Transforming Search with AI - Perplexity's Approach to Answer Engines

Explore the innovative world of AI-powered search engines in this 44-minute podcast episode featuring Denis Yarats, CTO of Perplexity. Gain insights into how Perplexity combines advanced search engine technology with large language models to deliver precise, rapid answers. Delve into technical challenges, the critical importance of speed, and the future of AI in search. Learn about the decision-making process between in-house and outsourced models, methods for evaluating result quality, and strategies for handling controversial topics. Discover the significance of diverse annotation teams and hear Yarats' personal anecdote about pitching to Yann LeCun. Understand the early indicators of success for Perplexity's generative AI application and the complexities involved in scaling while maintaining a focus on speed and accuracy.

- Introduction
- Denis describes Perplexity as an answer engine.
- Discussion on using third-party APIs and in-house infrastructure.
- Choosing between In-house vs. outsourced models
- Evaluating the quality of results and using LLMs.
- Building a classical search engine and custom parsers.
- Latency and quality trade-offs in providing answers.
- Handling controversial domains and providing unbiased answers.
- Importance of hiring diverse annotators and their influence.
- Denis's story about pitching Yann LeCun.
- Early signs of success for Perplexity's gen AI application.
- The hardest parts of scaling up the application and maintaining focus on speed and accuracy
- Closing thoughts