Large language models are increasingly used as simulated respondents in computational social science. Yet accurately reproducing the varied response patterns of real populations remains a known challenge, according to a new arXiv paper.
The paper, “Composable Parametric Demographic Injection for Values Survey Simulation,” proposes an approach called composable parametric demographic injection. The method is aimed at making simulated responses better match population-specific distributions.
The abstract notes that existing methods have struggled with this task. The new contribution is framed around improving accuracy for heterogeneous response distributions, though the abstract does not provide full implementation details.