My guess is when enough LLM tools have been created for various use cases, until the insurmountable flaws of their nature are finally broadly realized (of course not implying there's no usefulness to them).
Agreed. And if this fad lasts long enough for future generations to remember, they'll make fun of us for calling it “AI”. It's the equivalent of believing in the existence of witches and devils. It may be explainable in light of the standards of the time, but it still makes us look, er, unintelligent.