You have just beautifully articulated, in my mind, why AI is (for the foreseeable future) incapable of replacing us:

I have limited understanding of large language models, but I view them as if-this-then-that rulesets which return an average summary of aggregated answers. It's a regurgitation of the baseline. It's the common sense.

Since it is, by its nature, incapable of thinking outside the box (since its job is to provide the box), it is incapable of generating insights unless one of its rules is to break its own rules in a way that it understands not only improves on the common sense but will be acceptable to others (which would do much better than humans, since any sufficiently interesting idea is usually written off as lunacy.)

I could be wildly wrong about that, but that eccentricity that so many of us in the insights business share is the very thing that makes us valuable.