ChatGPT and the Velcro Test

This argument is correct, LLM are not AGI. However, that doesn't remove any utility from LLMs and ChatGPT. Only you should keep humans in the loop. It's a "chat" architecture, ie. you must have a dialog, between a human and the LLM. (or possibly some other AI components and the LLM).

While ChatGPT alone couldn't re-invent the Velcro or something similar, you could still use to brainstorm and come with some innovative solution to your problem. See for example, how such a conversation might go:

https://chat.openai.com/share/e7799549-b8bd-4e5e-960b-1f0f8c...

Even if ChatGPT came up with Velcro I wonder if it could figure out how useless Velcro is to 'button' pants pockets after they've been washed a few times and the Velcro gets clogged up full of lint—and that fact had been kept from it.

Could ChatGPT then follow up and make a sensible suggestion like "take your pants to your local clothes alteration shop and have them remove the damn Velcro and replace it with sensible secure fasteners—namely buttons—so you won't lose your mobile phone"?