It's important to understand that all ChatGPT knows is what's in its training set, and what was in the prompt. I see it all the time in programming - if you are

It's important to understand that all ChatGPT knows is what's in its training set, and what was in the prompt. I see it all the time in programming - if you are doing a boring, run-off-the-mill websites or plugging other peoples' libraries together, then AIs like copilot have a very high success rate. But try something new or unusual (even a bit unusual, like QNX), and the lack of training data means the amount of hallucinations becomes incredibly high, and the resulting code is often outright wrong. A lot of times that's fine - after all, there are programmers who spend their careers never writing an original algorithm, and they could definitely use the help. And in a lot of cases an original architecture is actually a bad idea - stick to the boring technology, create your architecture proposal based on re-blending previous projects with a tiny bit of changes. Nothing wrong with that. But school tries to be better than that (it does not always succeed, but it at least tries). That's why the students are taught how to multiply numbers by hand, and only then they are allowed calculators. And they must be taught how find their primary sources, before being allowed to use Wikipedia. And they must be taught how to write their thoughts in their words, before being allowed to ChatGPT. Sure, some of them will go on and never multiply numbers by hand, nor ever read a primary source nor ever create a highly original proposal. That's fine and even expected. Still, the school aims high even if not everyone can get there.