The issue is not that you can't get a model trained enough to output working code, it's the accuracy, efficiency, and reliability of said code that would matter. The output would still need to be checked and debugged by a human to ensure it even does what is expected of it. Most models at current can kick out templates that work most of the time, but it's absolutely terrible code. So if an engineer just relies on this crap all the time, suddenly you're trading good, performant code, for "eh, it works. One less engineer we have to pay." Also, AI is not capable of creating novel code, so no new ideas will come of it.

They'll be right back to hiring engineers soon enough because their existing ones are terribly out of practice due to relying on AI. This is a cyclical trend that happens every decade, this trend just happens to give boners to the people making business and hiring decisions.

[–] DudeImMacGyver@sh.itjust.works 5 points 2 months ago

That sounds like a spectacularly stupid idea