Aravind Srinivas

Co-founder and CEO of Perplexity, the AI-powered answer engine. Previously research scientist at OpenAI and DeepMind. PhD from UC Berkeley. Built Perplexity around the insight that mandatory citations eliminate LLM hallucinations — a discovery made accidentally while building an internal employee Q&A chatbot.

Background

Research background in computer vision and RL. At OpenAI, worked on models including early DALL-E work. Left to co-found Perplexity with the core thesis that the advertising model is fundamentally incompatible with truthful search, and that an AI answer engine with mandatory citation architecture could provide a structurally better product. Known for: founder case study approach to company building (Page, Bezos, Jensen, Musk, Zuckerberg as mental models); latency obsession; “magic metric” thinking for retention.

Appearances in this wiki

Key positions

• Citation as hallucination remedy: mandatory sourcing forces retrieval-grounded generation, eliminating wholesale hallucination
• Google’s AdWords creates structural incentive conflict between advertiser interests and truthful ranking — Perplexity’s subscription model avoids this
• RAG as “open book exam”: decoupling memorisation from reasoning enables accurate answers without massive pre-training
• AI curiosity (generating own research questions) + verified reasoning loops (checking own reasoning without a human) are the two unsolved capabilities needed for autonomous research AI
• Magic metric for Perplexity: “queries that delighted you” — fast, accurate, readable, reliable