What's this about?
Joon Sung Park — the researcher behind the "Smallville" generative-agents work — discusses Simile AI's $2B Series B, backed by GreenOaks and Index with Fei-Fei Li and Andrej Karpathy among the investors, running tens of millions of simulations for Fortune 100 clients like CVS, with digital twins reproducing real people at reported 85% accuracy. The host notes he started the interview half-joking and became serious partway through.
What exactly is a "digital twin" of a person?
A simulated agent built to answer as a specific real individual would, calibrated against that person's actual responses. At 85% accuracy on held-out questions, you can run a survey against a simulated population instead of a real one — which is why the customers are Fortune 100 companies rather than research labs.
What's the community saying?
The framing worth carrying is the shift from prediction to intervention: not just forecasting what people will do, but simulating how to change outcomes. The host's own caveat is worth keeping — today's frontier models still miss how humans actually behave, and human biases are part of what has to be reproduced.