Learning to program a computer makes it clear that a computer is incompetent. How do we find out if something is in a list? Does this one match? No. Does this one match? No. Does this one match? No. And on and on. One of the biggest hurdles I ever had to get over starting out was understanding how much of what made a computer feel “smart” was just a lot of really dumb things done really fast (and hopefully in the right order and with the right logic).

But that means if you don’t get that logic right, nothing else is right, and that’s where these AI tools lose their magic. When they get it right they’re impressive and when they get it wrong, it lays bare that your computer is very very stupid, just very fast

"Competent" and "smart" are two different things. A computer that correctly executes a program every time is competent. It's reliably doing what it is intended to do.

These genAI tools don't do that, though, so they aren't competent in the way that a computer is.

From that perspective, these AI tools are perfectly competent. They are programmed to take user input, run it through various LLM models and use that output to act on behalf of the user. And they are reliably doing that thing exactly as they are instructed.

The problem isn’t “competence” then, it’s a failure to create a program that does what the user wants it to do reliably. In this way, these AI programs are no more “incompetent” than any other poorly designed program. Any system that applies randomness to its outputs is inherently going to get things wrong because randomness implies a lack of accuracy.

AI is absolutely getting shoved hard and fast in places it’s just not ready for yet, due to the overhyping of its boosters. But personally I think these sort of articles and complaints that treat them as somehow unique from all the other poorly implemented and rushed software that has plagued the industry since the first paycheck was written is accepting the booster’s hype. It’s poorly implemented software running on the same dumb computers we’ve been using forever, of course it’s going to be getting things wrong.