Or what about simply choosing to engage with whomever you enjoy engaging with? Couldn’t this be self managed, people can mute others if they find them bothersome, curate their own experience, and that doesn’t impact the freedom for others to continue to read that content?
Why would we want more censorship, when Discourse offers all the tools necessary to customize what you see and read here? Only in EXTREME cases of spamming bot activity, threats to personal safety, DDOS, etc. would I think that any moderation is actually needed.
No, it is easy to understand; anyone who has learned a second language will understand this. It’s related to the fact that it’s easier to recognize a word than to recall one.
But it just means that you will be able to understand the slop once it’s generated, but not be able to see how horrible it is. Again, automatic translation is good enough now that it probably won’t misrepresent too badly what you wanted to write.
I am 100% agreeing on this point. I started flagging a few in posts that are too long, I also stopped reading them.
If someone does not put the time into thoroughly writing what I shall read why should I care to read? Or should I ask an LLM to summarise someone else’s LLM slop? Not what I want.
Sure, improving writing with some help might be something some people like. I am fine with my semi-okay, maybe too German style of long sentences that never end.
For package announcements, I do prefer the shorter ones that really focus on what a package is about and not try to give me all features in a whole novel – that is what docs are for (please do not LLM generate those either).
There was some doubt raised about the accuracy of automatic LLM detection tools. This is a rapidly developing field and papers published in 2025 are already out of date. Nevertheless the idea of automatic deletion obviously makes no sense if there is no reasonably accurate way to detect slop.
You can find academic papers claiming accuracy results conforming to any result that you prefer, so it’s easy to make a point by cherry-picking papers. But maybe this is a good place to start:
Van Vlasselaer, M., Van Droogenbroeck, F. & Spruyt, B. “Who wrote this? Evaluating the reliability of AI detection tools in higher education.” Int J Educ Integr22, 16 (2026).
The upshot is that Pangram has a 0% false positive rate and about a 95% slop detection rate.
(Even the other tools studied, which did a bad job at detecting slop, had essentially 0% false positive. I think this is important, because, as offensive as it is to encounter slop where we were expecting the thoughts of human beings, falsely flagging human text as slop would be worse.)
This is extremely interesting: AI (or automatic translation) is allowed to compensate for my language deficiency, but it should not augment other abilities that I have. It is even doubted that AI may augment abilities, because what AI produces is typically assumed to be “slop.”
What if I understand, say, a Bayesian model reasonably well, but fail to communicate that understanding clearly. Why should AI not help me overcome my lack of expressive skills?
Now, you may find this quite human of me and you will nicely help me, but at the same time, my deficiency is out there for anyone to read and attached to my name. That is something worth pondering: Why shouldn’t AI be allowed to compensate for other limitations in expressing my thoughts?
This isn’t a solution, just a personal impression, but I think it’s good that these discussions come up as often as they do. Just seeing that a topic like this divides people might make individual community members stop and think about their own use of AI.
English isn’t my first language, and I’m still new to Julia and have a lot to learn compared to most people here. Looking back, the reason I started coming here in the first place was that I wanted to know how the community feels about AI use. Before that I mostly spent my time on a small Discord server in my own language.
I’ve found that teaching, both in person and through writing, is immensely valuable in showing me, the teacher, what I don’t understand as well as I thought I did. It’s easy to become entranced by the illusion that you understand a thing, but when you try to get someone else to understand it, you suddenly realize that your understanding was imperfect.
I would never sacrifice this magic wand by getting a computer program to generate explanations for me. I would remain in the dark, falsely confident that I knew what I was talking about.
I can easily imagine fora such as this one devolving into machines talking to machines, with we humans as spectators, unless a hard line is taken against slop.
I can understand your reasoning. But think this through: If I am using AI before I asked you (as a teacher) for advice and if I reviewed and endorsed it, then I am still asking for your advice and I am not asking the AI for everything.
The difference is, that I have not chosen you to be my teacher for all that I do not know or cannot express, just for that, what remains after I consulted AI. It is a question of who decides what to be taught and by whom to be taught and what to be seen publicly. There is more to this than meets the eye, at first.
I think the question of the capabilities (is it slop or not?) of LLM is a red herring; the issue is authorship not the quality of the content. Imagine you use Claude 9.0 to write your posts and you don’t mention it, it would still be problematic because you didn’t wrote the post. You have an account with a picture and a tag name, the implicit contract is that the text that is is the box with your name was written by you. It’s true that there’s a bit of a blurry line when it comes to translation or minor editing, but if you’re confused about that just ask yourself “I am the primary author of that piece of text?”, if not don’t post it or at least put it in quotation.
This isn’t as complicated as it might seem at first blush. The Discourse guidelines and our community standards already cover it well… even though they were written a decade ago.
There are three or four fundamental guidelines that all content here must satisfy:
Be respectful
Be concise
And take time to improve the conversation! Think about what you are posting!
Obviously I cannot tell how much someone thought about their post. But I can often tell how little someone cared about what they posted… just by hearing the AI’s bog-standard voicings and structure.
It is true that I had been turning a blind eye to package announcements and changelogs… but I’m finding myself agreeing with @goerz here. There are other places to stash AI generated output on the internet. Here we talk to each other, even when we’re announcing changes and packages.
The question is whether you need to be in a “locked room situation in isolation”, e.g., are you allowed to consult with a colleague or read a book before you answer? And if that is allowed, why shouldn’t dialogue with an AI be allowed before you post?
It’s not as simple and clearcut as you make it sound. So effectively, I can discuss something beforehand with an AI and let it structure my argument and then I need to write this, say in German, to have it be translated. Is that ok?
Sure, if the final result is thoughtful, concise, and reflects your own voice and meaning, then that’s fine. The problems with AI output — insofar as we’re discussing here — are the inverse of what I just posted. They tend to be overly wordy, limited in meaning, and don’t improve the conversation. Anyone can similarly prompt their favorite model themself if they so wish.
Perhaps it’s a bit like “all toupees are bad,” but that metaphor runs into trouble quite quickly; don’t take it too far. Point is simply that it’s sloppy use of AI that’s slop… and that’s when people clock it as AI.
No you can’t. You may detect (presumably) AI output, but without dealing with actual contents you cannot judge how much human effort went in.
As an example. Yesterday I spent most of the day discussing with Gemini and Claude some something (this posting is produced with 0% AI, which I do not think makes my English better). The result of our discussions was a document finally written by Claude, with all telltale marks of AI. Still, it is a result of like 8 hours of my human work.
Furthermore, why should I care about the effort somebody puts into communication? In many cases am just interested in the result, whoever produced it and with whatever time/work/money investment. I see an [ANN], in most cases I can see at the first glance if it is interesting to me. In most cases it is not anyway. If it is, I can read it, and I can ask questions. Where is the problem?
Keyword on often. And yes, that’s still problematic IMO, because those “telltale marks of AI” are annoying because they’re typically meaningless dribble. They show a disrespect for the reader’s time and effort.
So, yeah, if you cared enough to spend 8 hours on the thing, I’d hope you’d spend another 10 minutes to clean up the prose a bit for the sake of your readers.