E35: Stability AI’s Emad Mostaque and Slow Ventures’ Sam Lessin Discuss Investing in AI

Episode Summary

Executive Summary: Sam Lessin and Imad (Stability AI) debate whether generative AI is truly disruptive or mostly a sustaining technology that will enrich incumbents. They agree AI will transform media, entertainment, coding, and therapy-like use cases, but disagree on where value accrues: Sam says distribution, data, and IP will dominate; Imad sees big opportunities in open models, regulated industries, and global-south leapfrogging.

Main Topics: AI as sustaining vs. disruptive innovation (Priority: 5/5): Sam argues AI mostly extends existing products and strengthens incumbents like Meta, Google, Amazon, Adobe, and Apple rather than creating many zero-to-one startups. Imad sees meaningful new markets, especially where gaps in service exist. Distribution, data, and IP as the real moats (Priority: 5/5): Both repeatedly return to the idea that the winners will be companies with distribution, proprietary data, and valuable IP, while pure model companies and generic app-layer startups are likely to be commoditized. Entertainment and synthetic media (Priority: 5/5): They broadly agree AI will dramatically reduce the cost of content creation and make entertainment more personalized, weird, and engaging, with a major impact on social feeds, porn, influencers, and media production. Education, healthcare, and regulated markets (Priority: 4/5): Imad emphasizes high-value opportunities in regulated sectors and underserved regions, especially AI tutors, healthcare support, compliance, and national models for governments and enterprises. Coding, search, and workflow automation (Priority: 4/5): They see strong near-term value in coding assistance, Stack Overflow-style problem solving, document analysis, and tooling that improves productivity, though Sam thinks this mostly benefits incumbents and existing workflows. Open source, private data, and global-south strategy (Priority: 4/5): Imad frames Stability AI as infrastructure for open, auditable, and locally controlled models serving non-aligned countries and private-data deployments, while Sam worries open web/data will increasingly close off. Social relationships, loneliness, and AI companions (Priority: 4/5): The discussion explores AI characters, AI girlfriends, therapy bots, and synthetic communities as responses to loneliness, with both acknowledging this will reshape online identity and social validation.

Key Arguments: AI is powerful, but for most consumers it is still best understood as next-token prediction and a sustaining technology rather than a full-scale disruption. The biggest winners are likely to be incumbents with distribution, data, and IP, not startups attempting to build from zero. AI will make entertainment radically cheaper and more personalized, but that does not guarantee new standalone businesses will capture the value. Generic model/app companies will struggle because models commoditize quickly and customers can switch to the lowest-cost inference provider. Open models matter strategically for governments, enterprises, and countries that do not want dependence on U.S. or Chinese platforms. Education and healthcare have large unmet needs, making them plausible areas where AI creates real new value, especially in the global south. AI will likely reduce the value of public web information by making scraping and summarization easy, pushing more content behind private walls. Coding is one of the clearest near-term wins because code is testable, self-contained, and highly compatible with current LLM capabilities. AI companions and synthetic social interactions may intensify loneliness, niche communities, and validation-seeking behavior, but they are not necessarily strong investment moats. Inference will become cheaper and more specialized over time, with the market shifting toward customized hardware and edge deployment.

Data Points: ChatGPT training scale: ~100 gigabytes from ~10 trillion words - Used to explain that models are vastly compressed relative to their training data, so even partial correctness is technically remarkable. Stable Diffusion training input size: 100,000 gigs of images - Imad cites this to show the scale of data compressed into a small model artifact. Stable Diffusion model file size: 2 gigabytes - Contrasted with the 100,000 gigs of images used to train it. Stable Diffusion app-store ranking: 4 of the top 10 apps in December - Referenced as evidence of how quickly image generation became consumerized. Character AI usage: ~2 hours/day average - Cited as an example of very high engagement in AI companion products. Microsoft Copilot code generation: 50% of all code on GitHub AI-generated - Mentioned as an estimate of AI-assisted coding adoption. Productivity improvement in coding: 40% efficiency gain - Referenced in the discussion of AI coding tools and Copilot-like workflows. AI tutoring outcome in education: 76% of kids literacy/university in 13 months - Imad cites an adaptive-learning deployment in Africa as evidence for AI-assisted education potential. Valuation of Stability AI: $1 billion valuation after raising $100 million - Used in the episode framing to describe the company’s market position and scrutiny. Therapy supply gap: Not enough therapists in the world - Used to motivate AI-based mental health and support tools. Young male virginity rate (US): 8% in 2008 to 27% in 2018 - Cited in the context of loneliness, dating, and AI companions. South Korea fertility rate: 0.8 - Used as an example of low birth rates in a highly digital/entertaining society. SVB outflow: $18 billion - Mentioned as an example of the power of narrative/story in finance and confidence.

Pivotal Quotes: "AI may well disrupt society at large, but it doesn't seem likely to disrupt many existing SaaS markets between now and then." — Nathan Levenson (host commentary): Sets the overall skeptical framing about startup disruption. "The biggest winners are the people with distribution data, right? That's the answer." — Sam Lessin: Core thesis on where value accrues in AI markets. "What if you had an AI tutor for every child? What does that look like? What if you had 100 AI tutors for every child?" — Imad Mostak: Illustrates the education-focused opportunity Imad believes AI can unlock.

Implications: AI will likely enrich incumbents, compress public information, and intensify competition in media and software. The biggest new value may emerge in regulated, underserved, or globally distributed markets rather than generic consumer apps.