Simulating Humans at Scale: Simile's Joon Sung Park

Summary

Joon Sung Park, founder and CEO of Simile, discusses his work on generative agents and large-scale human-behavior simulation. He explains why current frontier models are strong at objective tasks but insufficient for simulating diverse, irrational people, and describes Simile's commercial validation, including a CVS partnership. The conversation covers use cases in concept testing, product research, earnings-call simulation, and long-term ambitions for social-science simulation.

• Joon Sung Park describes Stanford's Smallville and the generative-agents research that led to Simile.
• Simile aims to build models that encode human diversity rather than only rational problem-solving.
• The company validated its simulations by predicting behavior of 1,000 Americans at 85% of human replication accuracy.
• CVS is cited as a customer using Simile for concept testing and market-insight work.
• Simile combines real-world data collection, including Gallup panels and interviews, with AI agents.
• The discussion covers applications such as product testing, multi-agent simulations, and earnings-call preparation.
• Longer-term ambitions include simulating bank runs, climate cooperation, democratic instability, and macroeconomics.
• No public stock or asset recommendation is made; most investable references are private companies or unnamed use cases.