This basically matches my experience in trying to get it to do the right thing. BEING VERY EXPLICIT AND ANGRY WORKS TO REINFORCE A POINT. Specifically telling it to not do a thing it will otherwise do is often necessary.
The only part that surprises me is `Output should be in markdown format`. Usually being that vague results in weird variation in output; I'd have expected a formatted example in the prompt for GPT to copy.
It understands most things without being given examples in my experience. Being explicit is helpful. Being angry is likely inconsequential, but I can't say for sure since I never felt the need TBH. What I can say is that the bot has become spookily like a person, "someone" that I conceive as helpful, courteous and friendly, albeit sometimes wildly wrong in spite of the assertive tone in the response.
I'll probably get used to it over time, as I get a deeper sense of how it works, and how it differs from real persons. ATM the distinction is blurred.
The only part that surprises me is `Output should be in markdown format`. Usually being that vague results in weird variation in output; I'd have expected a formatted example in the prompt for GPT to copy.