Comment by nullc
2 days ago
I've been describing this as "The LLM is an improv machine--- any situation you put it in, it tries to go with the flow. This is useful when you understand what it's doing, dangerous otherwise. It can be helpful to imagine that every prompt begins with an unstated, "Lets improvise a scene!"."
system prompts, and the code around "agents", all include dramatis personae
how so many people think it's a good idea to take this synthetic improv transcript and extract code, commands to run, emails to send on the job, etc?