This is part of the #LaunchWithAI series.Interested in sharing your story? Contact us here!
It’s amazing how generative AI tools like ChatGPT are rapidly being integrated into our lives. Every day we see meaningful progress in how Large Language Models (LLMs) can help users research, create content, and more. Now the question is, what is the next step in the evolution of AI chatbots?
Perplexed I am working on an answer. Founded in 2022, this San Francisco-based startup uses LLM to build a conversational search engine that delivers more accurate and intuitive results. The company’s technology is based on OpenAI’s GPT-3 language model, but as co-founder and CEO Aravind Srinivas explains, Perplexity leverages the potential of LLM to make search more natural and conversational. We are on a mission to be relevant and deliver more relevant results. The idea is to evolve from a “search engine” to an “answer engine”.
I reached out to Aravind to find out how Perplexity hopes to make it easier for users to answer questions, and how Microsoft gave his company an early competitive edge. , and also discussed what they had learned from the collaboration.
A fundamental human desire for answers
“We want to be the most knowledge-driven company in the world,” says Aravind. “We want to expand the productivity and knowledge of everyone on the planet. At the end of the day, people only want answers, not sifting through tons of links.” We built Perplexity as an answer-first platform.”
Microsoft CEO Satya Nadella also attended Removed the term “response engine” Recently, Aravind emphasized how Perplexity differs from other search engines and chatbots. According to him, Perplexity is a reimagining of question-answer interaction that “picks a needle out of a haystack” to provide users with concise responses.
“We are addressing basic human needs for information and answers,” says Aravind. “Our aim is to help people learn faster. With Perplexity it feels like you are talking to Wikipedia. We have the ability to understand your questions and deliver quality intelligence as a service.”
Aravind says his team found Microsoft’s platform to be the top choice when it came to building an intelligence-as-a-service model.
“We need to deliver super-smart AI models specifically for computing,” says Aravind. “Today, Microsoft is the only company democratizing access to very large artificial general intelligence (AGI) models so that any startup in the world can start building useful products on top of it. is.”
Perplexity’s peculiar AI application hints at the future
The excitement is palpable when Aravind talks about Perplexity’s technology, especially when he emphasizes how unique it is compared to others on the market. But as his company’s features become more successful, he expects other products to adopt similar features.
According to Aravind, Perplexity takes advantage of technological advances wherever possible, preferring a future-oriented and pragmatic approach. One example is the programming seismic shift that emerged when GPT 3.5 entered the market, allowing models to be programmed in plain English. Traditionally, programming an AI model involves collecting data and fine-tuning with gradient descent to update model parameters. However, with GPT 3.5 it is now easier to program using English instructions.
“English becoming a programming language is the biggest shift in computing history,” says Aravind. “So when a company develops a product, it is beneficial to be realistic. If programming in English can achieve something instead of buying GPU instances and training models, It should be done in English.”
Built on a “first in class” AI infrastructure
The AI innovation Perplexity is pioneering is built on Microsoft’s Azure infrastructure, which Aravind calls “first in its class.” He believes they have become completely dependent on this service and it is the reason why the answer engine works so reliably.
“There is no way we could build a product like ours without Azure and OpenAI,” he continues. “Microsoft is fundamentally responsible for both. There have been instances where other similar products have gone offline and they have stood by simply because we trusted them. Azure OpenAI service. As perplexity and user bases grow, we can only imagine more situations in the future where security, reliability, and scale are all fundamentally important. Partnering with Microsoft is the best way to build our products. ”
Aravind was also quick to highlight the exchange of knowledge between the two companies on the responsible use of AI. Despite its positive value, it is hard to see Microsoft not only recognizing its safety obligations to society, but also learning and iterating on where AI adoption could go wrong. It’s reassuring, he says.
“We are working with Azure to make sure there are safety filters in addition to the answers,” says Aravind. “We work with them and share what we learn. It takes a certain level of foresight to keep AI operational at high speeds and still keep the costs of using AI responsible. Yes, and Microsoft is doing it.”
Through Microsoft services and products like the Azure OpenAI Service, everyone now has access to a generic model that seems to do just about anything, says Aravind. He calls it a fundamental change in daily life and stresses the urgent need to rethink our daily work.
“People will feel like they are much more productive,” he says. “Instead of doing the monotony that makes work boring, we can 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 has come for startups to create great user-centric services without the heavy lifting of building LLMs.
“We are very lucky to be working at a time when these models continue to improve,” says Aravind. “It is in everyone’s best interest to start building a product on top of his LLM that is already the best available and the best model at the moment is his Azure OpenAI Service.”
He also encourages startups to consider how Microsoft can help. For Perplexity, membership in the Microsoft for Startups Founders Hub provided significant financial support during the early development of the startup.
“Founders Hub gave us a lot of Azure credits so we could stay thrifty while building on Azure,” says Aravind. “So saving money early on and building your brand on Azure is in the best interest of any startup. recommended.”
For more tips on using AI in your startup and accessing Azure AI services, sign up now. Microsoft for Startups Founders Hub.