It also can't seem to create icons from another set and match the stroke weight, curvature, and other stylistic choices of another simple icon set. Seems it struggles with tight constraints. 10 years ago you would think it would be able to handle the visually "simple" things over the more complex ones it can do today. Sorry, my disappointment is showing isn't it?

Of course the fundamental failure case is solving novel problems.

If it's in a book AI can do it.

Asking if it's in the book, maybe not.

Not in the book, definitely not.