“Byrne Hobart argues that OpenAI's memory feature is a "great thing for lock-in" because the LLM that a user interacts with the most will know them best, providing superior, personalized answers and creating a self-reinforcing cycle of usage.”