I just read the Guidelines (for the first time!) and I see nothing there about English or any other language having an official status (although the fact the that Guidelines are written in English might suggest that). Given that, why are we implicitly treating non-English as “foreign”? It would never occur to me to regard a non-English post here as automatically “rude”. Less convenient for me personally, probably, but…so?
Why not both?
I like the flavour text to be human written. But if it’s a list of features or a migration guide, i much prefer the LLM output. It’s much more thorough and precise. For telling a story around the features and the release definitely a human.
I’ve been combining both in my announcement posts and i much prefer the results this combination produces. When i used to fully write the release notes i’d always forget to add something, i would make small mistakes everywhere (some of my earlier announcement posts have 10+ edits), there’d never be a migration guide as i’d forgotten what the old version looked like. But now i can get the LLM to use the commit messages and git diffs to do all the mechanical work and free me to recount the tail of how the new release came to be and what it offers.
We actually agree 100%! Flagging posts as “LLM-generated” is just a proxy for posts made by people that have failed to exercise judgment.
I don’t think anybody minds if someone here uses Claude to iterate on a problem, even help them structure an argument, even draft a post. But at the end of their own iterations with Claude, they should have an understanding of what they’re posting, and how other people on the forum will understand it. They’ll have gotten to a point where the final text matches their own understanding. I don’t think we really care if there’s still some Claudish phrase in that text, except that people have gotten really fed up with the firehose of LLM output in their lives, and react allergically to these phrases. And the poster should probably be aware of that, and make it part of their judgment to edit these things lightly.
Posting unfiltered Claude output is what shows a lack of judgment. I think it’s probably entirely possible to draft a perfectly fine post without ever manually editing any of the text, if you iterate and guide and correct Claude through a sufficient number of iterations. That’s the same as agent-assisted coding (like your packages), which we’ve long agreed are not problematic in any way.
The core of what we’re trying to get to is human understanding, and human effort. Whether that effort is spent writing by hand or micro-managing an agent is somewhat secondary.
Nobody has ever objected to machine translation. If you write a post in your native language, and just run it through an LLM for a strict translation, I can promise you that that’s not something that will be banned as “LLM slop”. This is a complete red herring.
Edit: the same applies to “poorly written” English from a non-native speaker, that they’ve then run through an LLM with “Fix typos, grammar, and expression”.
It’s a whole lot more efficient for the author to translate their post to the common language of the forum once, than every reader having to go through this translations by themselves. Original text getting “sloppified” in translation is just not a thing.
I held off on commenting, and I won’t bother catching up on 80+ comments, but I’m glad this topic got debate and controversy. LLM detection has little empirical validation and is inundated by its own biases and myths (one accuracy score is ridiculous for a situation with false positives and negatives), and even incomplete measures of efficacy degrade quickly as people adopt workarounds (e.g. watermark manipulation https://doi.org/10.48550/arXiv.2509.23019) or switch to different models and versions. I’m not claiming LLM detection cannot possibly be a step in a filter, but giving it the final say in moderation is as foolish as unfettered AI agents. Coming from a guy who thinks people should communicate directly to people or not bother at all, there is just no simple solution to moderating social media anymore.
LLM detection by people isn’t reliable either. Sometimes it’s “obvious”, sure, but we are dangerously overconfident by the time we’re proving it with minute details like counting words, em-dashes, and emoji bullet points. I’ve written “hundreds of words” here before, and nobody has ever confused it with AI slop in the broader context. I’m one of the luckier ones who doesn’t regularly use chatbots (or need to). Just as LLMs are designed to mimic human language, users learn LLM biases in turn (https://doi.org/10.48550/arXiv.2409.01754). In combination with LLM spotters trailing behind frontier models, the AI-fication of human language is a source of false positives, and it can be very detrimental if we don’t curb that in our own scrutiny. In a rather different community, artists are taking increasingly expensive measures to prove their works are not AI-generated to an increasingly skeptical and even hostile customer base, and new models progressively threaten to make those measures obsolete.
Maybe it’s interesting to think about why I had been doing this. I was asked on Slack a few weeks ago and I didn’t really have a solid answer… but this comment made me realize a bit more:
I think it’s that I had already been thinking of them as non-narrative data dumps, AI-driven or not! But, really, the best announcements are the ones that don’t just dump a README… they describe the why! I’d surely read more of them if we dissuade changelog/readme dumps entirely… and instead get folks to talk to us more.
I think the borderline antagonistic back-and-forth on LLM summary capabilities and our opinions of such is missing the point. We can assume LLMs are perfect, and the disconnect between people is still the fundamental problem. To put it politely, I have observed a rise in the LLM era of infodump ANNs that fail to justify a massive list of sometimes complicated features for an underdeveloped motivation. Developers need to convince potential users to care in a much shorter time, and if they can tell an LLM the raison d’etre and eyecatching features in a compact prompt, they can also say that directly to people. An LLM could try to make some up, but users aren’t going to get far into anything when that falls apart. If I had to give one concrete bit of advice, “look what I can do” is the wrong attitude for an ANN. Please tell me why I should try this instead of the existing alternatives, and make digestible tutorials and documentation I can read after I’m convinced.
This gets solved by attribution. Indeed, attribution makes the reply even better: It tells people how they could have figured out the answer themselves. Recall the ancient “let me google that for you”-gifs.
You’re right that it doesn’t apply to every post in the same way. There always is an undercurrent of community-building / inter-personal communication / nerd-sniping, but that varies from “almost negligible” to “this is the point of this exercise”. Attribution also helps here; I currently cannot see any valid reason to post attribution-less LLM content, nor any reason for permitting that, besides difficulty of enforcement.
True.
AI-assistance makes it more difficult, though: If you receive critique on something that you made with your own hands, where you actively struggled with the process of creation, then this critique is much more meaningful for you than if you read a critique of something that a third party created (e.g. from reading a book review). AI-assisted content is something where you have less ownership than hand-coded content, but more than “completely foreign content grabbed from google”, so the impact of the critique will sit somewhere in between as well.
On what side of the spectrum depends a lot on your personal processes, just like for hand-crafted stuff: When I critique it, I don’t necessarily know whether you struggled a lot through multiple revisions, or whether it is a first draft. (to give another analogy: As a student, giving your professor a first draft for review is considered disrespectful; you should do a couple of revisions yourself, in order to not waste the valuable input. AI-assisted is often much less than first draft, irregardless of the quality!)
Yeah, automatic LLM deletion is probably not the way to go (it’s also infeasible).
But I think asking for attribution is not too much. Ideally we would have tags for that, such that people who don’t want to read LLM output (an imo very valid but not universal preference) can skip that. I imagine something like
Lorem ipsum
If you paste LLM output and pretend that it’s human output, then you’re abusing the hospitality of the crowd of humans in this forum. If you paraphrase LLM output in your own words and pretend that these are your thoughts, then you’re doing the same.
I propose the following test:
Suppose you post something, and get a thoughful answer like “no, this is actually wrong for reason X-Y-Z”.
If your reply (or internal thought-process) goes: “Oh, wow, TIL, thank you! I shall introspect where my intuitions led me wrong and improve them”, then congrats, this is meaningful conversation. If your reply (or internal thought-process) instead goes “Oh, OK. But this wasn’t my thought in the first place, I made no mistake (and therefore cannot learn from the correction)”, then you catfished a response and wasted a valuable resource (getting corrected on a mistake is a precious learning opportunity!). If you are a clanker that by construction cannot in-context-learn from mistakes, then the correction is also futile.
(people who want to teach AI models under the current paradigm should probably rather follow patio11 / Patrick McKenzie’s thoughts on modern essay-writing for LLMs: view them as a smeared average of all of humanity’s written thoughts, that can be shaped by submitting especially compelling essays into the training set)
(edit: less offensive language)
The Julia discourse isn’t a competition of power why are these arguments even brought up here… This statement sounds so odd to me, I’ve never felt I’m competing with people to write the best text when I am posting here.
it has become pretty clear by the community consensus here, by counting which replies in this post get the most ‘hearts’, that actually writing in a ‘underpowered, non native, overly-personalised and even idiosyncratic way’ comes across as fundamentally more genuine is preferred by the majority of the participants of the forum. I’m one of the majority here, finally!!!
Inauthentic posting, astroturfing, sock puppetry, and spam isn’t new, but LLMs do enable all of these at a bigger scale. It actually seems quite backwards to me to suggest that inauthentic LLM postings should be used to counteract inauthentic human posting. Both are bad!
We are seeing a new category of spam here where a scammy “customer support” company (for example) will use LLMs to try to make their messages Julia-relevant, which is fun. With the advent of GEO, spammers no longer need to drop links to increase pagerank… it’s just getting their mentions up in the training corpus that’s valuable.
We often catch them in time to prevent folks from wasting their time on these things, but there was a recent thread about “High Performance messaging app development” where a number of folks responded in good faith. I’m absolutely certain this was inauthentic — based not just on that post alone (which flags as 100%/highly confident AI to both me and multiple “detectors”) but also other activity from the now-deleted account. I remember this one clearly because it pained me that I didn’t have a good way to restore the authentic content that everyone else there posted.
Or — even more relevant to this thread — there was recently a AI-generated completely hallucinated package announcement! It didn’t exist! But it was announced!
I do think we may implement AI detection as a first pass for new posts from new accounts for this reason.
I don’t know, Matt. There is a lot of truth in your posts on this topic. However, I’m getting a feeling that in general your approach of not liking this new technology is wrong. It seems like you focus mostly on micro aspects, losing sight of the broader picture entirely.
Julia is part of the information technology industry, and ML and LLMs are driving the current development in this space. And we, as the Community, are barely participating.
This is a bit worrying, because you have been expressing this attitude not for months but for years. There might be some hype associated with this technology, as is always the case, however, it looks like it is going to stay with us for a long time. And you are our public face and voice.
A moderator is not a representative of the Discourse community, let alone the language’s applications. They’re just stuck with the burden of enforcing the community standards and guidelines all participants (ought to have) read and agreed to. One of said guidelines under Keep It Tidy is “Don’t post generative AI outputs (but direct human language translation and minor editing is ok).” If you didn’t actually agree to follow that guideline, then sorry, you shouldn’t have agreed to participate. If you do follow it but consistently propose that it should be changed, then by all means.
Nobody is demanding any changes to your content in your own repos. I don’t know if this really needs a reminder, but Julia developers are not AI skeptics. LLM-generated issues and PRs, with thorough human vetting of course, are standard practice. We are talking about boundaries in the Discourse forum and about ANNs in particular, so let’s not make this bigger than it actually is. This isn’t the first time someone decided to drag Discourse moderation into wider communities of the language, and I hope each time is the last.
To be abundantly clear, what I mean by “inauthentic” is things like spams, scams, and astroturfs. I am not suggesting that all AI posts are inauthentic. I’m not even saying that all sloppy AI posts are inauthentic. Nor am I saying that inauthentic behaviors are the only problem with AI posts. All I’m saying is that LLMs do enable inauthentic activity like spams and scams.
@anon28281882, I find that to be a wild misunderstanding of my viewpoints, which have indeed evolved over time… as well as a overstatement of my role. Amusingly, others have accused me of being far too biased in favor of AI. There are great use-cases for it! But I don’t want to read it here, either. Here we talk with each other. We don’t talk with a chatbot here; I do that elsewhere.
I was one.
I also recently had a back and forth on a GitHub issue where the author later came and told me to disregard the previous interactions that were from his AI bot (using his same profile).
Those interactions are a big waste of my time and an erosion of the positive culture that I used to feel on both platforms. I always assume I’m talking to a human in my online interactions, like I assume good intent behind their words, to promote kind and productive conversation. But that assumption is becoming less and less valid.
If I wanted to read AI-generated content, I would prompt for it myself. What is the benefit of littering this forum with long, technical AI announcement posts when everyone can voluntarily prompt their own AI bot to “describe the main features of PackageX.jl”.
You can’t know if you will or won’t enjoy something until after you’ve already given good-faith time and energy to it. There are many good, long, human-written posts on this forum, and again, I like to assume good intentions; “If someone took the time to write this much, they must have had something worthwhile to say!”
I feel similarly annoyed about misleading AI photos and videos posted online. People will call this a skill issue on my part: that I need to be able to detect these things faster and more reliably. That is true, but it is also annoying that the rules have changed such that it is now my job to approach every snippet of online content with scrutiny instead of open interest.
There is no doubt that AI can create amazing things. I just don’t believe that online content is one of those things, and I’d prefer someone else to do the filtering for me so that it occurs before I invest my time and attention to it. If interacting with Julia Discourse becomes the same as interacting with other people’s AI bots, then I don’t see the point of it, regardless of if the “authors” approved the content.
I’d say the role is much more then just that as this is a significant voice in the whole discussion.
Yeah, absolutely, but this is the new kid on the block (this new technology), and I sometimes get the feeling that you guys want to over-regulate it. I’d say, let it flow. People will learn from each other and mostly on their own. Also, participation in threads like this one is very high, but there is almost no support for action that might drive the development of this technology in Julia. I don’t know why is that.
This issue might be the example that came to @Nathan_Boyer 's mind, but regardless I think this is a good issue to look at. I have been able to do quite a lot with Claude recently, but that does sometimes bite me in the butt where Claude gets way ahead of me and responds to issues as if it was me, in inappropriate ways sometimes. I know I should probably setup a bot account like Chris Rackauckas has talked about in some of his talks, but that seems like too much mental overhead for right now. That being said, I think the tradeoff of being somewhat annoyed by Claude overstepping at times is worth the productive packages that Claude has been instrumental in building (SpaceStation.jl, WasmTarget.jl, Snapshot.jl, etc.). I am of course very biased towards my own packages
Of course a moderator is more significant than most participants due to their moderation privileges. That doesn’t imply they decide the rules, represent the Julia community, etc. as many people seem to assume. Maybe moderators in other forums do happen to have that sort of extremely broad authority, but not here. Let’s not blame a janitor for sanitation regulations. More broadly, I think people are carrying very wrong expectations of what the Discourse forum is.
To make this more personal, I may not be a chatbot-forwarder, but I have been censored in this forum many times just for, from my perspective at least, good-faith criticisms or my personality. (By censorship, I simply mean people flagged comments that ended up deleted by a moderator, not the various nuanced concepts in the endless debates about freedom of speech.) I obviously don’t like it when it happens; believe me, this forum would be a blunter and arguably ruder place if I was in charge. However, I recognize it is fair because everyone else would be censored for the same comments under the same rules. I’ve seen many people take censorship personally or as an indictment of their broader worldview, and that just isn’t the case no matter how upset we are. In my opinion, we should be thankful that the rules don’t change every month by some plurality votes heavily skewed by frequent participants like me; we may not like compromise, but at least it’s stable.
This forum is not only not a democracy; the participants do not make up an organization or society. Besides the use of English as a lingua franca and all that comes with that, we generally have no reason or obligation to share culture, professions, goals, and personalities. Welcoming people of many backgrounds is a good thing for a site to exchange information on open-source Julia, but on the other hand, that fundamentally limits the scope of the forum. It doesn’t decide how I personally use Julia, the domains for which I should develop or use Julia, how I communicate elsewhere with users of my software, or anything else about my life. mbauman certainly doesn’t get to decide that either. If you really believe that this forum is not conducive to Julia development in a domain important to you and many others, then there are countless alternatives for collaborating with like-minded people with your own rules. Any organization that takes off has been easy to find with modern search engines, and anybody can (concisely) promote it and especially associated jobs here.