Joon Sung Park's team chose simulation over personal agents because a useful assistant requires a deep, accurate model of its user's preferences and behaviors first. Understanding the person is a prerequisite for effective automation.
Prompting works when a model already understands a domain's underlying "physics" and just needs to react. Fine-tuning is necessary when the model must learn a new domain's fundamental rules, such as the nuances of human social behavior not present in its original training data.
Most decision-makers don't want a passive prediction of a negative future. They want to know the causal levers—the actions they can take now to avoid that outcome and create a better one. Simulation provides this by modeling causal mechanisms.
Simile's validation method involves collecting extensive data from real people, creating their "digital twins," and then testing if the twins can accurately predict how the real individuals behave in unseen experiments. This grounding in real-world data builds trust in the simulation's outputs.
Unlike frontier LLMs optimized for perfect reasoning, Simile's goal is to create models that are "as dumb as I am." To be accurate, a behavioral simulation must replicate the same mistakes, biases, and non-optimal choices that real humans make.
Joon Sung Park notes they are observing the beginnings of a scaling law for simulation. As they ingest more compute and high-quality human data, they see predictable improvements in the model's ability to accurately predict and simulate human behavior.
Early agent-based models, like Thomas Schelling's on segregation, used simple rules for agents ("red and blue dots"). Generative agents provide a leap forward by enabling high-fidelity models of people, allowing for richer, more nuanced simulations of complex societal dynamics.
While simulation can disrupt the $100B market research industry, its true total addressable market (TAM) is far larger. The ultimate goal is to inform every decision made by humans, for humans, making the potential value proposition nearly boundless.
Coming from a professional painting background, Joon Sung Park likens simulation to art. A simulation, like a great painting, is never a perfect copy of reality. Instead, its power lies in its ability to highlight the most essential essence of its subject to reveal a deeper truth.