This recent study by the University of Minnesota Law School was set up as a randomised controlled trial, where sixty law students were asked to complete four different legal tasks, either with or without GPT-4.