Copy the formatted transcript to paste into ChatGPT or Claude for analysis
Are you like a Chennai boy? Have you
grown up there all your life? My parents
live in Chennai, so I first go there.
What were these fancy ideas? I'm very
upset to hear that cuz I actually
thought my ideas are cool.
[Music]
Hi Arvind. Hi Nikico. Hi. This is a bit
weird for me cuz I'm doing
it this way. a conversation after a bit.
Okay. Yeah. Yeah. I wish we could be in
the same place.
Where are you now? I'm in San Francisco,
right? Yeah. I was traveling to Europe
last week. Okay. A lot more travel
AI-generated overview
Aravind Srinivas, CEO of Perplexity AI, discusses his journey from Chennai to founding a leading AI search company in this conversation with Nikhil Kamath. Srinivas explains the evolution of artificial intelligence from narrow applications to general intelligence, emphasizing that the breakthrough came from throwing massive compute power at simple neural network architectures rather than complex theoretical approaches. He details how modern large language models work through next-word prediction on internet-scale data, and how reasoning capabilities emerge from this process. Srinivas argues that future AI differentiation will come from agentic behavior—AI systems that can complete tasks autonomously rather than just answer questions. He discusses Meta's stronger competitive position over Google due to network effects versus search dependence, advocates for India building its own AI models and infrastructure, and identifies voice AI for Indian languages and dialects as an untapped opportunity for startups.
The AI breakthrough came from applying massive compute to simple ideas rather than complex theories. At OpenAI in 2018, Ilya Sutskever told Srinivas that AI is just two circles—generative AI and reinforcement learning—and the key is throwing compute at it, which proved correct over complicated academic approaches.
Current AI chatbots including ChatGPT, Claude, Gemini, and Perplexity have minimal genuine differentiation today. The real distinction will emerge in 2025-2026 through agentic behavior—AI systems that can autonomously book reservations, send emails, and complete multi-step tasks rather than just answer questions.
Meta has stronger competitive moats than Google in an AI-dominated future because Instagram and WhatsApp's value comes from network effects and human connection, while Google's ad business directly conflicts with AI agents that can bypass sponsored links and find optimal results.
India should build its own foundational AI models, not just applications, despite arriving at similar answers as Western models. Having domestic model training capabilities is essential for data sovereignty, inspiring local engineering talent, and participating in advancing AI capabilities like reasoning and agentic behavior.