“Divesh Makan observes that users of large language models exhibit very high retention and low churn rates.”