Episode Summary: Perplexity AI with CEO Arvind Seros

Introduction:

• Podcast: Unsupservised Learning Redpoints AI Podcast
• Hosts: Jacob Efron and Pat Chase
• Guest: Arvind Seros, CEO of Perplexity AI

Perplexity AI Overview:

• Next-generation search product, popular on Twitter.
• Raised at $500 million valuation recently.
• Goal: To be on top of the market in accuracy, reliability, latency, delightful UX, and constant improvement (including personalization).

Challenges in Modern Search:

• Understanding and reformulating queries.
• Selecting relevant pages and content for queries.
• Rendering the answer with supporting citations, images, or videos.
• Creating shareable content with permalink features.

AI-Assisted Interaction & Improvements:

• Co-pilot feature to assist users with follow-up questions as human curiosity is endless but articulation can be limited.
• Intention to make AI interactions as seamless as talking to a friend.

Using & Building Models:

• Started with out-of-the-box OpenAI models.
• Shifted to fine-tuning faster models, and now using and releasing open-source models.
• Vertical integration is key: all teams should communicate to develop quality products.
• Importance of keeping up with technological waves like open-source models.

Thoughts on AI Industry:

• Companies often described as “wrappers” around OpenAI’s GPT models but starting with such models is smart.
• Having a moat only matters if there's a product worth protecting.
• Specialization within the industry will remain important.
• Building on existing models initially is crucial, and over time improving upon them.

Search in the Future:

• In 10 years, search will be more about answers and task-oriented agents, similar to how we talk to friends.
• Focus on deep knowledge work, surpassing platforms like Quora and Wikipedia by maximally increasing 'knowledge velocity'.

Perplexity's User Experience:

• Focus on making the product intuitive for new users while maintaining functionality for power users.
• Challenges of balancing feature implementation versus core product vision and growth.

Company Evolution:

• Initially showed interest in search over databases, which had limited market appeal.
• Pivoted to search on top of Twitter data due to high user interaction and feedback.
• Decided to focus on consumer-facing search product due to virality and user engagement.

On Industry Trendsetters:

• Arvind's favorite companies in the generative AI space include LL Labs and PE.
• Interested in using AI for preserving memories and creating content like movies.

Regulation in AI:

• Regulation is premature as widespread economic benefits have yet to be seen.
• If AI is dangerous, more people working on it is better for safety.
• Caution around few entities controlling decision-making in AI development.

Final Thoughts:

• Perplexity AI has gained product momentum by building in public.
• More developments in generative AI will enable for new possibilities and applications.
• Perplexity AI remains focused on improving search and creating delightful user experiences.

To learn more about Arvind Seros or Perplexity AI, the podcast hosts recommend using Perplexity itself to search for information.

Listen to the full podcast episode for an in-depth conversation