social simulation
Ongoing research on agent-based social simulation with LLMs. More to come.
A research project on large-scale social simulation with language-model agents — studying how populations of LLM-driven agents interact, influence each other, and produce emergent group-level behavior.
The first question I care about is methodological rather than behavioral: not whether a single agent resembles a real person, but what a simulator has actually learned about a population — whether it recovers the distribution, the relationships among variables, which of those relationships matter, and how they shift over time. Only once that is pinned down does the substantive question become answerable: how norms, legitimacy, and institutions make cooperation durable across scales.
The project website and code are not public yet. More to come soon — stay tuned.
An early sketch of the wider programme this project sits in — three pillars, a tentative paper sequence, and a reading map — is kept at computational social theory. It is a statement of intent rather than of results, and it will keep changing.