Key Takeaways
- • Simile raised $100M to build population-scale human behavior simulations, positioning AI as a tool for modeling societal dynamics rather than solving optimization problems, with backing from Sequoia Capital.
- • Achieved 85% accuracy in predicting behavior of 1,000 simulated Americans, matching the consistency rate at which people reproduce their own survey answers—a benchmark suggesting predictive fidelity comparable to human self-reporting.
- • Frames technology as "GPU of intelligence" versus frontier models' "CPU" function—designed to encode human diversity in values, preferences, and tastes rather than maximize rational problem-solving, requiring fundamentally different model architecture.
- • CVS and other customers deploy the platform for concept testing and earnings call simulations, indicating commercial adoption in consumer behavior forecasting and investor relations use cases.
- • Envisions a "CERN of human society" capable of modeling systemic risks including bank runs, climate cooperation failures, and early signals of democratic collapse—positioning human behavior simulation as critical infrastructure for institutional decision-making.
Summary
Joon Sung Park, founder and CEO of Simile, is advancing a thesis that superintelligence development has overlooked a critical capability: accurately simulating how actual humans behave. Speaking at Sequoia Capital's Training Data forum, Park distinguished between two classes of AI systems—frontier models optimized for rational problem-solving ("CPU of intelligence") and Simile's architecture designed to encode the messy diversity of human preferences and values ("GPU of intelligence").
The company's core offering rests on demonstrated technical achievement. Simile simulated a population of 1,000 Americans and predicted their behavior with 85% accuracy—a figure Park frames as particularly significant because it matches the consistency rate at which humans reproduce their own answers on repeated surveys. This metric suggests the model captures behavioral authenticity rather than merely fitting historical data. The approach draws from Park's earlier academic work, including the "Smallville" generative agents study at Stanford, which explored how language models could generate believable human agent behavior in interactive simulations.
Early commercial traction validates market demand for this capability. CVS has adopted Simile for concept testing, and other customers use the platform to simulate earnings calls and investor scenarios—applications suggesting institutional investors and corporations recognize value in human behavior modeling beyond traditional consumer research. This positions Simile in the intersection of applied AI, social science, and institutional risk management.
Park's longer-term ambition extends beyond commercial applications. He proposes building a "CERN of human society"—a reference to the physics research institution—that could model complex societal phenomena including bank runs, climate cooperation dynamics, and the early warning indicators of democratic instability. Such a capability would represent a fundamental shift in how institutions approach systemic risk assessment, moving from historical econometric models toward agent-based simulations grounded in empirically validated human behavior patterns. The $100M funding round positions Simile to pursue this infrastructure-scale ambition while serving immediate commercial customers navigating consumer and investor behavior prediction.
The company's core offering rests on demonstrated technical achievement. Simile simulated a population of 1,000 Americans and predicted their behavior with 85% accuracy—a figure Park frames as particularly significant because it matches the consistency rate at which humans reproduce their own answers on repeated surveys. This metric suggests the model captures behavioral authenticity rather than merely fitting historical data. The approach draws from Park's earlier academic work, including the "Smallville" generative agents study at Stanford, which explored how language models could generate believable human agent behavior in interactive simulations.
Early commercial traction validates market demand for this capability. CVS has adopted Simile for concept testing, and other customers use the platform to simulate earnings calls and investor scenarios—applications suggesting institutional investors and corporations recognize value in human behavior modeling beyond traditional consumer research. This positions Simile in the intersection of applied AI, social science, and institutional risk management.
Park's longer-term ambition extends beyond commercial applications. He proposes building a "CERN of human society"—a reference to the physics research institution—that could model complex societal phenomena including bank runs, climate cooperation dynamics, and the early warning indicators of democratic instability. Such a capability would represent a fundamental shift in how institutions approach systemic risk assessment, moving from historical econometric models toward agent-based simulations grounded in empirically validated human behavior patterns. The $100M funding round positions Simile to pursue this infrastructure-scale ambition while serving immediate commercial customers navigating consumer and investor behavior prediction.
Go deeper than the headline
Research every firm in this story, track the ones that matter and get the daily briefing — free for 7 days.