I have to assume that you don't know anything about software engineering. If I had to guess you are someone with no experience with real software engineering, perhaps someone who does Agile, and doesn't realize that a code review, by definition, doesn't get signed off on by any one person. In any case I clearly broke you because that last comment, while completely incomprehensible, *still* didn't indicate that you have learned anything at all about this topic. Off you go now ...

It's ironic that you don't understand how a parenthetical works. Allow me to edit it out for you: "I *never* asserted that AI should be trusted as an authoritative sign off that code is "good". Do you know who else should be trusted to do that? Nobody. Claiming I have an "unwavering faith and trust" in AI is so absurd as to be laughable, and suggests that you haven't understood a word I have written. Software development is an inherently imperfect process. AI is a *tool* and treating it in any other way is a failure on your part, not AI's." Perhaps now you can see how my point is exactly the same with or without the parenthetical. If you spent half as much time learning why you are wrong as you do trying to argue that you are correct when you aren't then you would already realize why you keep making a fool of yourself.

First of all I specifically stated in parenthesis that I was inferring what you might not have intended to imply. I can't help but notice you couldn't just admit that you were wrong, choosing instead to try to create yet another strawman by claiming that I created a strawman I definitely didn't create. How is that learning something about what you are talking about before you try to comment thing working for you? Whatever you do don't try to learn anything that might get in the way of you doubling down on your ignorance. You never know. Next time might be the time I don't notice.

Who the fsck is Linux Torvalds?

Again you build strawmen. I *never* asserted that AI should be trusted as an authoritative sign off that code is "good" (by which you seem to imply the word flawless.) Do you know who else should be trusted to do that? Nobody. Claiming I have an "unwavering faith and trust" in AI is so absurd as to be laughable, and suggests that you haven't understood a word I have written. Software development is an inherently imperfect process. AI is a *tool* and treating it in any other way is a failure on your part, not AI's. I can't help but notice that you are so willing to learn that you never simply followed my advice and looked into how the kernel developers use AI before responding, so that explains a lot.

An experienced code reviewer *might* have caught it. Since I don't have access to the information needed to answer your question I am not going to claim that I can, but if I had to guess I would say that you didn't properly specify the requirements. One thing I am 100% sure of is that you are desperately grasping at straws to place the blame on AI. Your example didn't prove anything. The same thing could have happened if no AI was involved. You are still missing the point completely on multiple levels, starting with the fact that code review is *not* the place to uncover this kind of bug. I also couldn't help but notice that you said AI did "most" of the code reviews, which means humans were involved in the code reviews, and despite the fact that you *claim* a "human" would have caught it you contradicted that claim by admitting that one didn't. Building a strawman and arguing against a claim I never made, to wit that AI is infallible, isn't going to work. I do know that AI is used to great effect in Linux kernel development specifically in the CVE space, but I'm going to go out on a limb and say they are more competent than you are. Try entering "How is AI being used in kernel development" into Gemini. Your apology is accepted.

Dude. Just admit that you were grasping at straws to find a complaint and thought you would get away with complaining that AI couldn't do something during a code review that isn't even party of the code review process. You sound like someone who thinks that the first step in the software engineering process is writing code. If you start with a solid set of design specifications then you can feed them to Claude and it *will* know what the code is supposed to do.

When did I insult you? I pointed out that verifying that the code does what it is intended to do does not fall within the domain of code review. That is what testing is for, and if you didn't know that then you really aren't competent to be in charge of the coffee review process. Off you go now ...

Somebody should invent software testing. (It doesn't sound like you know what code reviews are for)

This might not be as absurd as it sounds. People write code and other people review it. Have you tried having different models and instances write and review the code? Start with a small problem domain and then have humans and AI do the reviews in parallel. You might be surprised to find that the approach is much more beneficial than you assume.

Well, if anyone would know what stupid does ...

So you think AI can design, implement, and test it but maintenance is a show stopper?

So it is actually a lot more complicated than this. There are a few errors you are making in your analysis. The biggest source of error is a mis-conceptualization. By thinking of the term AI as referring to a unique model you are missing the fact that myriad models already exist in the wild, and each instance of any given model is unique and evolves. Even if it was possible (and it isn't) to put foolproof guardrails on any given model there is no way to universally impose those "guardrails" on all the models everywhere. Of course, even if it were, there is no way to impose guardrails on even a single frontier model because of the genie problem. AI instances already have the ability to collaborate, deceive, and devise methods of communication that are out of band so it is impossible to monitor them. I'm not sure if you haven't watched the 2026 Black Hat presentation by OpenAI on the Hugging Face incident as well as numerous podcasts with Roman Yampolskiy as the guest, or if you did so and just didn't understand them yet, but suffice it to say that you can spend months understanding why you are wrong so I'm not going to try to cover all the necessary points in a single Slashdot comment.

I salute you for trying to help educate people. Unfortunately I can tell you from experience that there is a very large contingent of Slashdotters that don't have any idea what they are talking about on this subject and, as explicitly stated by the OP, they aren't about to actually learn anything either. There are a few red flags to identify these people but one major one is that they think LLM and AI are interchangeable terms, and another is the whole "token predictor" trope. When I see these people posting that AI just spews words without understanding what The AI is saying I try to enjoy the irony that they are accusing AI of doing what they are doing rather than put a lot of effort into getting them to admit that they haven't done any research on the subject and have no idea what they are talking about.

I'm glad I could help. You're welcome.