“In an experiment to classify suicidal intent on Reddit posts using GPT-4, removing the name of Professor Philip Resnick from the prompt context caused a significant drop in model performance.”