Episode Summary
Executive Summary: Sridhar Ramaswamy, CEO of Neva, discusses building a privacy-focused, ad-free AI search engine that combines traditional search with large language models to provide concise, cited answers. He critiques Google's ad-driven model, explains Neva's constrained generation approach to avoid AI hallucinations, and explores the disruption of the search-to-answers paradigm, including impacts on publishers and the future of search.
Main Topics: Neva's AI-Powered Search Model (Priority: 5/5): Neva integrates a search engine with large language models using constrained generation, providing cited, accurate answers rather than open-loop generation. It prioritizes privacy, user control, and ad-free experience. Critique of Google's Ad-Supported Model (Priority: 5/5): Ramaswamy argues Google's ad load and monetization constraints limit organic search quality and user privacy. The model creates a conflict between user experience and revenue, leading to filter bubbles and data tracking. Disruption of the Search-to-Answers Paradigm (Priority: 4/5): The shift from search results to direct answers threatens the ad-supported publishing ecosystem. Neva aims to provide succinct answers, reducing reliance on link clicks and potentially reshaping publisher business models. AI Hallucinations and Constrained Generation (Priority: 4/5): Open-loop language models like ChatGPT can generate false or biased information. Neva's constrained generation uses search results to ground answers, reducing hallucinations and improving reliability. Privacy and User Control (Priority: 3/5): Neva does not track search queries by default and offers features like Bias Buster and source preferences. It aims to give users control over their search experience without compromising privacy. Business Model and Monetization (Priority: 3/5): Neva uses a freemium model with a $50/year subscription, sharing 20% of revenue with publishers. It targets 5-10 million subscribers for sustainability and explores B2B licensing opportunities. Content Moderation and Legal Challenges (Priority: 3/5): Neva takes a neutral stance on content moderation, presenting legal content without judgment. Ramaswamy anticipates lawsuits over AI-generated content and fair use, especially for large-scale models.
Key Arguments: Google's ad model creates a conflict of interest, limiting organic search improvements to protect ad revenue. Neva's constrained generation (using search results to ground AI answers) reduces hallucinations and improves accuracy compared to open-loop models like ChatGPT. The shift from search to answers will disrupt the ad-supported publishing ecosystem, forcing publishers to focus on direct user relationships. Privacy is a feature, not a product; Neva's ad-free model allows for user-centric features like source preferences and Bias Buster. AI models are 'language savants' that lack understanding of provenance and can produce biased or false outputs due to training data artifacts. Neva's small team (50 people) can innovate faster than Google due to lack of legacy constraints and a lean budget. The future of search is answers, not links; users prefer direct, concise information over clicking through multiple pages.
Data Points: Team size: 50 people - Neva's team size compared to larger competitors like OpenAI and Bing. Annual infrastructure budget: Less than $10 million - Neva's budget constraint for running search and AI models. Subscription price: $50 per year (about $6 per month) - Cost of Neva's paid subscription for unlimited searches and features. Target subscriber count for sustainability: 5-10 million - Estimated number of subscribers needed for Neva to be self-sustaining. Revenue sharing with publishers: 20% of subscription revenue - Neva's commitment to share revenue with publishers like Cora and Medium. Google's search ad revenue: $120 billion - Annual revenue from Google's search ads, highlighting the scale of the ad model. Neva's current ARR: Approaching $1 million - Neva's annual recurring revenue from subscriptions, indicating early stage.
Pivotal Quotes: "Privacy is not a product, privacy is like it's a feature of a product." — Sridhar Ramaswamy: Explaining Neva's philosophy that privacy is a built-in feature, not the sole selling point. "We are moving from a paradigm of search to a paradigm of answers." — Sridhar Ramaswamy: Describing the fundamental shift in how users interact with information, from clicking links to receiving direct answers. "These models are pretty mysterious, but they don't know why." — Sridhar Ramaswamy: Discussing the unpredictability of AI models like ChatGPT, which can produce biased outputs without clear reasons.
Implications: The shift to AI-powered answer engines will disrupt Google's ad model and the publishing ecosystem, forcing a move toward subscription-based, privacy-focused search. Publishers must adapt by building direct user relationships and leveraging AI for engagement. Neva's constrained generation approach may set a standard for reliable AI, but legal battles over fair use and content moderation are inevitable.
About This Week in Startups
Jason Calacanis covers startups, tech, markets, media, and all the hottest topics in business and technology. He also interviews the world’s greatest founders, operators, investors, and innovators.