What's this about?
This practical introduction surveys pretrained models that estimate causal quantities such as treatment effects on entirely new datasets through in-context learning, without training a bespoke estimator for each study.
What exactly is “causal inference”?
Prediction asks what is likely to happen; causal inference asks what changes because of an intervention. Seeing that umbrella sales and rain correlate does not mean buying umbrellas causes rain.
Why is this worth noting?
Foundation models could make causal analysis faster and more accessible, but validity still depends on assumptions about confounding and study design that pattern recognition cannot wish away.