TBF, I think one can’t ignore LLMs. I think they work very well already as Stack Overflow replacement. It is especially good when only thing you have going for is an error message or a stack trace.
And it sometimes is extremely useful when you go back and forth and discussion converges towards a testable statement (one you have might have not seen in the beginning of the conversation). When it works it is magical, but it often goes awry.