Fireside Chat, Interview
'Future of Search in the Age of AI' Perplexity CEO, Aravind Srinivas | RAISE Summit 2024 | Paris
- The company aims to establish a self-reinforcing data flywheel from day zero, where AI-driven product improvements drive increased usage and data collection to further enhance capabilities.
- Product focus will converge on search and answer engine functionality rather than free-form creative tasks, prioritizing scenarios requiring rapid web navigation or tailored answers like legal contract review.
- To mitigate hallucinations, the strategy involves restricting outputs to existing web sources with verifiable citations, while maintaining neutrality by surfacing diverse human opinions on controversial topics instead of AI viewpoints.
- The business model prioritizes user alignment and long-term legacy building over immediate revenue, anticipating lower margins than Google's high-margin advertising model due to higher query costs and a nascent advertising infrastructure.
- Future monetization may leverage subscription models similar to niche data services, though average revenue per user is expected to remain lower than current advertising benchmarks until a new advertising form factor is established.
- Operational architecture will rely on routing and orchestration to match specific intents with appropriate tools, such as API retrieval for factual data or large models for complex queries.
- The default user experience will automate model selection to handle commoditization, hiding underlying distinctions between providers like GPT-4, though fine-grained settings will remain accessible for expert users.
- Partnerships with model providers may drive the selection of specific models for business reasons, diverging from the current neutral approach of model aggregation.