It's disingenuous to compare LLM outputs to script generated ones. They have more differences than similarities.

A script is reproducible, and fixable. They won't waste your time hallucinating issues. Note that this is different from "being wrong". A script can be wrong, and you can spend time debugging it, but that time is not wasted. You can fix the script, remove wrong rules, whatever.

Also, nothing exists in a vacuum. Script-generated review comments were never a widespread societal issue, LLMs unfeterred content generation is.

It's a bit insane how it seems that "deterministic and reproducible", what I would consider to be a cornerstone of good software development, has apparently become an afterthought in the age of LLMs. All that for the sake of perceived convenience.

Not even convenience at times, I think that for some the amount of technical-sounding text LLMs can output on demand is comforting. It feels serious. It feels like it's doing something important.

Meanwhile the rest of us have to deal with these low-signal/high-noise walls of text in commit messages and PR comments. Or ask another LLM to summarize them, I suppose. "Doing and undoing, it's all work", as we say in French.

Is the hallucination or “being wrong” the only major problem? Because if that’s the case I can’t imagine that problem will still be around in ~1-5 years.

I think it’s actually very similar to a script. It takes an input and produces an output. Certainly it has more variation than a traditional script, but it's generally going to return a very similar answer when the same input is given repeatedly, like when you ask an employee at a restaurant where the bathroom is. The verbiage may not be the same every time but the answer is always the same.