AI Summary
5 min readJoon Sung Park, founder and CEO of Simile, was inspired by science fiction visions of societies that have both AGI and large-scale simulations to guide decision-making. He believes the time has come to build the simulation pillar. His conviction emerged from years of research at Stanford, where he observed that large language models could encode human behavior from training data—and that with the right prompting, they could generate surprisingly realistic micro-behaviors. Simile is now an applied AI lab that simulates human behavior and societies, starting with commercial applications but aiming much higher.
From Smallville to a Company
The breakthrough came with a 2023 Stanford experiment called Smallville. Park and his team created a game town of 25 generative agents, each with a persona, memory, planning, and reflection capabilities. The agents woke up, went to work, formed relationships, and even organized a Valentine's Day party—one cafe owner named Isabella independently decided to host the event, invited customers, gathered materials, and the party actually materialized. Some agents forgot their invitations; others brought dates. The simulation felt alive.
Continue reading the full summary in the app — free to try.
Read Full Summary →Free • No credit card required
Never miss an episode of Training Data
Get every new episode summarized in your inbox — free, ~5 minutes to read.
No spam. Unsubscribe anytime.
What you'll learn
- 1 (00:03) **Science-Fiction Inspiration & The Simulation Thesis** - Joon opens by stating his conviction that simulations, alongside AGI, are a core pillar of advanced societies, and that the time is now right to build them.
- 2 (01:10) **The Smallville Experiment: Genesis & Surprising Results** - Joon describes the 2023 Stanford project that created a town of 25 generative agents with memory, planning, and reflection.
- 3 (03:36) **From Social Computing to Generative Agents** - Joon traces the intellectual lineage from his work on foundation model risks to the specific problem of simulating social platforms.
- 4 (08:02) **The Model Evolution: From Janky to Good Enough** - Joon pinpoints the moment when foundational models became viable for faithful human simulation.
- 5 (10:10) **From Research to Company: The Conviction Moment** - Joon explains the pivot from academic exploration to building a company.
- 6 (12:43) **Canonical Customer Engagement: The CBS Example** - A concrete walkthrough of how a Fortune 500 buyer uses Simile.
- 7 (15:27) **The Self-Driving Car Analogy & The Say-Do Gap** - Joon explains why real-world data collection is essential, despite the power of LLMs.
+ Full timestamped outline available in the app
Guests on this episode
Show Notes
The race to build superintelligence is producing models that keep getting better at objective problems, but not at behaving like actual people. Joon Sung Park, founder and CEO of Simile and creator of Stanford's "Smallville" generative agents study, argues that simulating human society requires a fundamentally different kind of model. He frames today's frontier models as the "CPU of intelligence"—rational, superhuman at problems with right answers—and Simile as creating the "GPU of intelligence," built to encode the diversity of people's values, preferences, and tastes. It simulated 1,000 Americans and predicted their behavior 85% as accurately as people reproduce their own answers. CVS uses it for concept testing; some customers simulate their own earnings calls. Joon's larger bet: a "CERN of human society" that could one day model bank runs, climate cooperation, or the early signals of a collapsing democracy.
Hosted by Sonya Huang, Sequoia Capital
More from this podcast
Training Data →