I've gotten pretty good at that (with other tasks)- there definitely are tricks in the way you word your commands, in particular with a nuanced question. ChatGTP is not always good with those, so you have to continue to reframe you question until Chat clicks with your angle. I find that conversing with it like another human is what gets the most accurate results (argue with it when it is giving you faulty answers and it will get to the right answer).Oh man I have been some conversation with it asking why it does something. Its pretty frustrating when you give it a specific "You must do this" only to have it do something else 2 prompts from now. I'm hopping the addition of gpt memory will help.
You always have to clarify the weighting of listed factors because the AI has a tendency to use the 'unbiased' part of the program to fallaciously discuss all factors as equal even when that is far from the actual case.
Anyway, just a few of my own observations. That would be crazy if AI could make better picks than any group of humans. It's seeing the future, which is kind of like Skynet. Someone call Sarah.