Codeberg disallowing LLM slop on their own platform: "They're restricting my freedom!"
Corporations pirating books and code to automate plagiarism, DDoS-ing the entire internet, and eroding trust in software ecosystems: "It's just a tool!"
Remote
Andrew Radev
@AndrewRadev@hachyderm.io
mastodon 4.7.3Software developer, Bioinformatician-in-training, Vim pluginist, part-time word inventor. Feed may contain traces of shitposting.
I, too, am an AI hater: https://anthonymoser.github.io/writing/ai/haterdom/2025/08/26/i-am-an-ai-hater.html
315 Followers
97 Following
42 Posts
Joined December 10, 2022
Blog:
GitHub:
Codeberg:
Replying to
This is in a PR where Shougo, another long-time contributor, communicates entirely in walls of unparseable AI slop text: https://github.com/vim/vim/pull/19413
What a pathetic state after decades of active, thoughtful work. "I asked the chatbot how to write this code", "Well, I asked my chatbot, and "he" doesn't like it". What a fucking embarrassment.
Replying to
Seems like there's a concentrated push to find someone, anyone, to blame for the complete flop of AI: https://2ndbreakfast.audreywatters.com/and-i-would-have-gotten-away-with-it-too-if-it-werent-for-those-pesky-kids/
[...] companies are responding to this failed revolution (that is, to the lack of interest, lack of demand) by ramming the "AI" down our throats whether we like it or not. Students didn't want to use Khanmigo, the Chalkbeat article explains. So "now Khanmigo is incorporated directly as a way students can get advice as they’re working through specific problems. A spokesperson said the organization made this change because 'students were not seeking out Khanmigo’s help as much as we had hoped.'" Don't want it? Too bad. It's part of the curricular infrastructure now, suckers.
Replying to
If I had to put down the snark for a bit and explain why people are blaming all the breakage on AI, it's because 1) a lot of it is *explicitly attributed* to AI (which the article even admits), and 2) the boosters insist that it's a 10x multiplier and it makes coding so much better, so when their code is still shit, it's kind of natural to point it out.
Add mastodon to the list of projects I use that is allowing LLM slop: https://github.com/mastodon/.github/pull/6
They "do not encourage" LLM-assissted contributions, but they don't ban them as long as you pinky swear you were super-duper careful.
"We are a small team supporting a production project; our time is best spent working with contributors who understand the project's technical requirements, and the safety of our users' data."
Then maybe you should *ban* the slop machines that generate large amounts of insecure and buggy code that you need to review much more carefully than anything else?
Replying to
Yeah, I don't know Kyle, maybe you need to step away from the chatbot, take a mental health break.
Boosted by @kkarhan@jorts.horse
Replying to
In a just world, everyone involved in this decision would be fired and barred from ever working in healthcare again.
It always impresses me to see AI boosters reach new heights of cognitive dissonance. This guy has decided to reject all code contributions: https://github.com/jesseduffield/lazygit/commit/aef4feb501d12c670bb9bbed92d9ff102509cd84
But also, he believes that "coding agents are capable of producing amazingly high-quality code", but also also, "AI generated code needs to be carefully reviewed and iterated on".
How is it both, dude? How are you keeping these two beliefs in your head at the same time? What an astounding feat of mental gymnastics.
Replying to
It's sooo difficult to determine the ethics of taking money from Anthropic, almost as if maybe the consequences should have been carefully evaluated before accepting.
For fuck's sake, another respected open source project to add to my personal slop list
Replying to
Wow, this guy wrote some tool that tells him changes in code quality in exact percentages, that sounds like a totally real thing you can do, or possibly complete horseshit, who can say
Replying to
@davidrevoy The CEO of OpenMagic assures us that Avian Intelligence will be casting 90% of all spells in the magical industry within 6 months. Companies can now fire most of their mages, I mean... Mages can refocus on important higher-level tasks.
Replying to on mstdn.social
@dazfuller@mstdn.social This, to me, is a perfect encapsulation of the value proposition of chatbots: do what you can already do, but worse.
The original image is available on the author's blog as a high-resolution PDF, explicitly licensed as Creative Commons BY-SA. It is *trivial* to find the article and get the canonical image. Instead, somebody asked copilot to regurgitate one, or possibly asked it to "upscale" a low-res copy of it.
Doing the work well would have been faster and easier, but the vibe must flow.
Blender has heard your concerns about taking money from Anthropic and they have carefully deliberated the issue and decided to... take the money anyway, lol.
"Blender is a tool for artists and creators, it’s made by humans for humans.", they say, while having commits "assisted" by claude code in its codebase.
That's pretty much the pathetic outcome I expected: Nothing substantial has changed and they're throwing in some hand-wavy feel-good phrases to pretend like they give a shit.
I want to support artists, I wouldn't mind giving Bandcamp a cut, since they're promoting the music, but... PayPal? Come on.
I continue to be pissed off at how difficult it is to give money to artists in exchange for their music/movies/books in the year 2026.
@afilina@phpc.social A year ago or so, I was reviewing bullet points for a group project that one of the group members obviously generated via LLM. I figured "fine, I'll fix them up". They were these generic marketing statements like "the project is highly performant, streamlined bla-bla" or something.
I kept thinking "okay, so this refers to X or maybe Y, so I can word it like..." and then I caught myself. These bullet points didn't refer to *anything*. They were just generic fluff. You could transplant these statements and put them on top of any project, and they would apply just as much. The context I was imagining was all in my head, *I* was interpreting these statements based on what *I* knew about this project.
If I had managed to "fix up" the wording somehow, the actual characteristics of the project would not have come across at all, because from the outside, readers would not have my context. What made the project interesting was nowhere in these points and just changing the style of text wouldn't fix that.
LLM output is a cognitive hazard. If you try to "fix it", you'll end up assigning meaning to the random text and then end up with something generic that is still just as bad from the outside. You have to start from a position of wanting to communicate something with your writing, and then fiddle with the style, otherwise you're just writing a horoscope.
Replying to
The deeply stupid article in question: https://www.forbes.com/sites/the-wiretap/2026/04/22/anthropics-claude-is-pumping-out-vulnerable-code-cyber-experts-warn/
I didn't finish it, my eyes started rolling so hard, I was worried it would cause some kind of permanent damage.
Replying to
Wow, it would be sooo alarming if developers were generating low-quality slop at huge volumes. "The ultimate risk", wow. I wonder if there are people out there that have been alarmed about this for like a year?
Replying to
@david_chisnall@infosec.exchange Every time someone brags about some good result coming out of an LLM, my first assumption is that this was just something that happens to be written down somewhere and has been regurgitated by the slop machine close to verbatim. It just gets harder to tell, because the slop machines obscure their output, making direct web searches not match as well.
> ‘You also have to create an environment in which entrepreneurs can be successful. Regulations in Europe and Belgium weigh too heavily on business. We need to find a better balance between growth, investment and risk-taking on the one hand and regulations on the other.’
> He points to the rapid rise of AI: ‘We have the European GDPR rules (that protect citizens’ privacy), we have an AI Act (that keeps the use of artificial intelligence within reasonable limits). It gives the impression that we want to have everything under control before we even start. There needs to be some room for uncertainty and experimentation.’
Fuck Belgium's central banker Pierre Wunsch
Food $200
Data $150
Rent $800
Tokens $30,000
Utility $150
someone who is good at the economy please help me budget this. my family is dying
Python's widely-used "requests" library now has at least one commit with a "co-contributed" tag by claude: https://github.com/psf/requests/commit/b2a1d33f571518ca9a6148e7da787cc5827f897a
What blows my mind about this is that the contents of this commit is... a version bump. That's it, a version bump.
This guy used the most expensive technology on the planet to perform a version bump. Bravo, dude, what a stunning performance.
Replying to
@cthos@mastodon.cthos.dev "In order to remain efficient, we are using the most expensive technology on the planet that has so far not resulted in any observable efficiency gains in any tech product"
Replying to
@cap_ybarra Dario calls up investors, shows them Blender's glowing praise of Anthropic, says look, open source people love us...
Replying to
@cap_ybarra Why do none of these amateurs use Ultimate Transparent Thinking Beast Mode, smh https://github.com/AungMyoKyaw/awesome-copilot/blob/71c95f17a6111a72e410201a6564a06975ee2989/chatmodes/Ultimate-Transparent-Thinking-Beast-Mode.chatmode.md
Replying to
@cthos When METR wrote a blog post about how the people they were talking to literally refused to do their tasks without AI (https://metr.org/blog/2026-02-24-uplift-update/) saying things like "my head’s going to explode if I try to do too much the old fashioned way because it’s like trying to get across the city walking when all of a sudden I was more used to taking an Uber.”"...
I realized that at this point, if there's another study and it shows these people being faster with AI, it's literally impossible to tell if it really makes them faster or if their brains are gone to the extent that they're helpless without their chatbots telling them what to do.
It's darkly funny, but also kind of worrying. It feels like we're beyond being able to test these issues, because there's nobody left that is in any way a "moderate" user. Either you're the kind of person who does their job, or you're the kind of person who outsources their entire brain to a random text generator and there's nothing in between.
Replying to
@cthos It's not really a "study", exactly, just a blog post where they say these findings are unpublishable because of this issue. They also don't include any of the estimations from the developers about how faster/slower they think they were. That, to me, was the much more valuable part of the study.
In this case, if they did try measuring that...
I found I am actually heavily biased sampling the issues … I avoid issues like AI can finish things in just 2 hours, but I have to spend 20 hours.
This person literally thinks they are 1000% faster with AI. Given the results of the previous study, how close do you think that estimation would be to reality?
Replying to
@tppetkov There have been attempts to compile them, see the "sibling repos" here: https://codeberg.org/small-hack/open-slopware#sibling-repos
Replying to
@ariaflame A foolproof method of writing "LLM contributions are not allowed" in the AI policy of your project? I think it would be pretty foolproof to write those words in that document.
Scrolling through a github PR diff and some of the javascript and html files just don't have any syntax highlighting, seemingly arbitrarily. I refresh and scroll all the way down and the syntax highlighting is there. Good job, Microsoft, very modern.
@noboilerplate Annoyingly, this doesn't seem to work if claude was tagged as a co-author: https://github.com/rails/rails/commit/546345c8f708bdadc2b3124ddb458673ae34c63d
Replying to
We believe the attacking group to be highly sophisticated and, I strongly suspect, significantly accelerated by AI.
Vercel: "We used slop machines to generate a bunch of code we don't understand. Clearly, the only way this can be hacked is by another slop machine"
Replying to
@ariaflame oh, you're looking for magic? I'm sorry, I didn't realize you were a child.
See, words don't do anything, actions do. But first, people need to decide what actions to take. When someone submits an 18,000-line "AI-assisted" PR, do you reject it on principle, or do you painstakingly review it in good faith? You're gonna have to decide what rules you follow and write them down in a little document, so that you can choose a specific course of action.