This isn't really my point. My point is that you can't ask the documentation to expand upon what little is already in it. Stack Overflow works a little better for this purpose, because you can ask follow-up questions and get more than one answer. But even there you have to wait days for answers, and sometimes you don't get any at all. That's really inefficient, and perhaps you forgot why you asked in the first place, or just gave up and moved on.

Not so with an AI chat system. You get the answer immediately, and you can ask as many follow-up questions you like, to your heart's content. This means that, sure, even if the code you get from ChatGPT has approximately as high a chance to work as "random code from the internet," you can't ask that "random code from the internet" to expand on why it's good or bad, why it was made in a particular way, or if there are better ways, or if you can error check it one more time, or complain that, "Hey, random code, your dumb code didn't work!" In that way ChatGPT is already leaps and bounds better than "random code from the internet."

Also I assume here that you aren't just throwing darts at the wall, but instead testing out code with a clear goal in mind. Then it can be wise to ask ChatGPT about the underlying system, which it will gladly answer, and also immediately.