No, the suggestion by the original author was to implement automated LLM detection and deletion of posts. That’s not good.
But OP want to ban LLM generated texts. And my “review” does not necessary include “edit”, which means I think post fully LLM texts is acceptable without any revision by human.
That sounds a bit too good as to be true. The discourse guidelines are stricter than you make it sound:
- Don’t post generative AI outputs (but direct human language translation and minor editing is ok).
This is different from allowing endorsed and reviewed content.
I agree, and not only that, but General registry policies were brought up as well, outside of discourse. This included mention of README documents and non-reference parts of package documentation, in addition to announcements, grouped together as “user-facing communication”.
It seems to me that some of these policies are perhaps quickly making less and less sense, if taken as face value.
I personally don’t align with this statement about lacking judgement, because I believe we are still talking about humans who choose to click the post button? The question therefore is not about the “judgement” of the LLM, they don’t possess that. The person contributing to the forum has the judgement, and of course we’d like that to be good. But whether that person used LLM tools in the generation of user-facing communication, might not be the correct facet to focus on.
Maybe the simplest solution would be disclousure combined with user choice rather than prohibition, e.g., not wanting to read something should not necessitate not being allowed to produce something.
Couldn’t we mark individual posts as “created with AI-assistance; reviewed, understood, and endorsed by human.” And we may even introduce a flag for a profile “regular user of AI-assistance.”
Then other users may simply decide to mute/filter either posts or even other users, when they do not want to see such output at all.
People, who are happy engaging with such content or with such users, may then judge simply by content. Normal moderation would still apply.
That seems preferable to a categorical ban to me.
Funny thing is, I still don’t even know what the original announcement was that sparked this flurry of opinions. Where is all this problematic content that has people in such a tizzy?
Thanks, that’s very kind of you. But which parts of which responses in this thread were written by an LLM? Any of it? All of it? Maybe I’m actually a bot? 
Why would you, specifically, get the slop flag?
I don’t know! That’s what I was commenting about, I’m really confused because I don’t see any announcements which bother me, and I went to the channel and looked through all of them in recent days. So I feel like I’m addressing a hypothetical here. But I don’t see the issue.
I think the issue of people who are not at ease with English is important. Do we have a rule here that postings need to be in English? Many of us can read other languages, and we can all paste text into a translation program or service. I don’t think an LLM ban will hurt such people. If English is required they can write their announcements in their native language and use machine translation. The output of, for example, Google Translate is not the same as LLM-generated slop generated from a prompt. And how can someone who doesn’t know English judge whether the slop expresses his or her thoughts?
My feeling is that in this forum we come together as humans. So we should be able to speak as such and there should really be no need to polish posts to a high sheen with LLMs. I think that people often just do that to save time, not because they couldn’t express their opinion before, even if they are not native speakers (neither am I). I’m personally fine with a short announcement post saying “hey I’ve got a new package for X, if you’re interested check out this more detailed description at URL” which then might contain generated stuff (with a disclaimer maybe). Then it’s clear what to expect and for such a short post no LLM polish is necessary anyway. As this forum is for talking to other people and not just pinning stuff to the wall like an ad, I think it’s not the right venue for the LLM long form, even if it has all the details, has been reviewed for accuracy, etc.
And I don’t buy the “I’ve reviewed it so it’s fine” argument in this specific context. It’s simply not you the person who wrote it. For me that’s all I care about. As a hyperbolic example, would you have a bot impersonating you call your mom everyday to ask her how she is saying “well I’ve reviewed the task and the bot output and it functionally does the same that I would have done”? Hopefully no, because it’s not about the information you get but about the communication itself and building community.
Thanks, that’s very kind of you. But which parts of which responses in this thread were written by an LLM? Any of it? All of it? Maybe I’m actually a bot?
I also don’t buy this one. There’s very long announcement threads these days with all the telltale LLM signs, em dashes everywhere, overuse of formatting, unusual phrasing. It’s really not hard to tell with these. Even if the content itself is fine, as I said, for me it’s a data dump and not something for a forum of humans. For me it’s not a question of “how do we police this” but simply a matter where I’d be happy if everybody could be on the same page. Just “let’s try to keep the human communication at the center here” and not “but how can you even tell haha, maybe I generated this, too”. Like, I would prefer it if we could embrace the imperfection of humans and human communication instead of finding all these arguments why automating it away to avoid a little friction is fine.
Is this that hard to understand? As a non-native English speaker, I am well trained for reading but not writing as well as speaking. So I can read academic research articles very well, but there is quite barrier to write. That said, before LLM, I am struggling to publish my research results.
As far as I can tell, there is no mention of English being the only language allowed, so I’d say people should post in whatever language they feel will be most effective.
Orfay Eemay, Atthay illway eebay igpay atinlay.
However I think that many people will continue to find that English is for at least some reasons going to be the most effective language, because – for better or worse – it’s become the global language standard for a lot of science and technical communication.
In that case, LLMs can be very helpful, you should read the reference from my first response to you : https://arxiv.org/pdf/2511.11687
I think they probably do need to understand some English for best results, but LLMs can likely do a pretty amazing job, when they can converse in their native language to provide the context and framing, and then utilize the LLM to generate prose that matches. It’s beyond what Google Translate does which is literal translation.
I mean, I work on platforms designed to help people utilize coding agents. I’m not writing out the code for these in some abstract specification language and then using a transpiler to convert them to Julia … I’m using the LLMs to generate code from the concepts and specifications I input, in a natural language form. And today, they are shockingly good at this.
I do understand code syntax, so that’s helpful in understanding the output of the LLM, but I certainly am not making sure it’s a “word for word” translation, it’s generative, and expands on my ideas, and then I make sure that the things expressed embody my original intent.
However your guess about whether LLM bans would be harmful? I don’t think you’re correct there.
Where is it then? Because I haven’t seen this unbearable announcement.
Because I haven’t seen this unbearable announcement.
I haven’t said unbearable, it’s not nice of you to try and discredit my argument through hyperbole and snark. I just said many of the latest announcement posts are probably generated and I would prefer if people just wrote themselves what they wanted to say to others here.
That description is part of the foundation of this post. From the original author of course. I didn’t intend to attribute that to you. But I wouldn’t say that my writing is hyperbolic at all, I’m responding directly to what was said from the very beginning.
Which ones are bothersome to you? Feel free to send to me through DM if you’d prefer.
So what about disclosure? Why should your preference for human bonding and human-only production lead to a ban, when a simple filter may suffice? Disclosure would also be declared instead of tool-inferred.
I didn’t say it should be banned, I just said I don’t like if people do it (post LLM output as if it’s their own voice). I don’t think it can be banned because it can’t be reliably detected anyway. And production is not the right word because it implies there’s a product (like code on github). I’m just talking about communication, a lively back and forth. That’s I guess why ANN posts are the point of contention here because they’re something halfway between communication and product.