This is a consequence of a type of cognitive bias - bad examples of AI are more easily detectable than good examples of AI. Subsequently, when we recall examples of AI content, bad examples are more easily accessible. This leads to the faulty conclusion that.
> AI so far has given us ability to mass produce shit content of no use to anybody
Good AI goes largely undetected, for the simple reason that it closely matches the distribution of non-AI content.
Controversial aside: This is same bias that results in non-passing trans people being representative of the whole. Passing trans folk simply blend in.