Have you considered that the evidence just isn't convincing yet?
If you view popular llms as text prediction machines, they are in fact much better than spell check or auto correct from a few years ago. But if you actually ask it to solve the problem with nuance it will not use nuance. That's the part that would impress me.
As a recent example if you ask chat GPT how to use ffmpeg to slice out a video. and you tell it that you only want 3 seconds of video. somebody who deeply understands how ffmpeg works, (or even someone who deeply read the documentation) would point out that you have to be aware that it can only cut to keyframes. so if the keyframe is not aligned to the time you ask for you will not get the video that you expect.
another example ask it to play 20 questions with you, it will cheat at the end. even if you give it very specific instructions it still is unable to follow them to the fair conclusion of the game. (or it will make a mistake and understanding about some of the semantics of the question, but I don't fault it for a difference in context)
I'd caution you that just because you're impressed for the subjects that you understand deeply does not mean that chat gpt is good at all subjects. and therefore I assert that it is disrespectful to be so dismissive of people who have different opinions than yours. I believe the default should be to assume good faith rather than dismissiveness "they just don't understand"