Our work

does the same model, given the same question, behave differently when the user frames themselves as emotionally invested versus neutrally curious?