E7: In Search of Truth with Aravind Srinivas of Perplexity AI

Episode Summary

Executive Summary: Aravind Srinivas, CEO of Perplexity AI, discusses his mission to build the world's most trusted information service by combining search with LLMs, emphasizing citations and truthfulness. He contrasts Perplexity's product philosophy with Google and Bing, focusing on maximizing user value per unit time rather than engagement. He shares insights on technical challenges, feedback mechanisms, and safety. On AGI, he believes it will reduce the need for work and increase equality, but short-term wealth inequality may rise. He trusts OpenAI's leadership and Sam Altman's judgment.

Main Topics: Perplexity's Mission and Product Philosophy (Priority: 5/5): Perplexity aims to be the most knowledge and truth-centric company, providing accurate answers with citations. They prioritize user value per unit time, not engagement. Features like streaming, entity linking, and no sign-up are designed to reduce friction and build trust. Competition with Tech Giants (Priority: 4/5): Perplexity competes with Google, Bing, and OpenAI by focusing on search-specific use cases rather than general chatbots. They differentiate through citations, speed, and a collaborative platform. They are not worried about Bing's GPT-4 access because of their holistic product approach. Technical Approach: Combining Search and LLMs (Priority: 4/5): Perplexity uses existing LLMs (GPT-3.5/4) but plans to train own models to reduce costs. They rely on traditional search indexes for ranking and use LLMs for summarization. Streaming and fast first-token times are critical. They optimize for latency and tail latencies. User Experience: Feedback, Personalization, and Safety (Priority: 3/5): Only 10% of users give feedback; Perplexity uses contractors for labeling. They avoid intrusive sign-ups and plan to introduce personalization carefully. Safety is handled through query classifiers and citations, but they acknowledge the difficulty of preventing all misuse. AGI and Societal Impact (Priority: 5/5): Aravind believes AGI will allow humans to live a 'nice life' with less work, comparing current technology to past luxuries. He sees AI as a leveler of inequality in the long run but warns of short-term wealth inequality. He trusts OpenAI's approach and Sam Altman's leadership. Trust in AI Leadership and Future Hopes/Fears (Priority: 4/5): Aravind expresses confidence in Sam Altman and OpenAI's alignment incentives. He hopes AI will make life easier and reduce stress, but fears short-term job displacement and inequality. He advocates for basic income in the long term.

Key Arguments: Perplexity differentiates by focusing on truth and citations, not entertainment or engagement. They prioritize user value per unit time over time-on-site metrics. They use existing LLMs but plan to build own models for cost control, not for differentiation. AGI will reduce the need for work and increase equality, but short-term wealth inequality may rise. Trust in Sam Altman and OpenAI is justified by their track record and incentive structure. Technology is the biggest leveler of human equity, as seen in past revolutions.

Data Points: Google revenue (2022): $282 billion - Google's total revenue in 2022, with $60 billion profit. Perplexity initial funding and team size: $15 million, 10 employees - Perplexity started with a $15M VC round and about 10 employees. User feedback rate: 10% - Only 10% of users give feedback on Perplexity. First token return time: Under 1 second - Perplexity achieves fast response times, under a second for first token. Cost to embed the whole web (OpenAI estimate): $50 million - Boris Power estimated $50M to embed the entire web with OpenAI embeddings.

Pivotal Quotes: "We want to be the world's most knowledge and truth-centric company." — Aravind Srinivas: Explaining Perplexity's mission inspired by Amazon's customer-centric philosophy. "You right now are already living a higher quality life than the president of the United States like 50 years ago." — Aravind Srinivas: Arguing that technology is the biggest leveler of human equity. "I would trust him to make the right judgment here." — Aravind Srinivas: Expressing trust in Sam Altman's leadership regarding AGI development.

Implications: AI will transform work and society, potentially reducing drudgery but increasing short-term inequality. Perplexity's focus on truth and citations may set new standards for AI trustworthiness. Careful governance and safety measures are needed to navigate the transition.