I have to refine basic queries multiple times and still then need to verify the answers.
So it's like talking to a person or using a search engine?
You seriously expect or want it to be infallible magic? And then, even if it appeared to be, you would stop verifying its work?? Sounds like an unrealistic and problematic way to view it no matter how good it gets.
I use an llm-- GitHub copilot--every day and I find it pretty helpful.
I would agree that ai/LLM right now is sort of floundering in most products, but your stance that it's absolutely useless trash just is not accurate