> It had not ocurred to me that anyone would /knowingly/ hire a programmer who
They don't. Hiring happens before the first commit. This incredulity about programmers not all being consistently excellent is what makes it sound like your practical experience is very limited.
Point of fact, you're here complaining that a product (ChatGPT 3.5) known to be substantively shitty is substantively shitty, and when people say "you know v4 is very significantly better, right?" you say "but that one isn't free". That gives a strong vibe that you know you're doing something bad and just don't care. Is that the vibe of consistent excellence? I would say no.
> submitted code that fails even to compile
You seem stuck on this business of not compiling when that failure mode is in fact extremely less consequential than literally any other failure mode. Wake me up when humans stop writing code that compiles and runs and tests just fine while doing the wrong thing literally all the time.
Asking how anyone can blindly trust ChatGPT code is a fundamentally broken question. You can't blindly trust human code either. Trust has to be earned by past demonstrations of performance within the specific domain.
A reviewer earns trust by consistently catching other peoples' mistakes. A programmer earns trust by consistently not making obvious mistakes in the first place. When I see a reviewer who consistently fails to catch other peoples' mistakes, I do not trust that reviewer.
Giving a human more inherent trust than you give ChatGPT without experiencing sufficient past performance to have earned it would be your own failure. Humans are very stupid and fuck up constantly.
Good point. I'll correct to:
It had not ocurred to me that anyone would /knowingly/ hire a programmer who submitted code that fails even to compile, as per ChatGPT.
Anyone who would knowingly could well be delighted to use ChatGPT instead.