The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch
The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch

20VC: The Best AI Companies Have Unique Data Acquisition Strategies | Will Simile Kill Kalshi, Polymarkets and NASDAQ | How to Sign Fortune 500 Companies As Customers in Weeks with Joon Sung Park, Simile

August 1, 2026

AI Summary

5 min read

In 2023, Joon Sung Park and his team at Stanford built a simulated town of 25 AI agents, each waking up, going to work, forming relationships, and even self-organizing a Valentine's Day party. The agents were powered by GPT-3.5, but the real innovation was giving them memory, planning, and reflection—concepts that later became standard in AI agent architectures. That project, born from a simple insight that language models trained on human behavior data could be prodded to act like people, became the foundation for Simile, a company now valued at hundreds of millions of dollars that aims to build a "foundation model of human behavior."

What Simile Actually Does

Simile is not a prediction company in the traditional sense. Park draws a sharp distinction: "No one really cares about prediction... What people actually care about is they want to shape the future." A company like Starbucks doesn't just want to know its sales will drop in two quarters—it wants to know what to do about it. That requires causal mechanisms and counterfactuals, not just correlations. Simile's models are designed to answer "what if" questions: if we change our pricing strategy, how will consumers actually behave? If we launch this product, how will the market shift?

Continue reading the full summary in the app — free to try.

Read Full Summary →

Free • No credit card required

What you'll learn

  • 1 (00:00) **What Is Simile? Core Thesis & Origin Story** - The episode opens with Joon Sung Park’s fundamental thesis: AI companies need a defensible data strategy, and he predicts single simulation sessions could cost $100M in 2-3 years.
  • 2 (10:00) **Simile’s Model: Not Super-Rational, But Human-Biased** - Joon draws a sharp line between frontier models (coding, math, super-rationality) and Simile (human error, bias, and subjectivity).
  • 3 (11:50) **Prediction vs. Shaping the Future: The Causal Mechanism** - Joon argues that pure prediction is overvalued; what enterprises truly want is to change outcomes.
  • 4 (12:53) **Data Acquisition: The Hardest Pillar** - Data is the biggest challenge, and Joon reveals his two-part strategy for defensible data.
  • 5 (14:56) **Beyond Surveys: Simulation as Generalizable Model of People** - Joon clarifies that Simile is not trying to be the next Qualtrics; it’s building a foundation model of human behavior.
  • 6 (16:59) **Ethics: Who to Work With and the North Star of Representation** - Joon addresses the potential for misuse of simulation technology.
  • 7 (18:07) **How Much Data Is Enough? The Flywheel in Practice** - Joon explains the data requirements and the unique feedback loop for simulation.

+ Full timestamped outline available in the app

Show Notes

Joon Sung Park is the Founder and CEO of Simile, the AI simulation company building foundation models of human behaviour; allowing companies to test how real people may think, decide and act before making a decision in the real world. Simile has now raised $300 million in total, including a $200 million Series B announced last week at a $2 billion valuation, led by Greenoaks and Index Ventures.

AGENDA:

00:00 We Will Pay $100M for a Single Query on Some Models

10:00 Why Stock Markets May Not Exist in 5 Years Time

15:00 The Best Companies All Have Unique Data Acquisition Strategies

19:00 The Best AI Companies Have Clear and Fast Reward Functions

24:00 How We Sign Fortune 500 Companies for $10M Contracts in Weeks

32:00 Does Similie Kill Kalshi and Polymarket? Prediction vs Changing the Future

42:00 Inside Similie's $300M Raise; What Every Founder Needs to Know

 

 

The Twenty Minute VC (20VC): Venture Capital | Startup Funding | The Pitch