Hey, have you ever noticed how a language model struggles with when to put a comma? I’m curious about the underlying rules it’s learning and how that compares to our human standards.