Steve Jang’s Post

Super fun to dive into a deeper discussion onstage last week at Upfront Summit with the ever-thoughtful Aravind Srinivas, co-founder and CEO of Perplexity, a portfolio company at Kindred Ventures! Among AI-native companies, there are still only a handful of products that have found great adoption today. Perplexity's answer engine for both in-the-moment information and deep research is one of the few to scale globally clocking at over 500M queries per month, and growing in adoption with both consumers and businesses. During our talk, Aravind announced the launch of Enterprise Pro version of Deep Research, which solves the more complex business, technical, and scientific analysis needs for knowledge workers. We also covered topics including open vs closed source models, the advantages of building products leveraging multiple AI models, and his personal journey as a first-time founder and CEO. Excerpts below: Regarding open source LLMs: “In today's political climate, open source models are a double-edged sword. On the one hand, they allow anyone to download, fine-tune, and deploy the weights on any data center, regardless of where it came from. On the other hand, there are concerns that open source is dangerous and can fall into the wrong hands, especially if exploited by competitive countries.” – Steve “If something is really dangerous, you probably want more eyes on it, not fewer.” - Aravind "A lot of the technology that powers our phones, laptops, and the internet is built on open-source software. If it works for the foundations of the digital world, it should work for AI, too." - Aravind On building products with unique customer value and trust in mind: "Only two features really help build trust with the user: source citations and a completely transparent reasoning trace. We are the only product that actually does both." - Aravind “For enterprises handling sensitive proprietary data, trust is everything. Having an abstraction between the model and the application layer ensures data security and control–far beyond what’s written in a boilerplate terms-of-service. This is where companies like Perplexity stand out." – Steve "Everything is a wrapper at some level of abstraction. The real question is: Can you deliver real value to the end user at scale?" — Aravind Regarding AI compute and inference: "The cost of intelligence is going down 8x per year. That means serving 50x more users at the same cost. That’s a massive shift." — Aravind "Model labs are spending billions on data centers. But do you even need that much? Open-source models challenge that assumption." — Aravind What's your own personal feature request for Perplexity? “My one feature request for Perplexity is for it to incorporate all your context–the ability to answer questions around your flights, upcoming travel, meetings. It needs to have more context to become a more useful assistant. We’re working on that." - Aravind Thanks again, Mark Suster and Upfront Ventures!