” Aravind says. “Microsoft is the only company right now that has democratized access to really large artificial general intelligence (AGI) models and allowing any startup in the world to start building useful products on top of them.”
Perplexity’s singular application of AI hints at the future
Aravind’s excitement is palpable as he discusses Perplexity’s technology, especially when he highlights how unique it is compared to other products in the market. However, he also anticipates that other products will adopt similar functionality as his company’s features prove successful
Perplexity has also taken advantage of technological advancements wherever they can, Aravind says, preferring to be both future-oriented and pragmatic in their approach. One example was the seismic shift in programming that emerged when GPT 3.5 entered the market, with its ability to program models in plain English. Traditionally, programming AI models involved collecting data and fine-tuning with gradient descent, which updates the parameters of the model. However, with GPT 3.5, you can now simply program it using English language instructions.
“English becoming a programming language is the biggest change that’s happened in computing history,” Aravind says. “And because of that, it pays to be pragmatic when your company is building a product. If something can be accomplished by programming in English instead of buying a GPU instance and training a model, you should do it with English.”
Building on ‘first-of-its-class’ AI infrastructure
The AI innovations being pioneered by Perplexity are built on Microsoft’s Azure infrastructure, which Aravind calls “first of its class.” He says that they’ve come to completely rely on the service, and credits it as the reason their answers engine runs so reliably.
“There’s just no way to build a product like ours without Azure or OpenAI,” he continues. “Microsoft is fundamentally responsible for both. There are even instances when other, similar products would go offline and we stayed up simply because we relied on Azure OpenAI Service. I can only imagine more situations in the future where security, reliability, and scale are all fundamentally critical as Perplexity and our user base grows. Partnering with Microsoft is the best way to build our product.”
Aravind is also quick to highlight the exchange of knowledge between the two companies regarding responsible use of AI. Despite its positive value, he says it’s encouraging to see Microsoft not just realize their safety obligations to society but also learn where AI deployment can go wrong and iteratively make improvements.
“We work with Azure on making sure there are safety filters on top of our answers,” Aravind says. “We collaborate with them and share our learnings. To stay committed to the cost of responsible AI use while still operating fast requires a certain level of visionary thinking, and Microsoft is doing that.”
Through Microsoft’s services and products like Azure OpenAI Service, Aravind says, everyone now has access to general-purpose models that can seemingly do almost anything. He calls it a fundamental change in daily life, and stresses that we’ll quickly need to rethink everyday work.
“People will feel like they’re going to become a lot more productive,” he says. “Instead of doing all the monotonous things that make work boring, they can instead focus on productivity. And at the end of the day, productivity is what makes us happy.”
Advice for other startups
According to Aravind, the time is ripe for startups to create stellar user-focused services without doing the heavy lifting of building an LLM.
“We are very fortunate to be working at a time when these models are just continuing to get better,” Aravind says. “It’s in anyone’s best interest to start building a product on top of the best LLM already available, and the best model right now is Azure OpenAI Service.”
He also recommends that startups consider how Microsoft can help. In Perplexity’s case, membership in the Microsoft for Startups Founders Hub provided critical financial support early in the startup’s development.
“The Founders Hub offered us significant Azure credits which allowed us to stay frugal while simultaneously building on Azure,” Aravind says. “So it’s in any startup’s best interest to save funding early on and build your brand on Azure. Microsoft has been super supportive in this way, and I encourage people to take advantage of it.”
For more tips on leveraging AI for your startup and for access to Azure AI services, sign up today for Microsoft for Startups Founders Hub.
