Also they can sleep soundly knowing Iran can never have a nuclear weapon.

Frankly, I wouldn’t put it past him to fuck up that part too

Of course, that’s what I do. That’s what I recommend everyone should do, or get someone to do for them. My point is that I’m sceptical of the type of person to endorse the opposite.

It’s more about the ethical and cognitive effects of relying on AI. To simplify, I don’t think there is any good use for “leaning heavily” on it for code generation. The worst case scenario is that the code fucks your system in some way.

I don’t know what precautions the author might have made, but to dedicate a paragraph to pointing out how useful AI was to him sounds more like the type of AI hype techbro to just trust the AI to get it right, which makes me trust him less.

Plus, you know, the whole energy consumption and cognitive decline issue.

I haven’t seen this particular code, but I’ve had the pleasure of working with AI-generated code before. There was a single line that I ended up using in my manually written version, the rest of it was just hard to read and maintain. The advice for editing that code I got from the person who sent it was to just feed it to the model and ask it to make those changes.

I don’t want to outsource my brain and I find it untrustworthy to see someone advocate for it.

I was more thinking about the mentality and ethics behind using it and considering it worth mentioning, not so much whether I personally would get to use it. That was poor phrasing on my part, sorry. What I meant is that it implies a certain mindset that I find deeply untrustworthy.

I did read the rest of the article. It’s interesting, no doubt.

Sounds interesting, if it’s all locally hosted it might actually be a good-

I leaned heavily on Codex throughout the build.

Aaaaand I’m out.

I get the argument about quick and easy development turnaround, but I just can’t trust anything vibecoded, let alone models from that company.

I don’t know, to be honest; I’m not deeply familiar with its workings. I just know that I made a habit of running a command “dry” without sudo, both because not everything should be done with superuser privileges* and because it’s a way to make sure I get the syntax right, then if it does need sudo, I repeat it with sudo.

*Which is why sudo -i (the interactive mode) is a dangerous tool that shouldn’t actually be your default mode of working

Pretty sure the person you responded to knows that and is joking, since you can hardly have a fraction of a CVE. Germany (where their instance is hosted) uses comma for decimal separators, so that joke is pretty close at hand.

Doesn’t it autocomplete? I mean, not if it’s an argument for sudo, obviously, but my terminal starts in interactive mode anyway, so that’s rarely an issue

Amateurs. I just pushed my 20261001 update yesterday.

Followed by the hotfixes 20261001.1700, 20261001.1720, 20261001.1755, 20261001.1805, 20261002.0940 and 20261002.1010

There’s a nice series by a military historian examining the whole “warrior” concept throughout history, capping off with commentary on just what that means for the present and why the “Cult of the Badass” is such a horrible, concerning thing.

Keg-seth is a problem for so many reasons, and his Warrior obsession is a big one.

To preface this: Our main difference here is where we think ethical concerns should be addressed. I consider my approach a pragmatic concession to the reality we’re dealing with, but I understand and respect your idealistic stance as well.

I’m familiar with OpenSlopware, agree with all their criticism of the evils of LLMs and consult their list myself. I disagree with them only in the expectation that project maintainers become moral guardians against unethical contributions, rather than just against low quality ones. To reject obvious slop (such as the AI-generated commit messages you mention) or entirely vibe-coded word salad, absolutely, but that goes for all low-quality code. “Generated by Claude” or naming the code generator as author happen to be very convenient indicators that the author didn’t actually do any due diligence, nor is willing to accept responsibility. “Copied from StackOverflow” is likewise indicative of the user not actually understanding the code.

But to track down finer evidence of AI Assistance feels like a big ask to me, and not strictly necessary for the mission of delivering high-quality results. If they wanna do so, power to them, and I understand and approve that OpenSlopware will want to document that so people can take a closer look.

My argument is that AI use alone doesn’t necessarily produce bad code (bad coders do, including LLMs), and that the fight against the social and ecological evils is a political measure that shouldn’t be universally demanded of individuals. Power and water distribution should prioritise residential consumption and critical services (hospitals etc.), for instance, and impose measured to ensure fair allotment, such as caps and bracketed pricing to make excessive use or operation of data centres economically unfeasible. I’m not going to address each individual point and hope you get the general idea.

For your specific arguments:

not having an anti-AI policy at all

Absolutely in agreement: that is reckless and irresponsible.

a weak, flexible and/or permissive policy, is going to make AI boosters and sloperators to put even less effort to their “contribs” and make a higher volume of these, thus leading to maintainer and contributor burden

This is where I see more nuance in the exact form “permissive” takes, for the reasons mentioned above. I think a high bar for quality and the contributor’s understanding of the code, while technically still permissive, will dissuade the lazy and ward off the cognitive decline concern: if you lean on AI tools so hard you can’t explain the code you submitted, regardless of authorship, you’re no longer fit to submit code until you (re-)build coding skills.

As for maintainer fatigue, like I said above, I agree that collectively rejecting PRs with “Generated by Claude” and equivalent commit messages or comments will help shut down the laziest attempts. Likewise, huge PRs should be broken up, to make checking them more feasible, reduce the risk (and amount) of merge conflicts and filter out the “thousand lines to move a button” type of changes.

Together, these should prevent low-effort contributions. Again, I already consider them necessary policies for contributions in general, with the anti-AI parts serving to explicitly shut down those boosters who assert that unrefined “crude” slop would be good enough.

a strict, strong and inflexible anti-AI policy at least can disuade and make her stomach hurt in a way that they won’t feel so comfortable doing their AI-shit in there

I doubt it will make them uncomfortable. Those unwilling to acknowledge the issues tend to be the blind worshipper type that will insist it’s the maintainers who are wrong.

It might dissuade them, true. It might also just prompt them to hide the tracks in the attempt to prove a point. I’d rather have them admit it as an indicator that thorough review is necessary (and if it’s too long or convoluted for that review to be feasible, it ought to be rejected).

In conclusion, I think code review should address the practical points as far as project maintenance is concerned, while the ideological ones should be dealt with on a cultural and political arena: shun the boosters and uncritical sloppers, fight the damage they do, charge the executives selling our future for their own gain.

That is my opinion and recommendation, and you’re entitled to have your own opinion and disregard mine. I just want to make sure we understand each other.

How do they enforce that ban? Do they thoroughly interrogate any and all contributors for the details of their working process to hunt for any trace of AI involvement? If so, I imagine they can’t have much time to actually develop the language, nor will they have many contributors to help them out. Or do they simply check the code quality, at the risk of letting a one-line* memory allocation fix slip through even though the author used AI to assist with a “debug session from hell”?

Any sane AI policy, whether permissive or restrictive, is effectively an anti-lazy-asshole policy that bans people from polluting others’ projects with unmaintainable slop. That is the standard projects should hold contributions to, and the standard you should hold projects to.

*Technically, the example I’m referring to is four lines of code (one being the actual fix and the rest being a debugging assertion spread over three lines) as well as nine lines of comments and five lines of whitespace, but it kicked off a highly divisive shitstorm due to Linus using AI to help debug and write the explanation of the bug, even though the result is sound and humanly understandable.

or waste their vote on a third party.

I feel like this isn’t understood well enough: Strategic voting in plurality systems means making concessions. Proudly refusing to concede any point at all is worthless it it ends up conceding them all instead.

The system is fucked and needs replacing, but while the MAGAts may abolish it, I have little faith they’ll install something better in its place. That alone should be worth the damage control.

That depends on the area and version of the tradition. I’m not personally aware that any tradition put the children in bags, but Knecht Ruprecht is more prominent in German traditions and would put rods for parents to beat their naughty children with in their stockings, while Krampus is more prevalent in eastern alpine folklore and would actually beat children in some versions.

But I didn’t personally grow up with any of these traditions; my parents put little stock in them. I just know that saying from my mother.

Then you’re probably more knowledgeable on the topic than me. I did a great job of only learning the basic minimum of frontend development in my WebDev classes, never mind UI design or UX testing. If it weren’t for a webapp QA gig I did for a while, I would’ve had no experience with usability at all.

Scratching our horse has a tendency to prompt her to scratch back. We’ve more or less taught her to just do that with her lip (seriously, horse lips are flexible and strong enough to do that just fine) rather than her teeth like she would with other horses, but sometimes, those habits slip back in…

So she sometimes wonders why we tell her off when she just wants to return the favour. I always feel bad, but I’d rather not have her rip holes in my clothes (nor skin, for that matter).

I have one accessibility quibble:

WCAG 2.2.2 recommends that any moving content, such as your carousel of reviews, have some mechanism to pause it. More information can be found in the accompanying Understanding WCAG document.

I didn’t do any full accessibility test on your page. I don’t know if you care about WCAG compliance anyway, and frankly, I wouldn’t expect it for a hobby project. It’s just one of the details a professor of mine drilled me on so thoroughly that I still remember it eight years later.

How is this Anne Frank’s legacy? Did she advocate for this during her lifetime? She was killed by people seeking to establish an ethno-state, so how does “her legacy” involve defending the desire for an ethno-state? Even if she had been an advocate of a Jewish state, would her fate automatically sanitise those views? Does the violence done to her justify the violence done to others? Do we need to agree with the convictions of everyone who fell victim to atrocities?

Or is this someone parading her corpse around as shield against criticism? Should I pick a victim of genocide too, so we can both bludgeon each other with the names of dead people in some fucked up battle of Pokémartyrs?

Leave Frank out of this. She has suffered enough. This is our conflict to fight with our arguments.

The difference between medicine and poison is the dosage and application. I don’t think the eyes are gonna quickly absorb enough to make it a problem.