zlacker

>>goranm+(OP)
I don’t see ton of value of playing around with prompts until you get the desired output.

I feel most of AI “engineering” goes to this. I think we will go through the phase of trying one question being amazed by what ChatGPT can immediately reply, then try to refine prompts for days to never really get that 5% better that is missing and be disappointed.

>>hartat+Lk1
The prompt engineering tricks here give more than a 5% benefit. I've had many problems where ChatGPT flat-out doesn't work without prompt engineering.