newsfilter.io
Interview, Fireside Chat

The $2B AI Startup Simulating Human Behavior with 85% Accuracy

  • Simuli (referred to as "Simuli" in the transcript) has raised a total of $300 million, reaching a reported valuation of $2 billion.
  • The company has secured investments from prominent backers including Green Oaks, Index Ventures, Fei-Fei Li, and Andrej Karpathy.
  • Simuli claims to predict human behavior with 85% or greater accuracy using a foundational model trained on data from over 40 million real people.
  • The company's technology is based on the "Smallville" experiment (2022–2023), which simulated a town of 25 agents with memory, planning, and reflection capabilities to demonstrate emergent social behaviors.
  • Simuli has moved beyond simple interaction paradigms to create "digital twins" of individuals and groups to model complex societal interactions and environmental contexts.
  • The core technology models human behavior by identifying the gap between stated attitudes (which often follow a normal distribution) and actual behaviors (which often follow an exponential curve driven by "activation energy" and optionality).
  • Data collection involves voice-to-voice interviews with representative US panels (sampled for gender, race, and ethnicity) to train agents that mirror the psychological profiles of real humans.
  • Simuli partners with polling firms like Gallup to access high-quality panels, employing attention checks and specific sampling intervals to mitigate bias and fatigue.
  • The company's SaaS platform allows users to query simulations with no limit on questions, generating responses alongside a "confidence score" (0–100 probability) to indicate statistical robustness.
  • A confidence score above 85–90% is currently treated as robust evidence for high-stakes decision-making by Simuli's clients.
  • Key verticals include retailers, financial institutions, entertainment companies, and CPG firms; notable clients include CVS and Banco Itaú.
  • Simuli warns against the "short-term optimization" pitfall, where entities lose interest after achieving a single win (e.g., a successful paper, customer, or fundraising round), advocating for long-term commitment.
  • The company's co-founders include Michael Bernstein, Percy Leung, and Laney Yellen, with a team of roughly 100 people comprising researchers, social scientists, engineers, and go-to-market specialists.
  • Simuli operates on a "forward-deployed" strategy to build stickier relationships with customers, similar to the trajectory of the coding agent industry.
  • The founder, June, notes that while they do not avoid any demographic for modeling purposes, they strictly exclude underage populations to ensure informed consent.
  • Simuli simulates societal issues like climate change coordination and Universal Basic Income (UBI) scenarios to understand causal mechanisms before policy implementation.
  • The founder anticipates the simulation industry reaching the maturity level of the current coding agent sector within 1.5 to 2 years.
  • Simuli explicitly rejects a $15 billion acquisition offer if it would hinder their ability to pursue their long-term mission of reshaping societal operations.
  • The company's founder believes the probability of living in a simulation is high, arguing that any mature civilization will eventually seek to recreate itself to understand its own nature.