How did Aravind Srinivas get started in AI?

Aravind's interest in AI began with deep learning research. His path was significantly shaped by an internship at OpenAI, where he received direct advice on focusing on key AI research areas like unsupervised and reinforcement learning.

What was the inspiration behind starting Perplexity?

Aravind was inspired by a blog post on 'how to build the next Google' and the potential of LLMs to improve search. He also envisioned a product where AI advancements directly enhance the product itself, creating a virtuous cycle.

What were the early experiments and demos for Perplexity?

Before becoming Perplexity, they developed a specialized search engine for Twitter data by organizing it into tables and using AI to generate SQL queries. This demonstrated the potential of AI in structured data search.

How does Perplexity approach user experience and product design?

Perplexity follows a 'the user is never wrong' philosophy, emphasizing creating intuitive products that directly address user needs, similar to early Google's approach. They focus on making the product experience seamless rather than making users adopt complex practices.

What is Perplexity's primary metric for success?

The primary metric is the number of queries per day, indicating user engagement and product utility. This metric is reviewed weekly and monthly to track growth and address any declines.

How does Perplexity plan to compete with tech giants like Google?

Perplexity's edge lies in its obsession with the user and strong product taste, aiming to provide an end-to-end experience from problem discovery to fulfillment. They believe Google's ad-driven model is an encumbrance, potentially creating space for a new player with a different monetization strategy.

What is the future vision for Perplexity?

The vision is for Perplexity to become a single platform for users to address problems, receive answers, and fulfill actions, moving beyond just being an intelligent search engine to offering a comprehensive experience.

How To Build The Future: Aravind Srinivas

Y Combinator

Science & Technology5 min read35 min video

Feb 21, 2025|138,977 views|2,751|83

YC Y Combinator

Save to Pod

Key Moments

TL;DR

Perplexity aims to be an intelligent search engine, competing with Google by offering a superior user experience and anticipating future AI-driven information needs.

Key Insights

Aravind Srinivas's journey into AI was shaped by early mentors and research experiences at OpenAI and Google.

Perplexity's core idea is to create a more intelligent search engine, not necessarily to "kill" Google, but to offer a better user experience.

The company's initial focus on verticalized search and enterprise data was challenging, leading to a pivot towards a more general, LLM-driven approach.

Perplexity's success relies on a 'dumb' approach to data processing, betting on the continuous improvement of LLMs for effective search.

The company prioritizes user experience and product taste, drawing inspiration from early Google's user-centric philosophy.

Competing with tech giants requires innovation in monetization and an end-to-end user experience, from problem to fulfillment.

EARLY DAYS AND THE ALLURE OF AI

Aravind Srinivas's fascination with AI began during his undergraduate studies in India, which led him to pursue his PhD in the US at Berkeley. A pivotal internship at OpenAI, where he interacted with figures like Ilya Sutskever, profoundly shaped his understanding of AI research. Sutskever's emphasis on unsupervised learning and generative models steered Srinivas away from the then-dominant trend of reinforcement learning, setting him on a path to explore the potential of generative AI and its applications.

BIRTH OF AN AMBITIOUS IDEA

Srinivas's time at Google provided him with insights from the book "In the Plex," inspiring him to consider founding a company that blends deep AI research with grounded product building. He identified search and self-driving cars as unique domains where product iterations directly fuel AI improvement, creating a virtuous cycle. This realization, coupled with a desire to build something with significant scale and ambition, laid the groundwork for what would become Perplexity, focusing on a model where AI advancements continuously enhance the product.

THE QUEST FOR A BETTER GOOGLE

Inspired by Daniel Gross's blog on building the next Google, Srinivas explored how query reformulation and the application of Large Language Models (LLMs) could vastly improve search. He envisioned LLMs automatically discovering effective search suffixes and classifying queries. Simultaneously, the concept of AI agents controlling mobile environments sparked his interest. This led him to connect with his co-founder, Dennis, with whom he had previously brainstormed agentic AI ideas, laying the foundation for their entrepreneurial journey.

INITIAL EXPERIMENTS AND A STRATEGIC PIVOT

Early experiments involved pitching a vision to disrupt Google from a "glass" interface, avoiding direct competition with traditional search interfaces. However, realizing the difficulty in securing data for enterprise solutions, the team pivoted. They creatively built a searchable database of Twitter data using OpenAI's Codex, organizing it into tables and employing a RAG (Retrieval-Augmented Generation) approach with SQL templates. This demo, built rapidly, garnered positive attention for its novel capabilities and conversational interface.

THE 'DUMB' APPROACH AND LLM ADVANCEMENTS

The team realized that building indexes for every domain was inefficient. They adopted a strategy of keeping data unstructured and relying on LLMs to do the heavy lifting during inference. This 'dumb' approach, which involved taking top search results and cached snippets to feed into LLMs for summary generation, proved effective as LLM capabilities, particularly instruction following, improved significantly around the GPT-3.5 era. This method reduced latency and offered a more scalable, generalizable solution.

TRACTION AND REALIZING THE POTENTIAL

The virality of Perplexity's initial launch, driven by user curiosity and the sharing of personalized search results (like biographies), provided early validation. The introduction of follow-up questions doubled engagement time and exponentially increased daily questions, solidifying the belief that the product had significant potential, far beyond a pivot to enterprise solutions. This traction indicated a strong user appetite for a more intelligent and interactive search experience.

NAVIGATING THE COMPETITIVE LANDSCAPE

Srinivas initially worried more about Microsoft's Bing Chat than Google. However, he observed that Google's homepage is too cluttered for Perplexity's direct answer format, and Microsoft's consumer product track record was less strong. Despite looming announcements from both Google (Bard) and Microsoft, he felt Perplexity had an opening due to its focused user experience and the inherent challenges these giants faced in adapting their established models and infrastructures.

A USER-CENTRIC PRODUCT PHILOSOPHY

Perplexity is built on a foundation of user obsession, mirroring early Google's philosophy that "the user is never wrong." This translates into designing products that adapt to users, rather than forcing users to adapt to the product (e.g., handling typos, autosuggest). The team prioritizes a seamless and intuitive experience, believing that magical consumer products, unlike enterprise software, should require minimal user effort and understanding, aiming to make the interaction as effortless as possible.

METRICS, TEAM CULTURE, AND GROWTH

The primary metric for Perplexity is the number of queries per day, reflecting user engagement and product utility. The team maintains a data-driven culture, reviewing metrics weekly and transparently sharing progress. There's an emphasis on a flat hierarchy, encouraging direct communication and collaboration without fear of reprisal. This culture fosters a collective drive for product excellence, with every team member acting as a user and advocate.

THE CHALLENGE OF EVOLVING THE SEARCH PARADIGM

The long-term vision for Perplexity involves moving beyond just providing answers to facilitating end-to-end experiences, including fulfilling user actions like purchasing products or booking travel. This presents a challenge, as it requires balancing the ad-free, informational experience beloved by early adopters with the monetization needs for mass-market adoption. The goal is to become the primary destination for users starting with a problem and ending with a solution.

BUILDING THE NEXT GENERATION INFORMATION EXPERIENCE

The future of search involves orchestrating various AI models, knowledge graphs, and widgets to provide seamless user experiences. Perplexity aims to build this complex orchestration layer, differentiating itself from competitors focused solely on foundational AI models. The strategy is to leverage existing open-source models and focus on product integration and user experience, believing this approach is more crucial for building the next generation of information access than solely competing on AI benchmarks.

COMPETITIVE EDGE AND LONG-TERM VISION

Perplexity's edge against giants like Google lies in its laser focus on user experience and product taste, rather than being encumbered by an ad-driven business model. While Google has distribution and resources, its search is arguably subservient to its advertising revenue. Perplexity aims to build a user-centric search experience that can evolve and monetize effectively in a future where AI agents and direct conversations become standard, requiring perseverance over decades to achieve.

Mentioned in This Episode

●Software & Apps

●Companies

●Organizations

●Books

●Concepts

●People Referenced

Building vs. Blaming: User-Centric Product Design

Practical takeaways from this episode

Do This

Prioritize the user's needs: Assume the user is trying to achieve their goal and help them, rather than expecting them to adapt to the product.

Focus on core problems: Identify areas where AI can fundamentally improve user experience, like search.

Iterate quickly and get user feedback, especially direct, honest feedback.

Value attention to detail and obsess over product quality.

Fight entropy by maintaining speed and agility as the company grows.

Consider the end-to-end user journey, from problem to fulfillment.

Avoid This

Don't blame users for not being prompt engineers; design intuitive interfaces.

Avoid becoming a slow, cumbersome company by actively combating complacency and ensuring speed.

Don't just focus on AI benchmarks; prioritize practical application and user value.

Don't solely rely on polite feedback; seek out critical opinions to find the worst bugs.

Avoid building solely on hype; ensure a solid product that genuinely serves user needs.

Common Questions

Perplexity is an AI-powered search engine that aims to provide more intelligent and direct answers than traditional search engines like Google. It focuses on summarizing information from various sources to give users a comprehensive overview, rather than just a list of links.

Topics

Future Of Search Data Organization

Mentioned in this video

softwareGoogle Finance

A product developed by Google, showing their history of building specialized vertical services.

personEloot Gil

Perplexity's first seed investor whom Aravind pitched to.

bookin the Plex

A book Aravind read during his Google internship that inspired him to potentially start a research-grounded company.

softwareWeb GPT

A project by John Schulman's team at OpenAI that demonstrated web searching capabilities for LLMs.

softwareGoogle Flights

A travel-focused product from Google that Perplexity sees as a model for future expansion.

personJohn Schulman

A researcher whose team at OpenAI developed Web GPT, an early web-searching LLM.

software3.5 Turbo

A specific, more capable version of GPT models that improved Perplexity's performance.

softwareOpenAI Codex

An early OpenAI model used for the Twitter search demo to generate SQL queries.

softwareGoogle Shopping

A vertical service from Google that Perplexity aims to replicate or integrate.

personAravind Srinivas

Co-founder and CEO of Perplexity, discussing his journey into AI and the founding of his company.

conceptSearch Engine

One of the core areas where AI can be integrated with product development for a virtuous cycle.

personPaul Buchheit

YC Partner credited with inspiring the development of spell check and auto-suggest features at Google.

softwareTruth Bot

An early bot at OpenAI designed to search the web and provide answers with sources.

organizationNEA

Venture capital firm that provided Perplexity with a term sheet.

conceptUnsupervised Learning

A key area of AI research highlighted by Ilya Sutskever as crucial for AGI.

conceptSelf-driving Cars

Another area where AI integration with product rollout creates a flywheel effect.

toolPolygraph

Found this useful? Build your knowledge library

Get AI-powered summaries of any YouTube video, podcast, or article in seconds. Save them to your personal pods and access them anytime.

Try Summify free