As large language models are optimized for rationality and objective problem-solving, their ability to simulate the irrationality and subjective values inherent in human behavior has plateaued. This necessitates a new modeling paradigm focused on capturing human diversity, not just super-intelligence.
One of Simile's surprising yet common use cases is simulating corporate earnings calls. This multi-agent simulation allows executive teams to test their messaging and anticipate audience and investor reactions, providing a rehearsal space for high-stakes financial communications before they happen.
Simulations can be categorized in two ways. 'Convergent' simulations reliably reach a stable outcome despite small errors (e.g., network hub formation). 'Divergent' ones can have many results (e.g., elections). The value of the latter is mapping the range of potential futures, not a single prediction.
Simile's founder views academia as a vehicle for breadth, where researchers explore many parallel theses. He started a company because it is a 'machine for depth research,' enabling a focused team to pool resources and relentlessly pursue a single, ambitious vision needed to bring a complex product to market.
The Smallville project, a simulation of 25 AI agents, demonstrated that generative agents could produce unprompted, complex social behaviors. One agent independently decided to plan a Valentine's Day party, invited others, and saw them attend, showcasing emergent social dynamics.
LLMs trained on online text often reflect what people say, not what they do. Simile bridges this 'say-do gap' by collecting real behavioral data and personal life stories through partners like Gallup. This grounds their agent simulations in reality, making them more predictive of actual behavior.
Beyond simple concept testing, AI simulations allow businesses to model downstream consequences. A car company can simulate how launching a new EV might change market perception of its entire gas-powered product line, revealing second-order effects that are impossible to test in the real world.
