AI derangement, diagnosed
A field guide to the anti-AI identity: the contradiction, the self-sealing logic, failure theater, the artisanal coders, and the one question that tells you whether you are talking to a skeptic or a belief system. With the receipts — the Codeberg vote, the Torvalds line, the car wash test — and a copyable checklist for telling governance from ideology.
There is a genre of video that performs extremely well right now. It runs about fifteen minutes, it explains that AI is a bubble and the companies are doomed and the technology is fake, and it closes by telling you that the good people are the ones who refuse to touch any of it. One of these crossed my feed last week and did big numbers in its first day. And somewhere around the halfway mark I realized the specific video does not matter, because it is the same video every time.
So instead of arguing with one video, I made this week's episode about the behavior underneath the whole genre. This article is the version you can keep: the pattern, the receipts, and two tools you can copy and use the next time one of these takes lands in front of you.
The definition
By AI derangement I mean the group of people who treat hating AI as an identity. In their world, AI is simultaneously an unbelievably capable and dangerous technology, built by a coordinated elite to reshape society against you, and a useless toy that cannot count the letters in "strawberry," propped up by companies that are about to collapse. Hold both of those claims next to each other and ask which one is true, because it cannot be both. When someone carries the most powerful-and-worthless technology in human history around in their head and never notices the contradiction, you are not looking at an analysis. You are looking at a membership badge.
The self-sealing machine
In a working mind, evidence moves you. Inside AI derangement, every possible observation lands on the same conclusion:
| What happens | What it proves to them |
|---|---|
| AI companies lose money | The bubble is collapsing |
| AI companies make money | They are exploiting everyone |
| Models get better | The corporations are dangerously powerful |
| Models fail at something | The whole thing is fake |
| AI replaces some work | Catastrophic unemployment is here |
| AI does not replace work | It was all hype |
| AI is expensive | It can never work economically |
| AI gets cheap | Now it will destroy even more jobs |
There is no result that could arrive tomorrow that would move the needle, and when literally everything that happens proves you right, you are not analyzing anything anymore. You have built yourself a belief system.
Failure theater
Let me be straight about the technology first: it does do stupid things sometimes. These systems are highly capable at some things and simply not capable at others, and one does not take away from the other. The strawberry counting failures were real. So is the car wash problem. A tool that is not a person will fail in ways a person would not.
The dishonesty is in what the genre tests. The viral clip never touches the things these systems are actually used for every day. It goes straight at the known weak spots, films the stumble, and presents the clip as a verdict on the technology. Turn that method around and it judges humans just as well: you cannot multiply 2,184 by 10,615 in half a millisecond, so you must be useless. A serious evaluation tells you four things:
- what system was tested, exactly
- what it was given to work with
- whether the failure is reproducible
- whether a stronger available tool does better
The clips tell you none of the four. That is how you know they are entertainment for the already convinced, not evidence.
What it is actually about
Getting great at something takes years, and the skill becomes identity as much as income. Then a tool arrives that gets almost anyone to seventy percent of that result with a few prompts and an idea. The ability did not shrink, but the scarcity of it did, and for people whose sense of authority depends on being special, that is the real emergency. You can watch the defense operate in real time, because the objection keeps moving: it cannot do what I do, then the quality is bad, then it is stealing, then it is soulless, then even if nobody can tell the difference the process makes it invalid. An objection that teleports every time you answer it was never the objection.
Open source ran this whole argument in public this summer, and it produced the cleanest receipts available. In July, members of one code-hosting platform voted 358 to 144 to ban projects that mostly consist of AI-generated code. The same month, Linus Torvalds told the Linux kernel community that "Linux is not one of those anti-AI projects," and that anyone who could not accept AI-assisted review was free to fork the project or walk away. Note what Torvalds did not say. He did not say the tools are sacred. He said judge the code: use a tool that finds real bugs, reject garbage whether a human or a model wrote it. Maintainers drowning in lazy machine-generated pull requests that take five minutes to produce and an hour to review have a legitimate complaint, and that complaint is about quality and burden, not about what tools say about a person's soul. That distinction is the entire subject of this article, and there is a checklist for it below.
The confidence diagnosis
The last piece is the one nobody wants to say out loud. People who get into real trouble with these tools, and people who need the technology to be fake, tend to share one trait: there is no strong sense of self doing the driving. If you feed a chatbot your panic, it errs on the side of caution, and so would your mother. If you let a chat window make your decisions, the chat window is not the problem. To be clear, this is not about kids, and it is not about people in genuine mental health crises; that is a real conversation about safeguards. It is about functioning adults who outsource their judgment and then blame the tool, and about experts who would rather rebuild their worldview than update their capability estimates.
Real confidence is not the belief that your skill stays scarce forever. It is trusting yourself to adapt when the rules change. One kind of person says: if these machines are coming, I am going to face it, embrace it, and use it to my advantage. The other kind constructs a reason the evolution is not happening. The sentence underneath all of it is simple, and nobody says it out loud: I do not trust myself to navigate a world where the rules that made me valuable are changing.
The kill test
Here is the tool this whole diagnosis compresses into, and it works in any comment section. Ask one question: what development in AI would make you more optimistic about it? A real skeptic has an answer, because a real skeptic is running a forecast, and forecasts update. If no possible answer exists, you are not talking to a forecast. You are talking to an identity, and you cannot argue someone out of an identity, so spend your evening building something instead.
Copyable: governance or ideology?
When you meet an anti-AI rule, policy, or take, run it through five questions. Quality gates pass. Identity movements fail.
- Does it judge the artifact (is the work good, is the burden fair) or the person (what does using the tool say about you)?
- Does it name specific, measurable harms (review load, error rates, provenance) or moral categories (soulless, fake, cheating)?
- Would the rule still make sense if a human produced the same output the same way?
- Can its holder name evidence that would relax it?
- Does answering one objection settle anything, or does the objection move?
Further reading
- Codeberg's announcement of its member vote and terms change (blog.codeberg.org)
- LWN's coverage of the vote (lwn.net)
- Tom's Hardware on the Torvalds statement (tomshardware.com)
The full argument, with all the heat this article leaves out, is in this week's episode. As for me, I am not waiting around.