💭 I promise not to use it for evil, but I think I have a system in mind that could make an llm optimally impersonate me, at least for short texts with little context