The early part of your comment said

seems to imply that there is a correct way to use AI.

LLMs are universally hugbots, telling you what they think you want to hear. That's always going to be a bad way to get accurate solutions. I think we agree on that.

There are prediction models that aren't LLMs which can be useful. Your comment was unclear on whether you considered LLMs to be the entirety of "AI" or not.