If we are talking autocomplete for IDEs, not generative AI. Also, a lot of the predictive code lines thing are just based off of what's most likely to come next, as ( most likely ) trained off of others work. Whereas I'd argue that is different enough from an LLM since it's basically the IDE version of the autocomplete on your smart phones keyboard app ( if you keep it on ). For LLMs, I would argue that they generate generic code that majority of the time needs cleaned up in order to even function since it will often, especially from back before I ditched it, look at your request saying you are using a specific language version and disregard because it thinks it can just use any code from that language and it'll work. Don't even get me started on how generative AI companies purposefully make their "software" write code in a way that is extremely bloated snd could easily be cut down to a couple lines just so they can say how their AI ( emphasis on artificial ) can write more code faster.

If we're talking filters on art programs like Krita, also historically not completely generative AI. Majority of them ( at least on Krita ) rely on transforming your art by whatever the code says it should do as compared to generating a new image/texture. No prompt nor a need to mass steal art online in order for these filters to work. As for even something like an autofill colourize mask thing, that relies more on some form of gap detection, which isn't considered generative AI, AFAIK. Just an advanced bucket fill tool.

If we're talking things like ESRGAN, sure, and I absolutely have problems with that as well, but they're ever so slightly lesser versions of the problems I have with LLMs and such.

I am well aware that some forms of it have existed for long before LLMs, but they were much more tame and weren't being used to destroy the internet and planet to the degree it's being used now.

Is the programmer reviewing the code ai shits out or blindly trusting it?

Depends. If they continue to use it on a regular basis, I give them a couple months before their ability starts to decline and theb apply yahoo to them. Otherwise if it's more of a one off thing, I still personally do not trust the code but wouldn't call them a yahoo yet.