They are still a statistical machines, without any ability for actual comprehension and interpretation. They try to give an output that statistically matches best the question/task - regardless of the correctness of it.

It gets very obvious (and annoying) when you ask some not trivial technical questions.