> Good doesn't mean perfect. it doesn't mean flawless.
"Good" means "I can trust it to give me code that is at least as good as what a moderately skilled human would produce". They still aren't there, even after years of development. They still regularly give you code that doesn't follow the correct logic, or which isn't even syntactically valid. They are not good, or even remotely good.
That's just your expectation. if it can do as much as the least competent human, that's already a huge deal. You're expecting it to think for you instead of assist you.
You know what it is capable of, use it accordingly. it saves lots of time in troubleshooting, and generating starter code. in some cases, it can generate full featured complete production apps that people are using without major issues on its own.
Even with your example, you have to fix syntax and errors here and there, instead of writing it from scratch. Which approach takes more time, that depends on the model, the code and you. like the author, your measuring stuck is humans for some reason.
You know it's not really "AI" right, that's just a marketing term. there is no intelligence involved. it's auto completion. your argument is like saying IDE auto completion isn't always great so it should never be used.