Limited tests in specific settings are always surprising and do speak to it's use, but are not a testament to it being intelligent.

I replied because you answered "you're doing it wrong" to a question of it's failures. It seems you dismiss the concerns of the smaller errors or failures without realizing the point being made. If it's "smart" enough to solve take international math medals and beat grand masters at go, but can't truly understand problems and anticipate needs or issues, to me it's not a genuine intelligence and in it's current form never will be.

It's not that they are not perfect, is that they have no concept of reality, and it's evident in their failures. Beyond this point I am not interested in trying to convince you.