Don’t post generative AI outputs (but direct human language translation and minor editing is ok).
It should indeed apply to the entire site, including package announcements. I’m going to take it a step farther and edit the Package Announcements category description to encourage folks to tell the community about their new package (or new version) instead of info-dumping a README or changelog. That’s why we’re on discourse.julialang.org instead of github.com or claude.ai or chatgpt.com.
It’s a wiki, so folks can continue iterating on it!
What is supposed to happen when someone violates the guideline? Because there is at least one recent (extremely long) package announcement that is 100% generative AI output, verified by Pangram. And it’s still up.
I would like my pkg ANN to be taken down so that I can redo it. In the 2 years plus of adding to it if let alot of llm content be pulled from my changelog and so on to be remade as a discourse post for a ANN. Guilty. Ill redo it manually if its removed. Its not that I cant but I was cutting corners.
Actually, this is the issue I found more irritating to follow.
Details Below
Dale asks me what OS I’m on.
That’s weird. We just went over that in the other issue.
Bunch of background info on icon details per OS.
This doesn’t seem relevant to the discussion, but okay.
Dale asks for clarification on location of the missing icon.
I already explained that in the initial post, but I can clarify again.
Dale makes a bunch of commits without waiting for my answers to his questions.
I guess he figured it out?
Dale says the issue is fixed in a longwinded way and gives me clean install instructions.
Okay, great. I’ll try that.
Dale tells me I shouldn’t be seeing this issue on MacOS, and I should reset my Mac’s Dock to fix it.
Wait, why are you giving me instructions for MacOS again?
Dale contradictorily says there is one remaining issue but everything is fixed now.
Okay, I guess I’ll check for myself.
Real Dale shows up and says everything above was AI.
Oh, that makes more sense.
Real Dale asks me to test the new version but gives slightly different reinstall instructions to the AI instruction above.
I’ll try the union, I guess.
Real Dale closes with a positive affirmation that was definitely written by AI.
Double plot twist! Real Dale was a bot the whole time!
Because that would be rude. I found the communication during the fix distasteful, but Dale (or his bot) did fix my issue, so I decided to just leave on a positive note. I wasn’t abused.
That’s a lot of confidence in your tool there. You’re providing a single value of 100%, are you saying that the distribution is a delta function centered at 1? What prior was used before applying your Bayesian step?
I’d like to know (for research purposes) which recent package announcement you’re referring to. You can share it via DM if you’d like.
As you’re reading my reply, please do so in good faith. Asynchronous communication over text across strangers is borderline impossible to get right. Please try to understand the person behind the screen.
For me the crux of the matter is the complete lack of effort behind a text posted in this forum. LLMs make “authoring” these posts possible, nothing more. You can rely heavily on LLMs to iterate on the structure, grammar, or vocabulary of the text. If you put in any effort at all during the authoring process, that’s immediately obvious. Whether using LLMs throughout that process is for the benefit of the aesthetics of the prose, or whether it helps you better communicate, is another subject entirely. I would never advocate for any kind of censorship based on subjective taste. However, I do take issue with these kinds of posts.
It’s getting increasingly harder to shelter what little time and attention I have left from the always-competing, ever so clever mechanisms that try to steal them away. Engaging with the community here is a conscious, deliberate choice. If all I was interested in was a quick look at new Julia GitHub repositories, I could just directly query the platform and be done with it. These announcement posts are published in a community forum. The community is a living, breathing entity. Every single interaction here shapes, at the very least, what we perceive this community to be. In a day of digital isolation, being the witness to everyone here passionate to share their craft brings me joy. Seeing people come together and collaborate brings me joy. People helping people brings me joy. This is a community I deeply appreciate and care for. I’d like to see it thrive, so that other people can benefit from the compassionate, excited folk that browse these forums.
These zero-effort posts are extremely easy to tell apart. To quote what I said elsewhere: “I’m often unable to engage as a community member simply due to a lack of trust. A lack of trust in whether what I’m reading is genuine, and has real effort behind it.” This lack of trust leads to isolation. It sure is detrimental to me, even if for sentimental reasons. At the same time, I fail to understand how these posts are beneficial to the respective authors, in any way. What is there to gain from copy-pasting an outsourced summary of something you don’t understand, compared to, e.g., just pasting a link to the repo? To me it feels like a lose-lose kind of deal. Can you really not spend 2 minutes of your time to do a very basic polish on text addressing other humans? Why should the others engage? It’s two minutes and basic courtesy.
@Benny (post #101) As for your first paragraph, I’m not blaming Matt for any misconduct by any means, of course. What I wanted to say (again) in my initial post in this thread was that this constant complaining about this technology can potentially affect the development of Julia as a language in the medium to long term. There are many open questions about this tech, some of them of real significance. However, when it comes to our presence in the field in terms of basic research and implementation, it looks like, as a Community, we are probably barely participating. But I guess this is more of a question for the CTO, Keno, than for Matt directly.
As for your second paragraph, oh yeah, some of my posts were also flagged several times. I didn’t like it. In general, usually, I guess it’s hard to make a judgement on the topic of flags. It looks not quite transparent. I’d say it is what it is, and I guess it is a form of convenience feature for the moderators.
As for your third paragraph, (even though I see it a bit as an off-topic), I actually like the forum. I recall writing once here that most of the people participating are really great. Of course, nobody is perfect. My main concern, which I’ve also expressed here, is that it looks like an extremely academic focused place. That came as a real surprise to me, because when I joined a few years back, my understanding was that a piece of software infrastructure like Julia is built for people, and by people, who are interested in creating something. My understanding has always been that programming languages usually sit very much on the applied side, and that the applied side is what drives their development. So it is a bit different then my initial expectations but I still see the software on the positive side and I’m optimistic about it.
I haven’t seen any “constant complaining” from mbauman about AI usage, just upholding plainly stated community guidelines, and frankly, there should be room for skeptics as long as they don’t violate said guidelines. But more to the point, any technology that cannot thrive in the presence of scrutiny and complaints was too weak to thrive in its absence. OpenAI and Anthropic definitely didn’t need people to stop complaining about AI, why assume otherwise for Julia? Moderation in one forum is an even smaller problem than public complaints because it has no power over our decisions and work outside the forum, no matter what our bruised egos might tell us.
Again, if you see some unmet need, you can and should collaborate with like-minded people in more suitable places. That’s true of all longstanding open-source Julia organizations, and Julia itself certainly isn’t developed on Discourse. I promise that taking a little time to write ANNs and talk to potential users from the heart once in a while doesn’t get in the way of that or any software development.
I agree, there seems to be very little to gain from automated deletion of posts as it pertains to a few announcement topics. The notion that an announcement thread simply existing which doesn’t tickle the fancy of some Discourse members should result in suppressing the communication from other members is incredibly misguided.
Again, we’re talking about piece of content which is optional to read. If you see an announcement thread, click into it, and decide it’s written in a way that doesn’t fit your taste, why not stop reading it and switch to something you find more enjoyable? We’re not talking about misuse of the forum, spam, bots, or denial of service. We’re talking about people who are creating packages they think might be useful or interesting to others, sharing that and gifting it to the community. Why not celebrate this a little more? Why is there so much negativity here?
I am personally addressed and accused of being “highly reactive, angry” …
when in fact, I’m just appalled that this community finds it acceptable to suggest auto deletion of user contributions, purely based on some notion of “taste” preferences. The suggestion is to use LLMs (which apparently are “sloppy”) to infer whether LLMs were used to generate some prose? I find that distasteful. I’m accused of turning this into a trial about me, when in fact I’m directly talking about the effect this has on all of us. Go read some of my comments in this thread, if they haven’t been censored yet.
If I click the button, they are my words, I’m the one deciding to post them. There is no chatbot running here.
Maybe a community should be a supportive environment? Maybe a community should not focus on criticizing and tearing at the work and creations of its members? Maybe a community should not be so quick to judge and make assumptions about the context other members work in, what they are comfortable with, their own preferred work habits? What do the rest of you think?
God knows I am not enthused about AI in its various guises, but I do question the tendency expressed from various quarters here to proscribe posts produced with AI (LLM, …). If it is good enough to write technical documentation (with a tweak here and there by the supervising human), why not post text written partially with that technology? I would probably not enjoy reading something entirely written by the tool, I admit that.
So, why not have a rule that a post produced (partially or fully with LLMs) must be labeled as such prominently at the top? That way, someone hating AI-produced text can skip it, and we are not overly prescriptive as to the communication technology in use.
You’re free to see it that way, but if your words are objectively primarily generated by AI, then the guidelines are clear about that not being welcome despite its rather loose enforcement.
Unconditional support and discouraging criticism are a tough ask from online strangers and an internal contradiction. What about being supportive of strong criticisms, yours included? If you find them distasteful but the moderators do not ultimately find them to be a violation of the guidelines, then…
Frankly I’m a little surprised this is your perception of the forum because from what I have seen, you have been relatively successful at finding users of your packages and positively communicating with them. Many posters don’t get a tenth of the positivity you have. Granted, maybe most of the negativity got deleted before I had a chance to see them, but that’s a point in favor of the current moderation practices.
I don’t think changing the guidelines like this is tenable. People who don’t want to read LLM-generated text won’t be any happier about scrolling through a feed or topic full of LLM tags, which they will be because LLMs write much faster, and that incentivizes avoidance. People who edit and post LLM texts won’t be happy about people avoiding them, and that incentivizes dishonesty. I don’t know if Discourse can even have some sort of LLM tag filter to attempt to segregate these groups to their satisfaction, but even if there is, a separate forum that doesn’t fetter LLM responses would be much easier (I mean, why manually press “reply”, assign some agents and “let it flow”). There’s literally nothing stopping such a forum from existing, nor would there be anything stopping people from participating in both forums per their respective guidelines.
I think this is a good idea. The only AI generated text I might enjoy reading is the one that got generated after my own prompt. But package announcements are not schoolwork. We are not supposed to evaluate writing skills or a learning process. So AI generated text should be fine in that context. I will always prefer human craftsmanship, but this should just be a matter of personal taste.
I’m part of the “we” in that sentence, Steven. Yes, I do feel that way. I also believe there are others of “us” in the community who might feel similarly.
The point wasn’t about how I feel, though. It’s about established principle for the act of publishing. If I press post, I’ve adopted the words and I’m answerable for them. For whether they’re accurate, whether they’re clear, whether they’re worth someone’s time. A journal editor is accountable for everything that appears in their journal, whoever drafted it. If I ask a friend to look over a draft and they rewrite half of it, and I read it and publish it, that’s still my post. Nobody has ever thought otherwise.
Directing work you didn’t personally execute is how research groups and large open-source projects operate. Nobody thinks a PI who guided a paper or a maintainer who reviewed a PR is misrepresenting themselves. Accountability follows the decision to publish.
Let’s separate that from a bot account posting on its own. It’s not the same as a person drafting with a model, reading it, deciding, and posting under their own name. There is no chatbot running here.
I’d also note this doesn’t seem as settled as your reply suggests:
So which is it? Is the standard human understanding and effort, or the physical act of typing?
One more thing, since there’s been a lot of appeal to the guidelines in this thread. It’s always to the same line, which appears in Keep It Tidy, in a list with:
Imagine the same complaint about any of its neighbours. Someone finds signatures unbearable, calls for automated deletion of any post with a signature as a matter of urgency, threatens to stay away if the trend continues. Same document, same section, same remedy. Is that believable?
The guidelines open in a completely different place, and a place I’d suggest has been quite neglected in this thread:
It says of everything that follows:
If we’re going to hold each other to this document, I’d rather we started at the top of it than the bottom.
You are not, and neither am I. stevengj is one of seven stewards with the authority to resolve conflicts regarding the core project’s community standards, which is broader than this specific forum. The rest of us can argue about principles and how things ought to be, but this isn’t a direct democracy.
As for the arguments, many reputable publications ban AI-generated submissions, and all of your examples involving collaboration among people only bolsters the sort of forum LeePhillips prefers and the current guidelines. You might have meant that chatbots are effective substitutes for people and therefore those examples apply to your argument, but obviously many don’t see it that way.
We don’t have to imagine, it already happens. Besides moderators leaving very temporary evidence of deletions, the AI generation issue comes up a lot more because LLMs write a lot very quickly, and those walls of texts LeePhillips complained about are far more noticeable. Let’s be real, some people are already posting mostly generated responses, and moderators have clearly been letting some of it slide (“bygones be bygones”). The guideline against AI-generated posts only exists because AI users have generally gotten so bad at producing concise and sincere thoughts in the last few years that some line had to be drawn to curb spam.
