Do any LLM products run the LLM code as it's being streamed? Seems possible for some scripts and then you can do things like incrementally show prints or a mutated matplotlib graph or whatever