This still goes back to the likes of Altman etc. who are driving the bullshit hype. The things you noted above as AI success used to just be called Machine Learning and there was no real dispute (even from "AI" haters) that they were useful tools. But they were specialised tools - just like almost every other successful tool.
The LLM pushers want you to believe that their product (it's real AI!) will solve every problem, and it's just one more GPU generation away from Commander Data or R2-D2.
But there's one thing that LLMs are really good at - passing the Turing test. Train them to always tell you your idea is the bestest idea ever in the history of ideas, you original genius, and it's inevitable that CEOs insulated from the real work will fall for the hype.