When you prompt an LLM, you’re not talking to one brain.
You’re collapsing millions of possible answers into one line of text.
Every reply is a probability hologram and your words decide which version of the model you meet.
In order to be very efficient with LLMs, you gotta understand how they function.