Ashokkumar et al. 2026

Large language models can predict the results of social science experiments. Ashokkumar, A., Hewitt, L., Ghezae, I., and Willer, R. (2026). Nature.

Relationship to this project — Strong positive counterexample

Forecasts 469 effects from 70 preregistered survey experiments with strong correlations (alongside systematic effect-size overestimation and weaker megastudy performance). It is the strongest evidence against blanket pessimism — but it is a forecasting task over studies, not subject-level simulation of a response surface. Strong effect forecasting is compatible with the fixed-panel composition failure studied here; the two sharpen the estimand distinction.