A guy in an AI safety forum pointed out my prompt engineering was lazy

I used to just throw random instructions at GPT-4 and hope for the best, but someone showed me how a single line about output format could fix my messy results... it was a total facepalm moment. Now I spend 5 extra minutes structuring context and examples before hitting enter. Has anyone else had a small tweak totally change their output quality?

3 comments

3 Comments

kellygrant18d ago

Small changes like that can save hours of trimming later.

felixlane17d ago

Wait @corah75 you were just raw-dogging prompts with no structure? I thought everyone at least had a "you are an expert in..." opener by now. Man, the stuff we put together before we knew what we were doing is straight up embarrassing to look back on. Can you even imagine trying to get anything useful out of a 50-line plea compared to a clean 100-word prompt?

corah7518d ago

Oh man, my early prompts were basically me pleading with the AI like "please just work" and hoping for a miracle. A little structure goes such a long way once you stop being lazy about it.