Large language models are becoming increasingly capable of imitating human-like responses, creating opportunities to test social science theories on a larger scale and with much greater speed.