I wonder to what degree generative AI came before e.g. as in the article an AI that does your taxes because it's hard to be wrong, creatively. If I ask an AI for a painting of a dog commanding a fleet of starships, I'd be hard pressed to define, specifically, an incorrect result. There's wiggle room. Some results are more wrong or right than others, but I don't know that I'd ever hand back a grade of 0. And even if I did, so what? The stakes are low. Try again.
On the other hand, doing your taxes wrong is a thing. You can enumerate the possible mistakes. They're codified in law[0]. And these mistake have big stakes.
Generative AI is "safer" than an AI to do a specific task that has consequences.
[0] Yes, yes, I know, there's wiggle room here too and that's why we have courts