Generally, I'm an optimistic about current approaches towards AI. I'm using ChatGPT daily for my work. I'm seeing the BS generated from it all the time but it provides many great ideas from which I create something useful (both writing and programming).
The sinus example scares me. It's obviously wrong. But what about all the subtle errors that will be generated?
Humanity's knowledge is bootstrapped from BS. But most of it is long forgotten. On the internet, the BS ends up in the training data of the next iteration. I guess curation of quality text corpora will be an important thing in the next years.