Updates to site guidelines and AI/LLM tags
308 points by 355E3B
308 points by 355E3B
Hey folks,
I've been having to perform an increasing number of moderation actions surrounding the vibecoding tag and AI/LLM related content. I started reading Lobsters because it had the most interesting viewpoints related to computing. During that time, I've found that this community is at it's best when it's trying to, in good faith, help others grow. In lieu of a larger and more rambling discussion on this point, we would like to try two changes encourage that behavior again and re-focus topicality around ensuring that content submitted to the site should, at a minimum, be respectful of a reader's time and intelligence.
The first adjustment is two updates to the about page, which try to provide a bit more clarity in what we consider topical and what constitutes self promotion:
Intro:
Lobsters is a computing-focused community centered around discussing links. Our goal is to learn, craft, collaborate, and share interesting viewpoints.
The administrator is Peter Bhat Harkins ("pushcx"), contact him or another moderator with any support issues. Lobsters was launched on July 3rd, 2012 by joshua stein (who is no longer running it, so please quit bugging him with site stuff) with careful design touches to encourage a healthy community:
At this risk of writing a mission statement, it seemed worth explaining what the site is and hopes to be to visitors from the first sentence of /about. (This original text came from an intro post.)
Then the Guidelines:
Topicality: Lobsters is focused pretty narrowly on computing; tags like art don't imply every piece of art is on-topic. Some rules of thumb for great stories to submit: Will this improve the reader's next program? Will it deepen their understanding of their last program? Will it be more interesting in five or ten years? Is this compelling material that will seed a discussion, or is it just an artifact?
Pending an upcoming deploy, we no longer allow submissions which are code repositories. While, at first blush, this may seem odd for a community which is primarily centered around programming but it does eliminate the bulk of the "low effort, self promotional" submissions we've been seeing on the site. We've also found that being dropped into a new codebase doesn't start discussions because the reader has to have enough subject matter experience to immediately recognize what's novel or otherwise worth talking about. Instead, we are more interested in content around code bases which cover why it was created, the purpose it serves, the challenges overcome, and the reasons why it's a useful artifact in the ecosystem.
Some things that are off-topic here but popular on larger, similar sites: entrepreneurship; management; careers; news about companies that employ a lot of programmers; investing; world events; anthropology; self-help; personal productivity systems; last-resort customer service requests via public shaming; "I wanted to see what this site's amazing users think about this off-topic thing"; and defining the single morally correct economic and political system for the entire world when we can't even settle tabs vs. spaces.
We're listing "careers" as off-topic to try to head off posts about interviewing, promotions, burnout, and farming. We have the job tag because it's nice to help people connect to jobs, but the topic of getting a good job, weaving jobs into a career over your lifetime, and work is a different thing than the actual programming itself.
Self-promotion: It's great to have authors participate in the community, but we do not want the community to be exploited as a write-only tool for product announcements or driving traffic to their work. As a rule of thumb, self-promotional activity should be less than a quarter of one's stories and comments in the last year. New users (first 70 days) are not allowed to submit self-promo.
This is how we've been enforcing the rule for a while, better to be explicit. It's unfortunately common that authors see their work posted, join the site, but then only use us for traffic. Hopefully this time helps them understand the difference between a symbiotic and parasitic relationship.
The second adjustment is that the vibecoding tag is now marked as inactive. It still appears on old stories but can't be applied to new ones. From my perspective, the tag was overloaded with six differing responsibilities:
First, it was the tag which was trying to capture the discussions related to the use and implementation of AI/LLM tools.
Second, it was a marker for content which is partially related to AI/LLM tools, tooling or companies.
Third, it was a marker for content which includes a discussion on or output from AI/LLM tools in a part of a larger work.
Fourth, it was a marker for content which was generated using AI/LLM tools.
Fifth, it was being used in competition to the spam flag, for really low-quality content which was generated via AI/LLM tools but without apparent significant human authorship.
Sixth, it was being used as a tag of shame by people who really dislike AI/LLM tools and used it to insult people at every passing mention.
The last couple years have been the biggest upheaval in the field for decades. While it's acceptable, within limits, to express frustration towards circumstance or technology, it's not acceptable to direct that behavior towards individuals whether they be members of the community or not. Other people can't be your punching bags and Lobsters can't be your boxing ring. That's not acceptable.
While `vibecoding` is no longer an active tag, we still feel that a subset of the tooling related discussions are relevant to the site. While this is reaching far back into the annals of computing history, tools like compilers, databases, testing, and vcs had similar points of consternation and concern when they were introduced. I'm not going to claim that I'm able to guess how the industry is going to shape up, but my gut feeling is that but my gut feeling is that it's going to be closer to something like we've seen with systemd than compilers. This new tag `ai-tooling` covers "The design, ux, development of AI/LLM-powered tools themselves".
As a fun addendum, there's a variant of Maslow’s Hierarchy of Needs which exist for websites, we kinda fell on the wrong side of that trying to migrate the site and get this out on a Saturday. Whoops!
Thanks, 355E3B
Fourth, it was a marker for content which was generated using AI/LLM tools.
Fifth, it was being used in competition to the spam flag, for really low-quality content which was generated via AI/LLM tools but without apparent significant human authorship.
I'd like to (continue to) voice my unhappiness with the current site policy around this.
I want a "Slop" flag. Lobste.rs is absolutely swarmed with slopped-out blog posts, and the moderation team seems incapable of recognizing or removing them. In lieu of moderators who can recognize slop prose, I want a community-facilitated moderation (or even just tagging) system.
To continue on the same line of discussion, the definition of spam was updated to read:
Spam which should be used for content that either is designed to promote a commercial service or for content that is created without meaningful human authorship.
It's not perfect but that is the flag for marking low effort/ai generated content.
While I understand the desire to have a "slop" flag, the "tag of shame" connotation and associated behaviors around the vibecoding tag was causing moderation problems.
OP named 6 ways how people were using the overloaded vibecoding tag: one was given a new alternative tag (the positively-connotated ai-tooling), one was deemed invalid (tag of shame), and one was given a non-tag alternative (flag as spam). What is the recourse for people that were using vibecoding for the other 3 reasons? Is it just now website policy that these are not worthy characteristics of being tagged, despite being quite widely used?
This is effectively choosing a side in the bifurcation of the website, and I think it's important to elucidate to the community if this was the intention of the tag changes and indicates the direction of the website going forward, or if this was an attempt at a neutral solution.
Here's how I would map them based on my understanding of the OP. Two of the six reasons map to the new AI-tag.
First, it was the tag which was trying to capture the discussions related to the use and implementation of AI/LLM tools.
ai-tooling tag
Second, it was a marker for content which is partially related to AI/LLM tools, tooling or companies.
this doesn't cleanly map to anything. invalid, add the new tag, or flag. what is "partially related?" industry news have already been off-topic.
Third, it was a marker for content which includes a discussion on or output from AI/LLM tools in a part of a larger work.
invalid
Fourth, it was a marker for content which was generated using AI/LLM tools.
invalid
Fifth, it was being used in competition to the spam flag, for really low-quality content which was generated via AI/LLM tools but without apparent significant human authorship.
spam flag
Sixth, it was being used as a tag of shame by people who really dislike AI/LLM tools and used it to insult people at every passing mention.
invalid
While I understand the desire to have a "slop" flag, the "tag of shame" connotation and associated behaviors around the vibecoding tag was causing moderation problems.
The shame is a feature, not a bug. This technology is not neutral by any means. Its very use treats vast swathes of labour as punching bags. Comparing its introduction to the introduction of compilers, databases, and version control systems seems disingenuous at best.
Concealing slop behind the spam tag makes it easier for slop to go unmarked. Introducing a new tag explicitly for promotion of the technology under the cover of ai-tooling is similarly disappointing.
I don't expect the moderation team to change tack, but I am disappointed that this is the direction it has chosen.
Shaming is toxic. I wouldn’t attend a garden party that habitually shames people because that’s a community dynamic I am not interested in, regardless of the justification for the shaming. In the case of machine generated posts, it is much healthier to not even allow them on Lobsters and that’s what the spam flag is for.
Though ‘shaming’ is toxic, I have no issues participating in a garden party where some ideology is not allowed. I personally do not believe modern internet spaces need to accomodate everybody and every desire.
I am not afraid to say I seek a community of like-minded programmers that are mildly to completely against LLMs. I do not really care that they are here to stay. I can sing their praise anywhere else, let me have a space where they are NOT accepted.
Cool, you're allowed to seek that. I hope that that isn't what this community ends up being after having enjoyed my time here for the last 5 years.
I am not afraid to say I seek a community of like-minded programmers that are mildly to completely against LLMs.
Please share with me if you ever find one and I will happily join. It has become impossible (in my experience) to find any such safe haven and would gladly welcome the same company you seek. I thought Lobste.rs was it when I first jumped ship from HN as it became a rotting cesspool of AI/LLM worship but it seems as though readers mods here have the same problem of not being able to distiguish between the desire for clean work, ideas and discussion and the delusions of GenAI culture.
Other commenters want to call shaming toxic, but I cannot think of anything more toxic than allowing a magically stochastic blunt object to beat the humanity out of our labors, it's champions demanding that we hand our thinking over to it, asking us to give up the very thing that drives our curiosity, innovation and agency. If that is the ethos the Lobste.rs community wants to adopt, perhaps I do not belong here.
I share your frustrations, but if we are to believe in our ideals, we can't invade spaces that don't belong to us and force them to contort. The admins of this website believe that conversations about technology can and should remain reasonably neutral.
I also think that's quite naive, and also unjust in the grander scheme of things, but nonetheless that doesn't give me any right to try to shape this community to my whims. No one really wins if two factions keep trying to out-argue each other by strength of numbers. I'll also be keeping my eye out for a more sympathetic community where I can be more open with my feelings, as I need somewhere and all my spaces have been taken away from me, but the admins are within their rights to say that it's not here.
Who invaded whose space? It seems, based on how discussions here usually go (hence all the anger about the vibecoding tag) and just from the comment upvote counts that people who don't want to discuss LLMs are the majority here. I also thought this was somewhat of a safe haven when I became active again earlier this year and have been disappointed that it's been trending towards LLMs sucking out all the air like they do everywhere else.
I wonder if there's a credible way to tell if that's a majority or a loud minority.
Even if such a way existed, I'm not sure if you could extract any useful information from that without doing more demographic analysis. It feels important to know how this changed over time, e.g. whether we pushed some pro/anti-AI people out even though they would like to remain here, or whether any astroturfing is taking place. You'd also need to assign value to different groups -- do you prioritize readers, commenters, or writers? So I'm worried that if any objective statistics are brought up, the inferences made from them will be very subjective.
This whole story reminds me of Mozilla with their "let's copy the anti-features of our competitor to maybe hopefully steal their users [and alienate the existing ones]" strategy. So even if it turned out that almost everyone wants to see LLM content, I just don't think it'd work out well for Lobsters, because at that point it'd just be HN with less fascism, and I think that differentiating yourself in some other way would be valuable.
Going by the upvotes, the people who dislike LLMs and want them out of Lobsters are a very clear majority.
Upvotes are anonymous, and comments regularly get an order of magnitude more votes than there are users in a comment thread. That seems like a fair poll of the spread of opinions of Lobsters' users, the commenters and the silent participants together.
If you look only at posts where there are pro- and anti-LLM comments available to vote on, the anti-LLM comment is always at the top, with at least three times as many upvotes. And that's in the vibecoding tag, which many anti-LLM users have muted.
I analyzed 90 days worth of lobsters content with Opus and Jev and it would appear it's a loud majority: https://radius.earendil.com/artifact/01m59h7w4ke3j89g3a0yzd33d8 (slop warning)
Of the 1,054 people who expressed a view, 62% leaned negative, 25% positive and 13% were balanced.
Considering there were 2,509 active commenters over that 90 day period then yeah, that's a very significant portion of the site actively expressing negative opinions (0.62 * 1054 = 654 out of 2509, 654/2509*100 = 26% of active commenters.)
Given those numbers, I now think a better fix here might be to rename "vibecoding" to "llms" so people can continue to filter out that tag (while fixing the confusion around the tag name.)
As someone who is cautiously pro AI, I just don't bother engaging with the debate around it on Lobsters most of the time. I imagine that means my opinion isn't counted?
This only measures comments, yes. And only comments that were attributed as being about AI.
To decide whether it's a loud majority or minority, it would have to be put in relation to the total amount of (active) users.
Two points on this post:
I vibecoded something" see /s/lkngrz as the message in the moderation log.First, if this were a story, I would be heavily included to remove it with I vibecoded something" see /s/lkngrz as the message in the moderation log
I don't know what this means.
Second, I'm not entirely sure that the analysis drawn from the data set is accurate? A large portion of the interaction that community members have come in the form of votes, flags and tagging suggestions.
I can only look at what is public data obviously. But I did have some discussions with the agent about what I could expect if I had vote and flag data, and it does not think it would change the results much.
But I did have some discussions with the agent about what I could expect if I had vote and flag data, and it does not think it would change the results much.
I was with you right up until here. How would an LLM know whether having private data would change the result? Why would you trust or relay its "opinion" on that?
I was with you right up until here. How would an LLM know whether having private data would change the result?
It's not so much about if the LLM has any private data, but I had it create various different scenarios of different voting patterns (based on public information we do have) an see if that would change the findings.
//EDIT: if you want you can read the reasoning here: https://radius.earendil.com/artifact/01m5ajp6hrf6wtrs40gbfy6x4v
Interesting analysis. I am not sure though that processing classifying all comments using LLMs will be well received.
What's missing in my opinion is, why if a majority is against AI are so many stories submitted about/tagged vibecoding? Are the submitters the same people arguing?
What's missing in my opinion is, why if a majority is against AI are so many stories submitted about/tagged vibecoding? Are the submitters the same people arguing?
About a quarter of the people "with a view" submit AI stories. The anti camp in the last 90 days submitted 191 AI stories at the time of snapshot, the pro camp 152. Interestingly around 40% of AI stories are not from people who participate in the discussion at all.
Also, a ton of these tags were added post-hoc. Not sure how many, but clearly some stories were originally submitted not to be AI stories, but ended up starting a discussion about AI regardless.
Thanks.
Yes, many stories were submitted without vibecoding but someone spotted an AGENTS.md or similar hint of LLM involvement. Makes sense that the submitters don't necessarily participate in that discussion.
Reasonably fascinating.
One thing I don't know from this is whether there's a point where neutral conversation on AI topics, or neutral responses to pro AI comments are included as being pro-AI (same for responses to anti-AI comments). To me these sort of comments on an article tacitly accept the idea that this is what we want here and these are the people or conversations that we want in this garden even without being positive or negative.
Two sets of comments were classified: all 7,627 surviving comments under vibecoding stories, and 1,516 comments elsewhere on the site that mention AI by name (LLM, Claude, Copilot, OpenAI, vibe coding and similar terms). About half of the first set, 3,619 comments, and 912 of the second turned out to express a view. Most of the rest were ordinary technical discussion that happened to sit under an AI story.
So given only half the comments expressed a view, the rest seem like they could be treated as some sort of broad quiet majority right? This is probably one of those difficult to measure things.
Invade is too strong a term, you're right, I'm letting rhetoric get the better of me.
I mean that the space belongs to the administrators of the servers and maintainers of the software that runs on them, it is for them to decide. Unless they setup an explicitly democratic institution, if they wish the space to act in a certain way, and can put in the effort to make it so, that is how it shall be.
doesn't give me any right to try to shape this community to my whims
We are (were?) part of this community. That gives us the right to try to shape it how we think keeps it a community. GenAI is community-destroying, antisocial technology.
Shaming is toxic. I wouldn’t attend a garden party that habitually shames people because that’s a community dynamic I am not interested in, regardless of the justification for the shaming.
Shame is one of the few effective alternatives to brute force that our species has developed over time for group discipline. It is toxic. It works. It is a proportionally mild response to a technology as toxic as generative AI.
Shaming people on Lobsters will not make the technology go away. But it does fill your mind with hatred which has negative effects on you and makes other Lobsters users’ lives worse too.
Nobody on Lobsters realistically expects the technology to "go away". There are just people who do not want Lobsters to be part of the same death knell as the rest of the internet.
I don't think endless in-fighting, purity checks and scolding each other to ensure no one touches LLMs would help lobsters avoid a "death knell" either, so perhaps we need a secret third option.
Ideally, it would be marking all AI discussion as off-topic, both as posts and as bickering over it. There are a ton of places where you can talk about AI, but not so many where you can talk about programming and technology, without having it drowned out by AI.
It may make the people posting it on Lobsters leave Lobsters, though.
Even if we assume for a second that the goal of removing LLM users from the site is a reasonable one (which I don’t think it is at all), that would need to be a site rule. Bullying is not a replacement for moderation.
But tagging a post as "on a broad topic i don't want to see" is not bullying. The fact that some people take it as such, and leave, is enturely those people's problem.
Shaming people from a community temperature perspective is if anything very ineffective. When it's (indirectly) encouraged by a community's staff you're going to get a lot of problems, which I can say from experience.
Basically it was a community on Discord that by its nature (technical support) accrued a lot of "quiet" bigots. At the same time, the community itself has explicit rules against bigotry. Staff also would set a pride flag each year as a support for LGBTQ members, which were a good chunk of regulars and staff.
Predicably, every year a chunk of those bigots would crawl out of the woodworks and show their ass in public channels. Staff would ban them for obviously breaking the rules. Eventually, due to staff not being able to constantly monitor everything, you'd eventually get some bigots to stay around for long enough until a mod comes online to boot them.
The server regulars would typically mock these bigots in the meantime, encouraging and arguing with them in the hopes this caused them to say something worse. Eventually it became staff policy to encourage this behavior; basically if a staff member was online, saw this was happening, policy was to just sit back a bit and "let the server regulars have their fun" (or to even join in themselves.) This seems reasonable at first; the bigots need to be booted anyway, what's wrong with some catharsis?
The problems however accrued slowly. The first, and most immediate problem, is that the choice to set a pride icon went from "show support to LGBTQ members" to "let's see how many people we can ban today for being bigots". The second problem is that it began to encourage bullying of members; even if someone wasn't being a bigot, if you could just catch them into saying something out of context, you could call staff sanctioned harassment on them. Thirdly, a lot of LGBTQ members just... lost interest in the server; it was mostly non-LGBTQ members after a while who got a kick out of bullying bigots. It drove a wedge between the people the adjustment of the rules was supposed to be nice to, simply because people didn't want to deal with this kind of bullying all day, even if they definitionally wouldn't be the subject of that. The server got extremely toxic in a way which rules couldn't easily express. (It also had deeper seated issues with moderators giving warnings for things they'd themselves proceed to break the rules over less than a week later, to the point where the "best" way to appeal a warning was just to catch a mod breaking the same rule.) Ultimately, the best way to come down on this would just have been to cleanup the bigots whenever they showed up, and inform users to not dogpile and to just report instead. (I wasn't staff in this community to be clear, but was/am still friends with a bunch of former staff in this server, and most will readily admit the server was a mistake, and in the times I've moderated communities myself since, rules forbidding dogpiling largely originate from trying to prevent this sort of thing.)
At first, public shaming may feel good. In practice, shaming only works in the absence of actual moderation. To bring it back to lobsters and the (hopefully) less serious problem of LLMs; moderation taking any position is preferable to the previous situation, where the vibecoding tag got quietly overloaded (as it's actual use vs its description don't match up) to shame people, quietly boiling up the temperature as it brings mostly the same kind of comments out of the woodwork. Even as someone who really doesn't want to bother with reading LLM slop if I can avoid it, it just got really tiresome. Staff taking the position to eliminate the overloading of the tag is preferable to doing nothing and then looking around months later and wondering why everyone despises each other.
This is a thoughtful post and matches my experience. There is something truly insidious about dog-piling, and in the long term it becomes a case of "the cure is worse than the disease" for online cultures that support it. I don't say this from a moral high horse of not getting the appeal of the dog-pile when it suits me. But lots of things that feel good in the moment aren't good for you.
Shame isn't actually that good at discouraging behaviors, instead often encouraging people to hide the behaviour better.
I agree with the issues around a "tag of shame", but I wonder if there might be a way to reframe this as something more like a "content warning" to make this healthier for the community.
I imagine a system where users can suggest warning tags that are not used to categorize posts, and do not show visibly on posts by default, but that users can set filters for so they can choose to avoid content they do not want or are not safe to engage with.
This keeps the content categories strong while allowing finer resolution for "I never want to see a mention of REDACTED" again. Basically lets users crowdsource the "open it, see a mention of REDACTED, close it" process.
This mechanically is very similar to the dreaded "tag of shame", but instead of publicly tarring and feathering the post, users without the filter can't see the tag without going digging, and users with it filtered shouldn't even see the post. Huge change in the friction and discomfort of the experience, while still letting people opt out.
I imagine a system where users can suggest warning tags that are not used to categorize posts, and do not show visibly on posts by default, but that users can set filters for so they can choose to avoid content they do not want or are not safe to engage with.
This is roughly similar to the ATProto labels concept and the entire work around stackable moderation. I do think it's an interesting concept and one that I would like to explore for the site, but it would be a pretty large codebase lift that would need to happen.
So fair. So valid. I meant to put a comment or "but not to suggest a bunch of work for you", and apparently forgot!
Perhaps a browser extension could be developed by users interested in prototyping this. In the heyday of tumblr, it was common to use "Tumblr Savior" which was basically an external tag-filtering system that didn't post anything visible to people not using it.
Perhaps a browser extension could be developed by users interested in prototyping this. In the heyday of tumblr, it was common to use "Tumblr Savior" which was basically an external tag-filtering system that didn't post anything visible to people not using it.
Consider this the final push I needed. I'll work on such a thing as I have time.
FYI, https://shinigami-eyes.github.io/ has somewhat related mechanism, perhaps ready for reappropriating.
I viscerally hate reading AI prose: it’s disrespectful to the audience (when undisclosed up-front) and makes me feel like my sense of taste over writing style is being polluted by homogenized AI-speak. The actual content of the article is almost irrelevant.
I’d like a way to know whether an article is human-written before diving into it, and if I can’t do that, I will have to call it out in the comments. I don’t think this will make the community feel less toxic.
Per https://lobste.rs/c/lmt2ys, you can now use the Spam flag for low effort/AI-generated content.
355E3B says that user behaviour around having a 'tag of shame' for 'this is slop' was causing moderation problems. I can completely see how it's tiring to to play judge in variously-heated "don't post this shit / you can't prove it's shit / this shit is useful" discussions, and sincerely hope that the mod team finds it less stressful to triage spam-flagged stories.
Either way should work to keep the slop off Lobsters.
Unfortunately, not all LLM-generated prose is spam or low-effort. And I’d argue that a tag of shame (in which the “shame” aspect is very subjective) is much better than actual angry comments.
From my observations, we had plenty of angry comments and name-calling even with the vibecoding tag. Perhaps I am naive to think it can't or won't get worse.
There is a way to address that problem directly: stricter moderation of comments.
I don't filter any tags, and my only complaint about the Spam flag is the lack of adoption. But the discussions are not good for me. I don't want to surround myself with folks who get angry instead of getting curious.
With the removal of the vibecoding tag, it is now necessary to flag such poats as Spam, even if you wouldn't otherwise.
I empathize with this, and also really dislike AI-written prose most of the time, but I suspect that actually achieving this is impossible.
Language is infectious, and the people who spend a lot of time digesting LLM writing will naturally tend to write like that as well. After all, these are statistical models of human writing we're talking about; it wouldn't be a property of their output if people didn't actually write like that.
This leads to people accusing anything with a whiff of LLM as being slop, and frequently it has felt like an overreaction to me. The knee-jerk reaction is often worse than the thing they're reacting against.
In my experience the vibecoding tag did not have the effect of removing AI-written articles from the front page. It marked some of them but it wasn't even most of them.
it seems the moderators have to explain that use of the "spam" flag at every week or so. this is a ui problem, so i opened https://github.com/lobsters/lobsters/pull/2260 to remedy this.
One problem is that people disagree on the definition of "slop". To some it's "low-quality generated content with little human involvement", to others it's "any content touched by AI". If we get that kind of tag, even a minority of people using the second definition can community-tag half the submissions, and we're back to the useless vibecoding tag again.
Low-quality content (generated or not) should be removed, not tagged. Hopefully the new "no code repositories" rule will weed out a lot of stuff (I've seen it work in other programming communities), and allow our moderators to spend more time on the contentious cases. In the meantime, a low vote count on a submission can be used as a red flag.
nd adjustment is that the vibecoding tag is now marked as inactive. It still appears on old stories but can't be applied to new ones. From my perspective, the tag was overloaded with six differing responsibilities:
No: Slop is the automatically generated output of chatbots presented as if the "author" did the work themselves. Chatbot "created" content is built entirely upon stealing the work of others, and in the case of "vibe coding" is purely theft and laundering of open source licensed code. This last point is a major reason for corporate support of chatbot code: for decades companies have attempted to steal OSS code, and now they vibecoders have given them an "acceptable" way to do so.
As much as I may agree with your sentiment, if you insist on defining "slop" that broadly while other people have a narrower definition, you just make it very hard to communicate with people you should be trying to convince. If you want to efficiently push against GenAI, you need to meet users where they're at, and keep in touch with modern practices, in particular the ones that avoid producing slop.
This is fair. I don't particularly care about the particular name, but perhaps ai-prose would be a less ambiguous tag/flag name. (Although I and everyone I know consistently uses the latter definition...)
To me, ai-prose describes a text mainly generated by AI, which seems to fit the updated definition of spam and should be moderated away. It misses cases of partial LLM use, and output types like images, code, math proofs, debug assists, etc, If you want an all-encompassing but neutral tag, maybe ai-used would work ?
Isn't that flag the existing "spam"?
Not really, IMO. LLMs can output informative slop. But maybe we should rather flag that as "melodrama writing style" (not sure about the name!) instead of "slop": it’s basically impossible to prove something is LLM-generated, but we can totally recognize that stupid writing style which does not bring any value from a technical standpoint.
I think the rules should simply be: no llm generated posts or articles. Just provide the prompt, that way people don't have to read the slop, but rather the actual ideas from the author.
One thing I am perpetually confused by is that people propose this as a “solution”.
I agree with you about not wanting to read fully LLM-generated posts, but “just provide the prompt” is not at all how I have ever successfully used an LLM. There is always a ton of iteration, plus the non-determinism, plus the context that isn’t captured in the prompt.
Not a post per se, but a really straightforward example: “Give me a summary of what I’ve worked on in the past two weeks based on my Obsidian Daily Notes. What loose ends are there that should get wrapped up?” is a prompt I could type into a harness and get a useful result out of. Me telling you that that’s the prompt I used will not generate anywhere near that same output.
Similarly if I were iterating on a blog post and shared the entire sequence of prompts, that still wouldn’t help. If one of those prompts in the session was “ok, I like the section layout now but figure 3 is hard to understand. Let’s rework that so that it clearly identifies each component and the data types going into and out of them”, that still doesn’t help because you using my initial prompt may not have even given you a document with a figure 3, and if it did it might not look anything like what mine did.
I've worked around this issue a bit by publishing the initial prompt and then linking to full transcripts, which can be a bit of a pain to export. I don't know if anyone ever clicks through to see them though.
I think a "slop" flag would only invite negative discourse. IMHO its a downvote button. So perhaps we should upvote and value the non-slop more aggressively.
AI-generated prose is not necessarily slop.
I still want to know though, because most of the time I won't want to read it.
I can accept AI-generated documentation that's been reviewed as correct by a human, but that's not the kind of content I'm looking for on a site like Lobsters. I want that to exist so I can find it when I need it.
If text is trying to convince me of something or explain the rationale behind a project I need it to be majority human written - AI as a spellchecker/grammar/translator is OK, but the moment I see LLM text trying to be convincing or add flavor (it's not X, it's Y!) I feel like my time is being actively wasted.
It may be a description of novel, human-imagined concepts, e.g. Wouters' genuinely mind-blowing new language Goose, which you probably haven’t heard of because the post here was immediately downvoted as spam.
It may be generated documentation of human written code where the author hasn’t had the time to write the docs themselves.
On a link aggregator, I want to see things that humans wrote. The project could be interesting, but slopped out prose doesn't it make it more interesting. It just means more noise for me to dig through.
may be the cleaned-up prose of someone with poor grammar/writing skills and/or dyslexia
Dyslexia can mean many things and have many causes. I'm not sure how much an LLM can truly "help." In any case poor grammar and poor writing skills can only be improved through practice. Letting an AI slop machine slop out prose for you harms both the writer and the reader. I have no problem if you write a clumsy sentence, I do that all day. If it takes you 10,000 clumsy sentences to write one beautiful one you've still got a better quality ratio than any LLM on the market.
It may be translated from the author's native language.
I think it's bad manners to machine translate something for me. Translate it yourself, Nabokov style, or have someone else translate it, or let me machine translate it myself.
I think Goose language would have been better received if it was introduced by a blog post explaining the language and design rationale, instead of linking the repo directly. I had a look at the docs and found them hard to read while grand claims are made about the programming language.
We fundamentally disagree. I do not ever want to read AI-generated prose. It is a cognitohazard.
When reading the post this was my first reaction. But slop probably has the same capacity/vulnerability of being used as a "tag of shame" even in scenarios where the post isn't slop but used an LLM.
I find it unfortunate that this update portraits people who prefer not to interact with LLM-related content as assholes who want to shame others. This is not a neutral technology; I have to deal with enough consequences from LLM usage in real life that it's just not fair to sum up my reasons for avoiding LLM-related content as just shaming. I can both berate LLM usage and respect its users, but I expect some basic respect in turn, like not subjecting me to LLM fatigue and then treating me as irrational when that affects me.
I have enjoyed Lobsters for a long while because it enabled me to talk and read about craft without constantly having to see articles about this catastrophe-in-progress. It's disheartening to start reading an article that promised a human connection, only to see they asked Claude instead of doing research and experimenting. That wouldn't be a problem if only this didn't happen in like half of the technical posts.
I already essentially abandoned HN for this reason, and I only enjoyed Lobsters because I could skip content that is uninteresting to me if it was marked with vibecoding; and if the title still sounded interesting enough, I could prepare myself emotionally. As others have pointed out, vibecoding worked well as a trigger warning, so adding the tag was erring on the safe side rather than overuse. I could always disable the filter if I was comfortable sifting through more content.
If the tag is going to be removed and no alternative exists, I'm not sure how I'm supposed to keep using Lobsters. While I respect that moderating is hard, it rubs me the wrong way that the only representation I get is "stop making people punching bags". I think the writing has been on the wall for the past two months or so, but losing a community still hurts :(
I think it's also necessary to consider where Lobsters stays in relation to other programming forums. HN and Reddit have been increasingly pro-AI, so if you're seeing more opposition to AI usage here, I think it's not because "AI haters" got louder, but because many of them migrated here from other places after being pushed out. There is a niche for LLM-free content and discussions, and I believe that Lobsters fit this niche. Explicitly marking LLM-related submissions as on-topic and not requiring annotation will cause Lobsters to lose these users, and while I respect that you don't have to cater to everyone, I don't think this alienation is healthy in general or will lead to a good outcome for Lobsters as a community.
I disagree with this premise entirely, lobste.rs has never been a place free from any technology even during the crypto bro nightmare hype cycle, just because you personally don't like a technology doesn't mean it's been an unwelcome topic of for meaningful conversation.
I ignored the rust tag when I was tired of the rust hype train too... what I am seeing now is wholly different, people have been tagging things with vibecoding when it hits their personal threshold in a vindictive way to signal this very vocal group of anti-LLM users to dogpile on, when it is objectively not a low quality article or even related to LLMs in many cases. A (assumed) minority of users is holding the rest of us hostage with their perceived moral superiority and if it continues we will lose our community.
This is astounding to me as the parent posts are describing a dog piling of pro-LLM users. I'm sorry that the advocates can't elevate themselves rhetorically beyond complaining about their criticizers but personal attacks like your post are the reason for this thread.
I can both berate LLM usage and respect its users
What the moderation team is saying, and the AI vegetarians are missing, is that berating LLM usage on a computing forum is becoming unmanageable.
This could become a bastion for the holdouts, but that would come at the cost of constructive discussion around a lot of developments in computing.
I think more importantly, in an environment where LLMs are becoming ubiquitous, we need to be able to discuss practices around them. I applaud the people who have switched and donated to the BSDs for standing true on copyright and LLM contributions. I'm happy for those who refuse to interact with slop. In computing today, that is a privilege.
I have old friends from graduate school, in corporate, who have managers forbidding typing code by hand. Internal "token use" metrics are rampant. We're veering back to LoC as a measure of productivity, for goodness' sake. Telling people to quit their jobs or leave teams isn't a meaningful avenue for many people whose livelihoods depend on programming.
we need to be able to discuss practices around them
These topics are discussed to death all over the internet. Even if you do "need" to discuss them, you don't need to do it here. But it's exhausting for many of us that this is the only topic people want to discuss anymore.
I haven’t found a serious place to discuss computery aspects related to research, engineering, or using agents for coding.
HN has too much shilling and uncritical acceptance of ideas, and the few Discord channels I’ve come across are self promotional, and thus inappropriate for discourse. I don’t use social media much so I am likely unaware.
I like the invite-only aspect of lobsters because I think that helps retain its content quality standards. If there’s anything like that, lmk.
For reference, I think it’s important for different spaces to exist, and if lobsters wants to be the non-AI/LLM software and computer engineering place, then what’s wrong with that? (Provided it’s not at the expense of misbegotten ideas that all agentic coding projects are slop, or that people don’t produce slop all by themselves). I admire the mods for trying to find a reasonable compromise, and I don’t envy the position they’re in for having to deal with it.
The community could try out the ai-tooling tag for a while, and see how it works shrug
I don't tell anyone to avoid using LLMs in their job if they're in a situation where that affects their life conditions. I don't go around commenting about our lord and savior manual coding on every vibecoder's post either. There needs to be a place for topics like that, and I'd rather that place was politically closer to Lobsters than HN.
What I'm asking for is to also have a space where I can isolate myself from those discussions and talk about other topics. I don't want to see anything related to vibecoding, much less comment on it. Lobsters with a filter on vibecoding used to be that place, and I valued this site a lot for that feature, and I don't want to lose it.
I have read many of your posts on optimization and Rust, and have always appreciated them. I value your contributions to this site, and just computing generally.
I wonder if the solution here is a community extension/plugin, like Return Youtube Dislike, "Return Lobsters Vibecoding"? It's an afternoon of work, and could alllow users to even implement more advanced filtering (users, keywords, Pangram API calls/tagging, domains, etc).
Given the number of upvotes to your original post, I think you might not be alone in wanting this.
I worry that the direction computing is headed in generally might ultimately alienate most people, but in the meantime, this could answer the needs of both moderators and AI vegetarians.
Thanks, I appreciate that.
I think making an extension makes sense as a plan B, but if it's not officially supported, it's bound to be unpopular and best-effort. It also means that people who submit LLM-related stories, who most likely don't have qualms about AI-related content and thus won't have the extension installed, will not mark their content, so only community tagging is left.
Not to mention all the technical difficulties, like copying/adjusting the Lobsters community tag system and filtering logged in users to workaround bad actors, etc.
I believe first-class support for trigger warnings is the only real solution.
I would still use such an extension if it existed and worked well enough, but I don't personally believe in it enough to create one myself.
Fine, that can be an option too.
Opt-in community custom tags (i.e., vibecoding, created by purplesyringa, 300 members, auto-hides applies after 10 votes).
This should satisfy the requirements of everyone. Also opens the door to all sorts of cool positive curation, i.e. exposing your groups tagged posts as a feed.
EDIT: This appears to be equivalent to ATProto "Stackable fliters", which apparently are a decent technical lift. So the options are DIY or wait for that to be prioritized.
It also means that people who submit LLM-related stories, who most likely don't have qualms about AI-related content and thus won't have the extension installed, will not mark their content, so only community tagging is left.
Presumably, that is the point (though I still agree that first-class support for community tagging would be more valuable than a third-party extension, it would still be community tagging).
Those who dislike LLM-related (or LLM-adjacent, to an arbitrary degree of adjacency) content get to do their own tagging, with the threshold they themselves consider proper. Those who have no qualms about LLM-related content likely don't even know what constitutes the current anti-LLM acceptance threshold, so why should the onus be on them to do the marking?
But you could discuss them here, and people did every day. The "vibecoding" tag allowed for people not interested in those posts to skip past them.
But you could discuss them here, and people did every day. The "vibecoding" tag allowed for people not interested in those posts to skip past them.
One person deactivated their account due to an article they submitted being tagged vibecoding, and that was rightly considered a problem worth addressing.
Multiple people have already deactivated as a result of these tag changes and made their issues with them clear.
It feels like we don't really have a good outlet to give feedback on these changes and that feedback in this thread is unlikely to lead to action, so I'm sure there are going to be more departures. It really is a shame.
Do you sincerely believe that @FiloSottile is the only person who has deactivated their account or stopped participating in the environment as a result of the current negative atmosphere towards anything even tangentially related to LLM-usage?
No, nor did I say they were. In both cases we're losing valuable contributors due to things that can be changed and it's worth addressing. I don't think the changes that were implemented are going to make things better for either side, except that some people won't have their feelings hurt by the word vibecoding. The negative atmosphere didn't go anywhere.
There is an impossibility here of catering to those who open a link, see any amount of LLM-generated content, close the tab, and flag the post as spam, and those who don't mind it/are excited about the results regardless. I'm hopeful this will cut down on much of the noise (esp. slopped together projects w/o explanation), but I suspect that posts detailing work only accomplished as a result of LLM use will now slip through, and I would definitely appreciate a tag for this as well. I do think LLM usage is shameful practice, and I do not want to engage with it, and I'm far from alone in that, and I'm certainly not leading crowds with pitchforks to those websites. I still believe there should be some such tag such that we can filter away such posts if we don't want to engage with them.
It's unfortunately been pretty difficult to try and keep the properties of the tag without also bringing along the negative externalities it's existence has caused. I've thrown around the idea of adopting the ATproto "stackable moderation", which would offer a better way to apply these overlay filtering long term, but that's gated behind some serious technical work and a few UX issues.
With regards to posts detailing the work which was only accomplished as a result of LLM usage, my suspicion is that the post themselves will also only be acompished as a result of LLM usage and already fall under our definition of "spam".
I would like to voice my continued opinion that your all's "negative externalities" are much-overblown.
The context of this thread is this other one pointing out that @FiloSottile (the former lead of Go's cryptography team and maintainer of several notable tools including mkcert and age) deactivated his decade-old account because of the vibecoding tag being applied to his (entirely human-written) blog post (about a result found by an LLM, so the tag's application was consistent with how it was used elsewhere).
Possibly you don't think that's a negative externality, but I certainly do.
I mean, he rightly pointed out that it's really hard to find an llm-friendly and fash-unfriendly place on the net, and from personal experience, he's right – he's so close to getting the point.
Also, I can't imagine what it's like to write thanks to a list of people spanning multiple lines when the result is the work of 30 minutes of a data center brute forcing it (as the Anthropic post tells it).
That’s basically exactly what I’m looking for. Leftist but not so rabidly anti AI.
such places do not exist, it's like asking for a rabidly pro-microsoft leftist space.
That was a very swiftly applied purity test.
I think maybe there's a confusion of terms happening here.
To not put too fine a point on it: If you view yourself as a leftist, what do you actually mean by that?
If you're a leftist in the "opposes hierarchy,, pro-socialism, pro-anarchism" sort of way, supporting pro-fash AI companies like Anthropic and OpenAI invites a contradiction. Thus, you cannot be "pro-LLM" and "anti-fascist" without hypocrisy.
Unless, of course, you just mean "less racist than MAGA" when you say "leftist". There's a wide berth of political identity not explored in this comment (and some of it is quite stupid and self-contradictory, but not all).
(I'm using the Royal You in this comment, because it applies to many people and sometimes even myself.)
Can someone be not-anti-LLM as a technology, anti-LLM-companies, and antifascist? It would seem to me that not being against LLMs as a technology doesn't exclude someone from being leftist, anticapitalist, etc. Not all use or development of LLMs need be tied to these giant corporations. I believe there is value in LLMs as a tool and that it is possible to build/train/run them in an ethical, anticapitalist, antifascist way. I don't see how someone can't be pro-LLM and leftist, but I understand how someone can't be pro-LLM-companies and be leftist.
In practice, the technology is inherently tied to the kind of immense wealth and power concentration that makes it viable to produce, and the kind of informational homogeneity that they produce in turn.
That doesn't mean that you can't use it ethically. Opinions will differ from person to person, of course, but while I'm pretty far to the left and avoid AI like the plague, I can see how you could use AI for e.g. vulnerability research while mitigating all the major ethical concerns I care about.
In particular, you can download and use an open-weight model without benefiting AI companies at all. (Accidental) plagiarism is a major concern of mine, but it doesn't apply if you don't use LLMs to generate code. Etc, etc.
I am still skeptical of people who claim to use them ethically, because they do often end up doing free advertising for AI companies anyways, or paying them hilariously large amounts of money for tokens.
But, while "leftist and interested in genai" might be a suspicious combination, it's not an inherent contradiction, and I'm pretty annoyed that we're making such black-and-white statements such as this one upthread. That doesn't benefit the discussion at all.
In practice, the technology is inherently tied to the kind of immense wealth and power concentration that makes it viable to produce
Kind like nuclear power plants: the stuff requires such an enormous investment, you do have to concentrate an enormous amount of wealth and power to build it. France did it with the power of the (mostly elected) state, and taxes. (Maybe the US too, I don't actually know.)
Or hospitals. Building a ton of them all over the country does require redirecting an enormous amount of wealth and power to this single purpose. France did it with universal health care, mostly financed and organised by worker organisations.
Or moon shots. To prove Capitalism was better than Communism (not only, but come on, it was in big part to stick it to the Soviets), the US massively invested in public education to get enough bright minds to increase the tech level, and of course the publicly funded space program itself. I love the irony on this one.
So yeah, much wealth and power is needed to make those LLMs. Totally agree, as long as we're all aware that it doesn't necessarily mean private wealth and power.
https://www.youtube.com/@PaduTeam are explicitly communist, and not especially anti-AI. In fact they see it more or less like yet another loom: the real problem is who control the stuff, and how they're going to use it to screw workers, as they always do whenever a productivity wonder gets invented.
Of course, the conditions under which the thing is produced in the first place, as well as the resource requirements, are serious concerns they seem to be aware of. As such they aren't pro AI. But they're not especially anti either. At least not right now. Maybe its on the long list on stuff to stop producing once the world is finally liberated from the Empire, and we implement The Glorious Global Communism.
Anyway, here's your anecdotal counter-example. I'm sure there are others, though I don't expect they're common.
Yeah, I can't say I care particularly much about one guy on Bluesky, even if he's famous.
It's not the "being famous" part that makes me think it's a loss, it's the "former lead of Go's cryptography team and maintainer of several notable tools including mkcert and age" parts.
(Plus I thought he had good comments, even when I disagreed with him.)
Sure, I mostly mean the "one guy" part. I agree that driving people off is a negative externality. I would put forth, however, that a lot more (also valuable) people have been driven off from all the slopped-out blog posts than have from Lobste.rs's imagined opinions on LLMs / passkeys / fascism (???).
(Like man, I dunno, "people disagree with me about passkeys" is a very weird reason to quit a site. I'm quite shocked that this was the impetus for moderation policy change.)
(Like man, I dunno, "people disagree with me about passkeys" is a very weird reason to quit a site. I'm quite shocked that this was the impetus for moderation policy change.)
This was pretty clearly not his reason for leaving this site, and I’m not sure why you imply otherwise. His “last straw” (not initial) reason was being tagged “vibecoding” on an article that has nothing to do with a vibe-coded project (codebase written without looking at its code), instead of being received excitedly or at least with curiosity about the new results and his explanation/putting-in-context of them.
Which, to be clear, is fine – you may not care about the topic due to who discovered the results, I may not either, but he clearly did, and thinks the majority of this platform is like the two of us in that example, not him, which makes it a classic “culture” non-fit (in hiring terms), which is pretty understandable.
Lobste.rs's imagined opinions on LLMs / passkeys / fascism (???).
The fascism part was referring to communities like HN I think?
Yes, it's pretty clear that he's saying "one of these three is disagreeable to any site I go to", not "all three of these are disagreeable to Lobsters".
That wasn't clear to me -- I read his BlueSky post as attributing all three traits to Lobste.rs. That post does make a lot more sense in the context of tech communities as a whole...
With regards to posts detailing the work which was only accomplished as a result of LLM usage, my suspicion is that the post themselves will also only be acompished as a result of LLM usage
This often holds but the number of cases where it does not is numerous. With this information, the best choice for us who do not wish to see the results of LLM usage is to flag even human-written posts about LLM usage as spam. There is no longer other recourse.
Or to tag them with ai-tooling ? If that's actually what they are about.
The problem is that many of them will not be about ai-tooling, but a result of it; this is what I would hope to avoid.
On negative externalities and given the new guidelines, would the former head of Go's crypto team have been less exercised if their post were tagged "spam"?
Given that the post was written by a person and not an LLM, it should not be tagged, or rather flagged as such.
"should not" and what is likely to happen are two different things. Given a not insignificant portion of people here would like to hide anything to do with LLMs and that is the closest flag left, what do you think will happen?
With regards to posts detailing the work which was only accomplished as a result of LLM usage, my suspicion is that the post themselves will also only be acompished as a result of LLM usage and already fall under our definition of "spam".
Unless I'm misreading you I don't think this is going to be consistently true. Like, the proximate cause of this latest kerfuffle: I think it is fair to say the result it describes was a result of LLM usage, but the post certainly had meaningful human authorship.
work only accomplished as a result of LLM use will now slip through, and I would definitely appreciate a tag for this as well.
If such content is welcome on this site, then I definitely agree with this, I'd be filtering that out by hiding this hypothetical tag. I think this is what a lot of people wanted or assumed the vibecoding tag was for... Like the post says, it became an overloaded tag.
I am assuming the mods will try out this new ai-tooling tag and then will be willing to reassess a little bit down the line. Hopefully this second tag for work only accomplished as a result of LLM will be considered. Again, assuming such content is allowed on this site --- which I'm personally not clear on whether or not it is.
What you said is true. The userbase of lobsters has bifurcated on the question of AI, maybe it's time for the website to do the same.
work only accomplished as a result of LLM use
But you can never know, right?
I loathe this argument. Sure, you are functionally correct, but you can never know if anyone is lying to you. I hope people would have the common decency to disclose, and otherwise when I find out they haven't, they end up in my ublock list. I'll miss some things, perhaps, but that misrepresentation is their failing, not mine.
Besides, it is almost always completely obvious when people don't disclose.
Why does that matter? If you need to hide it it's probably because it's a low quality project. The recompilation scene isn't shy about AI use. It's seriously impressive. I just don't want to see it, even ignoring ethical issues (and I won't), it has nothing to do with how I interact with programming. I have as much use for it as emacs posts. Just let me filter it.
I don't see any way of being a "LLM neutral" community to be honest.
I am mandated to use LLMs at work and have to spend all day dealing with people throwing slop at me. I really do want a way to filter it out of my personal life as much as possible. It's not that I think the technology is totally without merit (although the companies pushing it are immoral at best), I just really want to see anything else in my spare time. Having no easy 'ai adjacent' tag means I can't do that. Widening the 'ai' tag a little and providing an 'ai-tooling' tag doesn't really accomplish that.
I also think comparing it to systemd is disingenuous at best. Yes, systemd was controversial at the time, but systemd isn't powered by illegal polluting gas turbines, shredding irreplaceable books, making people's hobbies completely unaffordable, committing literal crimes that you or I would go to prison for, etc.
I don't think it's unreasonable for people to take a personal moral stance against it and not want to see it in their recreational time.
The vibecoding tag was the thing that kept this site tolerable in the deluge of crap that's subsumed the web. With it gone I don't really see myself sticking around. What a bummer.
Likewise. I may concede that there might have been a few false positives, but hiding that tag was like closing the floodgates. Oh well.
I agree, I would really like to be able to filter out blog posts that talk about work coded with LLM assistance. I have a feeling that this change will increase the noise-to-signal ratio in my Lobsters feed quite a lot, where signal = content entirely produced by a human. Previously filtering out vibecoding was enough to keep my feed clean.
(unless I am misunderstanding -- does retiring the vibecoding tag mean that vibecoded projects are now considered off-topic?)
My reading is that projects are now banned generally ("we no longer allow submissions which are code repositories"), as is most promotion of your own work, that LLM-generated blog posts are also banned (should be tagged as spam), and that human-written blogs about LLM-generated code are allowed. Although of course they should still be interesting or novel in order to get upvotes etc, or will fall off the front page.
I wish this was stated more clearly in the OP, rather than having to be derived from a series of separate rules.
I got the impression that blog posts discussing a project's insight is still allowed (e.g., "here's a new borrow checking mechanic" v. "here's my new language/compiler; you need to read the entire repo to understand this new borrow checking mechanic").
My understanding is that many vibecoded projects were already off topic (and this was part of the confusion around the vibecoding tag). The post introducing the vibecoding tag said,
I'm going to start removing [posts] ... showing off things that are made with vibecoding tools, unless they're a deep dive into implementation.
=> https://lobste.rs/s/lkngrz/new_tag_vibecoding
The site description for "spam" allowed posts to be flagged for insufficient human authorship. And now this post states that linking to code repositories is not allowed. And so someone promoting a slop project should be flagged.
But, and I think this is a big "but", technical posts describing an LLM-driven rewrite or the technical merits of an LLM-produced project are considered on-topic and have no tag. (For, example the recent Anthropic post describing their vibecoded Scion project I believe would fall into this category.) This doesn't effect me personally so I don't feel strongly on this, but it seems weird to me that there's not a tag for (paraphrasing OP) "marking content which includes a technical discussion on output from AI/LLM tools."
Same. This was my safe haven for the short while I've been here after escaping from HN, but I don't foresee myself sticking around either. Anyone have any recommendations on where else to go?
awful.systems is the right community of people, but you might not like their main comm's focus on sneering at TESCREAL. I think they have a comm called goodtechnology or similar for positive discussions of slop-free things.
I am new here but I was so happy to find a remaining site focused on the art of coding when the rest of the world is now just "how to tune LLM". I'm a bit sad that it feels this last dam may fall soon with that news. I have seen it all around me, even with talented folks, their technical skills crumbling because according to their own words it does not matter anymore (even if they initially were in the camp "it's just a tool"). I don't mind if people want to talk about how to manage LLMs, which is much more akin to a manager trying to micro manage their engineers in the vague hope they do as they are told. I want to talk about the art of code, even if some believe it's a lost art. Is this the site for this? I can't see how the mission statement of the website mentioned by OP and the disappearance of AI flag are compatible.
I hope its removal leads to greater use of the spam tag for low-effort slop. If not, though... yeah.
Fair enough. I realized PTSD with what happened elsewhere may have tinted my reaction here. Fear is never a good guide. I hope you are right and things will remain great here.
I was also filtering out the vibecoding tag. I hope we find another way to discourage the constant influx of vibe coding relating news. I probably will stick around as the moderation here is otherwise excellent and the discussion very good. :)
I agree, sadly.
I don't see how repurposing the spam flag to also include LLM output is going to make things better. People will complain the same for being "unfairly" tagged as spam (of course they will!).
In my opinion it will make things worse. Despite complaints, vibecoding-for-all was clear and it didn't prevent anyone what wanted to read that content from reading it, while "ai-tooling" plus "spam" are going to be misused and the conclusion will be content harder to filter by people that don't want to read anything related to LLMs.
I feel like there is a missed opportunity here to make everybody happy. If the reason to not use "vibecoding" was "a tag of shame" that it would push people away, what do the moderators think this new situation is going to accomplish?
It took 5 meta posts to get to this change. I don't see why we can't have one meta post every week if things aren't working for everybody.
the conclusion will be content harder to filter by people that don't want to read anything related to LLMs.
This is why I've advocated for the tag to be called "llms".
same. I am hoping there can be an additional 'tag of shame' for content not explicitly related to AI/LLM tooling but still in some way contains LLM generated material.
From the track record, it seems likely that such a tag would get applied to human-written posts about software that contains (or is suspected of containing) some LLM-generated bit somewhere.
It seems unlikely that such tagging would make discussion better or would be useful as a filtering signal.
Agreed. I'll try and keep an open mind while sticking around for a while and watching the actual outcome of this, but I fear the worst.
While not perfect, the tag and it's usage was the main differentiator between this site and a few comparable ones for me, and this change has the potential to destroy all of that.
Now would be a great time for someone to set up a new community explicitly focused on the craft and joy of programming and computer science. I don't think I can do that but I'd love to help such an effort.
Yep, it'll be yet another victim of the "asymmetry of attention" attack waged by all the LLM mongers. Generate reams of crap using stolen IP with zero effort and expect the rest of the world to spend their limited attention on it, no thanks.
How would you feel about six tags corresponding to the six identified overloaded meanings in the OP?
Some of those meanings are already covered by flags, so you shouldn't need six new tags.
Without conding 'tag of shame'ing, what combination of flanges does one need to.hide so as to avoid all 6 of above mentioned categories?
If there is no way to avoid any LLM (related|adjacent|generated|assisted) stuff, I see more free time in my future. Shame though, I liked Lobsters for however long it lasted.
How would you classify a blog post where a person talks about their latest creation they have entirely vibe developed? It’s not ‘ai-tooling’, doesn’t mean it’s necessarily spam, but some of us would want to avoid this increasingly popular type of content.
The ‘vibecoding’ tag did fit perfectly.
EDIT: I have to say I will miss posts that are just some projects repo. This is my main way of finding cool stuff, or to spend a pleasant 10 minutes reading some stranger’s code, and they are not always accompanied by a PR post that can be easily shared on social media, which would be allowed on Lobsters. If this site is for programmers, we can deal with a github link, thank you.
The ‘vibecoding’ tag did fit perfectly
It did if that’s what people applied it to.
The term “vibecode” was created by a guy describing using LLMs and never looking at the code. That was all it was.
It immediately (I counted days) became deviated and started to mean any use of AI, any discussion about AI, any meta about AIs, etc. Lobster wasn't at fault here but the meaning of the term was completely misunderstood from the get go by most of the programming forums.
Now we have people like Linus Torvalds who hand wrote millions of line of code merely mentioning LLMs and being tagged as vibecoding. It’s a useless tag.
Look at a perfect example: https://lobste.rs/s/k5omum/brain_sandwich
A human written post about the negative externalities of AI and how it’s changing the industry, extremely worthwhile discussion, is tagged the same as a post about some dude who vibecoded an entire code base or used LLMs to completely write an article.
The term “vibecode” was created by a guy describing using LLMs and never looking at the code. That was all it was.
It is worth pointing out that, in the “agentic” world, that’s ironically more or less what a lot of the big voices/influencers have come back to. LLM-generated code that’s also LLM-reviewed and LLM-fixed-up before commit, with the human operator essentially never looking at any of it and just trusting that all those layers of agents will manage to get it right.
I agree with that, agentic is definitely the end state of vibecoding as some people want it. Sort of vibecoding with high level managerial processes, which funnily reflects how human programmmers have been looked at throughout the decades. We’re fast approaching 1:1s and performance reviews for agents.
Lobster wasn't at fault here but the meaning of the term was completely misunderstood from the get go by most of the programming forums.
The Lobsters tag was defined as “using AI tools”.
Which is wrong by definition. What I mean is that it’s not only Lobster who misinterpreted the term, it’s become an issue in all forums.
Detached from the industry at large I’d say, at all workplaces I know of “vibecode” is not looked at kindly and is not equated to AI tools.
You yourself said the common usage of the term deviated within days, across the broader Internet, not just Lobsters. You can play prescriptivist but it may be a losing battle -- I for one won't limit myself to definitions from the likes of Karpathy of all people.
Use the tag woild have been a good fit for that narrow case.
Though if they're not talking about how they develop it (their process and tools aka ai-tooling) then it's just promoting a project and thus off topic and can be flagged as such.
talks about their latest creation they have entirely vibe developed
I see increasingly more people not writing any code by hand anymore, so this is the new norm.
As I mentioned in IRC, at least this is something and I hope it works well. Lobsters has done many things that I wouldn't have done that seem to work well, so I hope it works. I do not envy the moderators.
I have mentioned this repeatedly: the LLM mania has either flooded communities or polarized them. Just yesterday, I answered someone asking for communities with no LLM talk for their mental health. The flood is clearly hurting people, and I think these actions will complete the flood. The vibecoding tag was the second most filtered tag on Lobsters and now all those users are losing a way to mitigate the flood, because ai-tooling is much more narrow. (I won't filter, but I can care for people who do.)
What remains is the spam tag. There was a recent controversy about an article about research done by an LLM company. They quite obviously did this to promote their commercial services. Will that article help anyone improve their next program?
What constitutes meaningful authorship? I don't think this covers what hundreds of Lobsters users want to filter.
I think communities affected by the flood and polarization have a single option: choose. I fear these changes imply a choice I cannot share, and that makes me sad because Lobsters has been wonderful to me.
I cannot stand in the LLM side. I have quit my job and embarked on a significant (and impactful, negatively in many ways) career change because I'm apparently one of the few people who can read this article from Amnesty International about how generative AI is in conflict with international human rights law and not ignore it. I asked @pushcx to read it (and he told me he would try to read it) and now I'm asking the entire moderation team to read it and justify how it cannot affect moderation. I do not have high hopes, because I have not seen this article sway anyone.
I fear this change will hurt people. I fear this will be one of the last steps of me losing my garden party.
I hope people who lose this garden party will find a new one or I hope I am mistaken, I have been wrong before.
On minor technical things, I think forbidding repository links creates the wrong incentives. I have reviewed my upvotes and while I don't upvote many of those, on my old upvote pages there's roughly one code repository upvoted per page. I have discovered quite a few interesting projects this way. I fear some of those projects I discovered really do not have easy alternative ways to be submitted. And I think submissions targetted by this policy change will just use some low-effort technique to skirt it, because we have seen constantly that people will try to skirt policies in similar ways. I expect this will be addressed in a few iterations, but honestly I do not understand this change, it just seems to create perverse incentives.
The burden of proof lies upon the claimant. Support your argument here: how does the paper force a change in moderation tactics?
I do not disagree with you generally (and in part, I'm rooting for what you're saying), but the reasoning in the text is thin in places, for example, the claim ...
Crucially, the systems’ ability to influence decision-making processes may impair autonomous thought formation and critical discernment, instead facilitating the transfer of biases held within the training data to users of the particular generative AI system in question
... is not immediately substantiated by research. If the negative effects of generative AI are how you reason your claim, then they should probably be evident if you're arguing they constitute a change.
Several thoughts and questions.
Recently a story that was, effectively, news about the future (and likely shutdown) of an open-source project was removed as off-topic because the immediate cause of the news was an acquisition by a large company, hence the post was "business news". The immediate cause of the moderation changes being announced here was the vibecoding tag being applied to a story that boiled down to "an LLM vendor did a thing", including the author directly crediting the vendor and saying they wanted to send a gift to employees of the vendor as thanks. If "open-source project has news because of a company" is off-topic, how is "LLM vendor did a thing" not off-topic?
Like many people, I apply a rough guideline of "if someone couldn't be bothered to write it, I can't be bothered to read it". If someone submits clearly LLM-generated content to this site, will it be possible to report it on those grounds? Will it be removable on those grounds? This doesn't rule out people submitting stories about using LLMs, just imposes a requirement that they put in the bare minimum of effort to actually write it up themselves.
It is virtually certain that there will be posts using the release tag linking to release announcements of new models and new associated tooling. Currently, the general expectation on this site seems to be that release should not be used alone, but should be accompanied by another tag indicating the language/technology/domain of the released thing. For example, the recent Python 3.15 release was tagged release and python, the recent Ubuntu 26.10 release was tagged release and linux, etc. Will there be a tag which LLMs and LLM-related tools are expected to use for release announcements, and which can be applied by user suggestions if a submission author does not apply it?
Same question as the last one for submissions about practices and techniques; when I write articles about using Python, I tag them python, and if I somehow forget to I expect the tag to get applied by user suggestions. If someone writes articles about using LLMs, why shouldn't there be a tag for that?
Regarding "punching bags": I am skeptical of whether LLMs represent a genuine revolution in software development. For this, I have been called a Luddite, a terrorist, a transphobe, an enemy of human progress, and much more to boot. If it's going to be made impossible to filter away LLM-related submissions when browsing this site, will the moderators commit to responding as swiftly and severely to such rhetoric as they did to Filippo's complaints on social media today?
Finally: it is clear this set of changes was immediately precipitated by, for lack of any better way to describe it, someone with a relatively large social-media following making a post to rile up their followers to anger against this site. Do you really want to set the precedent that this is a way someone can win an argument with the moderators?
It is the thing that the LLM vendor did (found the source of NIST curve constants) that is on topic, not the fact that an LLM vendor did it.
LLM-generated content itself should be flagged as spam, and releases of LLM models/tooling should be tagged ai-tooling along with release. Articles about using the tooling could also use this tag. The article about NIST curves was not about using LLMs, it was about cryptography.
The LLM vendor did it as a promotional stunt to get attention, and we both know that.
Also, if it had been done with, say, a particular language’s tooling which made the task more amenable, it would (should) have been tagged with that. So why not tag it when it’s done with an LLM?
I disagree. If someone had written a Rust program that reverse-engineered elliptic curves, the blog post by an third-party cryptographer that links to that program at the end wouldn't merit the [rust] tag. My rationale is that such a post is neither of special interest to Rust users nor something that would be considered objectionable to people who don't want to use Rust.
I'm sure you're right that Anthropic published this result for marketing purposes, but Felipe didn't - it's just an interesting result in his field of expertise.
Recently a story that was, effectively, news about the future (and likely shutdown) of an open-source project was removed as off-topic because the immediate cause of the news was an acquisition by a large company, hence the post was "business news". The immediate cause of the moderation changes being announced here was the vibecoding tag being applied to a story that boiled down to "an LLM vendor did a thing", including the author directly crediting the vendor and saying they wanted to send a gift to employees of the vendor as thanks. If "open-source project has news because of a company" is off-topic, how is "LLM vendor did a thing" not off-topic?
With regards to the Deno acquisition and removal, I've written about this previously.
I'm not sure it's a direct comparison between the two postings. To use a semiconductor example the Deno post feel more like Annapurna Labs' post that they've been acquired by Amazon Web Services, and the post around the NIST Seeds feels closer to a AWS discussing a technical achievement around the Graviton architecture.
Like many people, I apply a rough guideline of "if someone couldn't be bothered to write it, I can't be bothered to read it". If someone submits clearly LLM-generated content to this site, will it be possible to report it on those grounds? Will it be removable on those grounds? This doesn't rule out people submitting stories about using LLMs, just imposes a requirement that they put in the bare minimum of effort to actually write it up themselves.
If the content is clearly LLM generated, it's flaggable as spam. This is also why I explicitly call out that we want content that is "at a minimum, be respectful of a reader's time and intelligence".
It is virtually certain that there will be posts using the release tag linking to release announcements of new models and new associated tooling. Currently, the general expectation on this site seems to be that release should not be used alone, but should be accompanied by another tag indicating the language/technology/domain of the released thing. For example, the recent Python 3.15 release was tagged release and python, the recent Ubuntu 26.10 release was tagged release and linux, etc. Will there be a tag which LLMs and LLM-related tools are expected to use for release announcements, and which can be applied by user suggestions if a submission author does not apply it?
As mentioned in the post, the tag is ai-tooling.
Same question as the last one for submissions about practices and techniques; when I write articles about using Python, I tag them python, and if I somehow forget to I expect the tag to get applied by user suggestions. If someone writes articles about using LLMs, why shouldn't there be a tag for that?
With some restrictions, that new tag is ai-tooling. The issue with the vibecoding tag was not tagging, but the behaviors and the associated moderation interventions around it.
Regarding "punching bags": I am skeptical of whether LLMs represent a genuine revolution in software development. For this, I have been called a Luddite, a terrorist, a transphobe, an enemy of human progress, and much more to boot. If it's going to be made impossible to filter away LLM-related submissions when browsing this site, will the moderators commit to responding as swiftly and severely to such rhetoric as they did to Filippo's complaints on social media today?
If you feel that another user's conduct towards you doesn't meet with community norms and site policies, feel free to reach out to any of the moderators.
Finally: it is clear this set of changes was immediately precipitated by, for lack of any better way to describe it, someone with a relatively large social-media following making a post to rile up their followers to anger against this site. Do you really want to set the precedent that this is a way someone can win an argument with the moderators?
I really do hate to say, "look back through the sysop comments", but from a timing perspective, we've discussed this direction as an option we would take well before any of the social media discussions started.
As mentioned in the post, the tag is ai-tooling.
At this time, the post says:
This new tag
ai-toolingcovers "The design, ux, development of AI/LLM-powered tools themselves".
If someone writes up their thing that they used Python to do, currently we expect them to tag it python. If someone writes up their thing that they used an LLM to do, the description in the post does not seem to me to imply they'd be expected to tag it ai-tooling.
To be absolutely clear: I'm saying that just as there are tags for various languages and domains and platforms and we expect people to use them, there should be a tag for things which made or were made with significant use of LLMs. If that is the ai-tooling tag, then I think that needs to be explicitly communicated. If that's not what the ai-tooling tag is intended for, then another tag needs to be created, quickly, to fill this role.
The issue with the
vibecodingtag was not tagging, but the behaviors and the associated moderation interventions around it.
I don't agree. I read Filippo's Bluesky thread (and responded in it), and ultimately the objection seemed to me to be that there was a tag users could choose to filter if they didn't want to see posts about things done with LLMs. On Bluesky Filippo described this as a "scarlet V". To me that reads as saying that any tag which would be applied to stories about LLM use/LLM-assisted projects is unacceptable and would be interpreted as persecution of LLM users.
And I don't think this site needs to tolerate that sort of stance. Every time I submit something related to Python here, the tag box reminds me that a couple hundred people filter python and therefore won't see my post. And I just get on with my day. I don't take to social media to complain about it, or to claim the site is branding me with a "scarlet P", or to say that people here who criticize or dislike Python are inherently "unhinged" or "not discussing the same reality" or any of the other hyperbole Filippo used today in his rant about this site and the vibecoding tag.
Neither Filippo nor anyone else is entitled to the attention of others on their blog posts. If I don't want to read certain posts, then that's that. I don't have to justify myself any further or come up with a good enough reason to pass Filippo's bar for being in "the same reality" or whatever other nonsense. "No" is a complete sentence and a sufficient answer, and I worry that this site is starting to cave in to the demands of people who, at least as indicated by their behavior when told "no", do not seem to think so. I also think this site is better off without those people.
If someone writes up their thing that they used Python to do, currently we expect them to tag it python. If someone writes up their thing that they used an LLM to do, the description in the post does not seem to me to imply they'd be expected to tag it ai-tooling.
That's because we're interested in languages and how they can be used per se, but AFAIK we don't tag every Linux program C or every Android program java either, unless it is somehow relevant to the conversation. Similarly, we don't tag posts with emacs, vim or whatever-IDE the person used to make their program. I would only really find it useful to tag it as ai-tooling if it explained exactly how (in technical detail) they got the LLM to build their program.
It often is relevant to know what language an Android app is written in since there are multiple possibilities. Same sometimes for Linux programs.
More generally, though, you seem to assume LLM generation is so common as to be an assumed/unremarkable default, which is something I’d push back on.
More generally, though, you seem to assume LLM generation is so common as to be an assumed/unremarkable default, which is something I’d push back on.
I see so many posts where people involve slop machines that it's not helpful to tag them just to indicate this, which is one of the reasons the tag is rather useless. And of course some don't even disclose their use of LLMs, which also means it's not as useful a tag unless one does some digging and then add the tag.
Anyway, I wish it wasn't the default, but that train seems to have left the station somewhere earlier this year.
I see so many posts where people involve slop machines that it's not helpful to tag them just to indicate this, which is one of the reasons the tag is rather useless.
Maybe I'm reading you wrong but my wish to not read llm-anything stuff does not decrease with it's abundance increasing. If I'm not able to filter it out it decreases the signal to noise ratio.
Seems like a good change. Site purpose clarity is nice to have, and I think ai-tooling will be useful both for discovering and for filtering out articles about LLM-related tools. Thanks for working on this with everything else going on!
hugs to the moderators. this feels like a no-win situation. i'm as anti-AI as it gets but half of this comment thread is really depressing.
First, I appreciate the tricky task of addressing this. So thank you to everyone involved.
With the tag now gone, I'm not sure where that leaves those of us who can't invest the time and energy into manually sifting through mountains of content that is, from our perspectives, at least uninteresting. For example, if someone is categorically uninterested in any submissions involving Rust even tangentially, they have an effective way out filtering out such content via a tag. An exclusionary choice is not necessarily a value judgment; it's an assessment that a set of tags, as used in practice, is an effective proxy for their preferences.
Not sure how I feel about this. Are people being shamed for vibe coding, or are they feeling ashamed for their posts being labelled as something that people filter out? This isn't a rhetorical question, I have the tag filtered so don't see the discourse.
I don't want to see posts related to LLMs on lobste.rs, but I also wouldn't harass authors over it. For that reason I'd like to still filter posts using vibe coding, but would want shaming authors in comments to be moderated. I'm not certain that criticism of LLMs counts as shaming an author for using them.
The connotation of vibecoding is that no effort went into writing the code. This is often signficantly false. Hence it's interpreted pejoratively when applied to something someone cares about.
Forgive me, but isn't the goal of generating something with an LLM to take away the effort from the human by definition? I can see how a human can put in effort around what the LLM has generated, and therefore the whole thing wasn't vibe coded.
I'm not sure the phrase "vibe coding" itself is the pejorative, it sounds more like the practice it is describing has some inherent negative connotations that any other name would also carry. If the project in question used LLMs enough that the post mentions it, then surely it meets some threshold for being marked as such, whatever the tag's name is.
Sorry, I think I understand better now. Is the problem that the tag implies the whole thing was vibe coded and took zero effort, but it's used to tag projects that are only partly vibe coded and otherwise took effort?
We don't allow entirely generated vibe coded submissions, so I think the tag means only "partially used vibe coding".
Then the tag should have been renamed rather than removed. (Even ignoring for a moment that I know multiple AI users who use "vibecoding" for generated code that they review and adjust, so I'm not sure if your interpretation is universal.)
Steelman: s/the connotation/a connotation shared by more than one person on lobsters when it's applied to a project level thing/ - see 74 votes on https://lobste.rs/c/sruweq
I often will jokingly refer to vibslopping some work, and that's fine. You might say you're spamming your friends your new track or whatever. But when slapping it as a label on someone else's work, you're being careless as to how it's interpreted. The 6 definitions in the first post are the problem, which of these should be renamed and how?
Retiring this tag is the right approach IMO.
Removing the vibecoding tag might have been a good idea but overall this sounds a lot like an implicit, more than an explicit, endorsement to post more LLM content. Or it will lead to that, I fear.
For the repo post changes, I'd really have to check my comment history if I have a strong opinion. Maybe we're losing out on some cool things, and we'll definitely losing out on any "interesting" new version releases (but I fully agree these are highly personal, I wasn't interested in 80% but was very happy about the 20%).
So there's going to be no way to distinguish LLM-generated blogposts and code repositories in submissions that isn't opening the link and wasting my time on it? The slop flag or an "this article is primarily AI-generated" would be a great introduction for that reason alone.
If you read the announcement: there's going to be zero LLM code repos going forward.
In the sense that there are going to be zero repos at all, which seems like a baby-and-the-bathwater situation.
It also explicitly (in my reading at least) allows "here's why you should look at my repo"-style blog posts. I think that's fine in itself (I like those posts, hell, I've done it myself!), but there's also been a pattern of people abusing them to launder slop (where the repos would have quickly be removed as "I vibe-coded a thing", the posts would often be allowed with the tag).
I have disliked most repo-submissions for a while now, mostly because the guidelines didn't encouraging linking directly to a project's Readme file so you had to scroll past a list of filenames, and that a short summary about why this project was submitted was discouraged.
Having to put together a post motivating pointing to a repo is at least a small speedbump hopefully encouraging human involvement (with all the caveats that come with that nowadays).
It's still subject to the Bullshit Asymmetry Principle. Sloptomizers just need to have Claude shit out a blog post for them to go along with the repo.
Sloptomizers just need to have Claude shit out a blog post for them to go along with the repo.
Sure, but then it can be flagged as spam, is my understanding.
Sloptomizers just need to have Claude shit out a blog post for them to go along with the repo.
This would fit the definition of the spam flag.
This is true, but a blog post has its own associated metadata. Was the blog created 10 minutes after the repo? That's a red flag.
Thanks, I should clarify blog posts "about LLM-generated code repos" instead of just "code repos". Only "low quality", LLM-generated posts are to be marked as spam (which doesn't mean they'll necessarily be removed, either). There should be a way to distinguish "this is work primarily done by humans/about work primarily done by humans" vs "this is about primarily LLM-generated work", without marking it as spam. ~fzakaria has done great work for Nix with LLMs and written great posts on that work, but as someone with the vibecoding tag filtered I wouldn't want the whole homepage to be blog posts about work LLMs have done.
I am in a generally negative headspace at the moment due to unrelated issues, so I will have to let this decision play out for a while before deciding what to do. For a long-ish time I have been contemplating just leaving the site, and maybe this will be the final push.
I don't envy the mod team. Going from the mission statement, it's impossible to have a forum focused on computing that excludes GenAI now. The "vibecoding" tag was the compromise - if you don't want to interact with this stuff, just filter. That's now gone, as far as I can see. I would have preferred a straight re-name to something that wasn't seen as pejorative, but oh well.
I've been so disappointed with the decision making here this year. I know I should appreciate the work the moderators do, but I can't help feeling betrayed and rejected. Comparing LLMs to systemd is wild to me. Maybe this just isn't, or is no longer, the place I thought it was. My gut feeling right now is that I should leave for my own mental health. But it's also the last community I really have in tech. I respect a lot of the people here, but I worry they will move on too, and I don't think there's really anywhere to go. It's a very lonely world right now.
I agree that the vibecoding tag was overloaded but these changes work out better for the pro-LLM crowd which is a real shame.
Thanks. I understand that there isn't any perfect or pleasant solution to this problem, but I think that the one that you guys picked is one of the viable ones. You might need to add another flag reason eventually so that it's more obvious that pure llm output is considered spam, but for now I think the set up you've described is worth giving a shot.
I think this is a helpful change. I, like many here, have serious ethical reservations when it comes to LLMs, but an old man once told me "people don't get better by making them feel worse." People shouldn't be made into punching bags. Their shit choices should be called out, slop should be discouraged out as a waste of everyone's time and attention, but we've had slop for way longer than LLMs. Echo chambers don't make people better, no matter how right you think your side of the room might be. If you're not listening and thinking, you're part of the problem. Being "right" doesn't mean shit if you can't also be kind.
I don't love this change but I think it's an improvement and I really appreciate the work that goes into moderating the site. Thanks.
shaming aside, I've found the tag to be very useful to indicate that the submission either talks about LLMs or uses LLMs to produce something. If we assume, that we want such a tag (or a pair of tags), then renaming it from slop->vibecoding->harness_engineering and whatnot is pointless, because the meaning will be quickly contaminated with the sentiment.
Perhaps "Slop" as "Spam" equivalent for heavily LLM-generated content and "LLMs" as a marker for /discussion/ about the technology?
AI-tooling is also too vague, because vibecoding covered only LLMs, while AI-tooling applies to many less controversial technologies.
edit: this proposition still does not resolve the issue of people wanting to filter out also llm-news (like Fable hacked a shed or something), but at least it is a middle ground (removing vibecoding tag is not a middle ground, it is conceding).
Pending an upcoming deploy, we no longer allow submissions which are code repositories.
This really bums me out.
Historically, a lot of the best things I have seen on Lobsters are code repositories where the README is the project's de facto website. I have put a lot of effort into READMEs, and others have as well. LLM jockeys won't be affected; it's just as easy to shit out a Github Pages site alongside whatever code. Please do not ban repositories just because too many slop hogs have lowered the average value of a code repo.
Thank you for listening and making these well considered changes. I’m sure you won’t please everyone but I for one am hopeful they will shepherd a return to more pleasant thoughtful conversations about the art of programming
I think this is a mistake. Frustration with LLMs that was localized to a mere tag will now leak out into comments, making the community feel far more toxic.
(I speak for myself here. If an article is generated using an LLM, yes, I’m going to call that out, because I treat LLM prose like second-hand smoke and it pisses me off when I’m forced to read it unwittingly.)
If anything, the "vibecoding" tag led to more commentary about LLM usage rather than less, so I don't see how removing the tag will increase the number of off-topic comments.
The answer to the second concern is that articles generated with LLMs should rightly be flagged as "spam" rather than commented on.
Not all LLM-generated content is spam. But I’d still want to know.
It's all fit to be flagged as spam because that's part of the flag's definition. But similar to the "vibecoding" tag, this is not obvious from the name and I think there would be value trying to make tag and flag names like this more intuitive, or just adding more.
I guess this is the last chance to post it here so here's a little bit of stats... I made this are we slop yet? page a while ago to keep track of the tags on the front page. For a long time, the vibecoding tag was OK-ish, only occasionally surpassing other popular tags like rust (and, sadly, security... Was the security flag so popular before LLMs?). But the stats from the past whole month has definitely made vibecoding the most popular tag on the frontpage. (Interestingly though, the vibecoding tag count has not gone up significantly. Does that suggest reduction of other tags or increase of vibecoding-tag-only posts?) But anyway, I really do appreciate the efforts to update policies regarding these.
Thanks for the writeup. Moderation–particularly around a hot-button issue as contentious as this–is not easy, so kudos for keeping a level head and thank you for all the work you and the team have put into keeping the site clean and for updating us on the new policy.
My gut reaction is that I'm worried about seeing more and more low-effort and generative AI-centric content flooding the zone. One of the things I like about lobsters over sites like Hacker News is that I can get some peace from the incredibly boring onslaught of people doing things with LLMs. I don't find this interesting, personally, and the vibecoding tag did a reasonably good job at keeping the feed human-only. We'll see over time how well the new system works overall, I suppose.
My gut reaction is that I'm worried about seeing more and more low-effort and generative AI-centric content flooding the zone.
Yes, and I was similarly knee-jerk tempted to read this change as a loud contingent's success in painting themselves as victims, but instead: (1) this is a compromise to keep enough people happy enough, as many moderation decisions are, and (2) on the flip side this may indicate to more people, who may have previously been filtering the vibecoding tag, that they should use the "spam" flag for submissions they view as "slop". Time will certainly tell if this change steers this site closer or further from being just another tech link aggregator.
I'm already 200+ comments late, so many good things have been said. Some less nice things, too. I've been complaining about this situation, so the least I can do is to offer my own thanks and (quick) thoughts.
Firstly, I'm mostly glad a decision was made, even if I disagree with it. In the end, it's better for all involved (staff and users) if Lobste.rs doesn't turn into a war zone over ethical disagreements, however important they are. So thank you 355E3B, pushcx, Irene and aleph.
Of course, I would have preferred a more ethical approach to this decision. Koala has made this point most eloquently, so I'll link to it.
355E3B, you said that you "started reading Lobsters because it had the most interesting viewpoints related to computing," and that was absolutely the case for me too. However, what y'all taught me consistently over these last twelve years is that the ethics of those viewpoints are important, too, and sometimes even outweigh the potential benefits.
Now, I amwas filtering vibecoding, so it's possible that I missed some kind of acknowledgement by the staff of the issues at play, in which case I apologise for bringing that up.
Where does that left me? Frankly, I don't know. These last months, I've increasingly quit spaces where I had the bad experience of getting excited over something, only to find out that LLMs were used in a key role. So, maybe that will happen here, too. Or maybe I'll still enjoy enough of the site to make it bearable, who knows. Either way, it's been twelve wonderful years.
A big part of the problem is that the tag was being so overused and misused that I don't see how anyone could have a clear picture of this site's community if they were hiding everything with the tag. Hopefully things can be good with people appropriately flagging spam as spam and off-topic as off-topic, and we can avoid beating dead horses in the comment sections of the things that remain. If the only thing worth saying under a story is a complaint about how it was made, then that seems like a good indicator it doesn't belong here.
I'm a fairly strong AI proponent (worked at OpenAI for a bit on Codex, use many billions of tokens a month). I think it's a good thing to remove the vibecoding tag for all reasons mentioned. Thank you.
My bias acknowledged, I think it's reasonable for people that choose it to not want to read predominately AI generated work. That often doesn't have an overlap with the new ai-tooling tag. I think we need something neutral which allows users to tag correctly. ai-generated or gen-ai is a reasonable tag that would help this. slop is overly perjorative and judgmental and suffers from the same problems a vibecoding. spam is incorrect in the same way as vibecoding was.
I want to talk about tooling in this here garden party that's developed with AI, with people that also want to do that as I think that's a reasonable take and something that the software industry broadly accepts. There are lots of people that disagree with this noisily. I want something that makes it easy for those noisy few be avoid interacting with me.
Given the fact that you're banning a lot that was previously tacitly allowed, and adding… well, a tag even of as-of-yet undetermined usefulness, this isn't the worst change in the world. Although given what the front page looked like before I had an account and could filter the vibecoding tag, even if 50% of that is deleted the pleasantness of the site will certainly still be drastically reduced.
However, this has been really nagging at me for the past hour:
… tools like compilers, databases, testing, and vcs had similar points of consternation and concern when they were introduced.
Directly and conspicuously ignoring the novel aspects of LLM technology that opponents and proponents love to point out literally every single opportunity, very solid and not-at-all fallacious argument there. Seriously, I really believe that the majority of both sides could unify together under the banner "AI is not at all in equivalence to compilers and databases and version control."
I really believe that the majority of both sides could unify together under the banner "AI is not at all in equivalence to compilers and databases and version control."
idk if you missed it, but Linus Torvalds recently made the same poor comparison in an email thread while calling the person he was responding to unintelligent.
Which is why I made sure to say "the majority of both sides" and was considering but ultimately didn't add "except for people making the same fallacious argument that 'you have to accept it like any other tool'".
But when not making that argument to try and force people to accept it, boosters love pointing out how unique and novel and all the capabilities it has versus anything that's come before at every opportunity; just like opponents point out why that means it has to be treated differently from "just another tool"
I truly hope I will not have to leave the site due to it being filled by meat proxies. So far it has been quite pleasant to use with the existing tag system. I would not change a working solution.
Interesting that compilers, databases, testing and vcs are compared to as having similiar proponents and opponents situation, yet they still retain the tag. Does the "tag of shame" argument not apply to them as well?
The comparison with compiler/db/vcs, is meant to show that the new ai-tooling tag describes what it is, rather than how it was made. It's a simple objective category, orthogonal to any "proponents and opponents situation", that people may be interested in or choose to filter out.
I really appreciate this thoughtful explanation of your reasoning for all this, and I hope the change makes the site a better place! Thank you.
It is unfortunate that AI boosters get incredible leeway to have policy changes made that benefit them at all costs, without considering the vast majority of people whose labor and livelihoods are economically and socially impacted by their destructive behavior.
As many have said before, vibecoding is a term that was coined by an AI booster. I find the term quite accurate - coding based on some magical vibes or feelings at the spur of the moment or with no concrete plan. I don't see any negative connotation implied by it... quite a few boosters have embraced the term, and don't see any baggage or problems with it. The only "problem" it presents is describing a submission with increased accuracy. Why would you want to hide or shy away from this label if there is nothing wrong with it? That's the naive question: the answer is roughly, "so the boosters get to control the narrative". They get to promote whatever they want, and you aren't able to call them out or even acknowledge their true motives.
Boosters get all upset when faced with anything less than special treatment. They are loud, irresponsible, factually inconsistent, and deliver flawed arguments at every turn. None of these behaviors are worth catering to or amplifying.
A creative and very reasonable set of changes. Thanks for spending your weekend on this stuff.
If this kind of community schism happened 20 years ago, 5 "AI free" lobsters forks would already have been created with their own growing communities figuring out where and how to draw the lines. That's one of the core features of the internet that used to be taken advantage of much more frequently. I have a feeling the rise of social media and the oppressive nature of big tech platforms has had an effect on user agency.
I like the direction the mods are taking here. But I'm also open to exploring other alternatives. If anyone does create their own link sharing community please link it below.
Unfortunately it's much harder to run a public site these days due to the tsunami of DDoS coming from AI scrapers.
It takes more than 12 hours to spin up a high-quality fork site. If such a site were made, with high quality moderation like we've come to enjoy from pushcx and co, I'd go there immediately.
In other words, what's stopping you?
Thanks, I appreciate these changes and the explanation behind them.
I respect people's desire not to avoid posts that are largely LLM-generated or human-written posts about work that is mainly LLM-generated, but I think "vibecoding" as a tag was too broad, and Fillipo's post was a good example of the tag being used in a "spoonful of sewage" way that always felt more like a witch hunt than a good faith way to help people filter out largely LLM-generated work.
So is there no longer a way for me to browse lobste.rs in a way that gives me high confidence that I won't be reading posts that involve usage of llms?
Communities are defined by who the exclude (by definition, otherwise it’s not a community it’s all of humanity- a community is a sub group that not everyone is part of).
The most important decision a community makes is who to include and who to exclude. I hope this is being done intentionally.
There’s no ‘neutral’ decision in allowing AI posts to fill up the front page and be unfilterable. In doing so you exclude people who want to avoid AI content.
If that’s the decision you want to make go for it but know it’s excluding a subset of people as much as the alternative is.
The last couple years have been the biggest upheaval in the field for decades. While it's acceptable, within limits, to express frustration towards circumstance or technology, it's not acceptable to direct that behavior towards individuals whether they be members of the community or not. Other people can't be your punching bags and Lobsters can't be your boxing ring.
Thanks for sharing this perspective. I consider myself quite skeptical of AI and share many of the ethical concerns raised here. But the negativity directed at certain people and projects has sometimes felt more like a witch hunt. I've increasingly felt that the atmosphere in this community was becoming toxic, and that's not the kind of environment a garden party should foster :).
The changes look good. Keep up the good work and thanks for moderating this garden party.
So there are no way to specify and filter submissions where LLM played a major role? I mean ai-tooling tag looks wrong for recent NIST and Antropic vcs ripoff stories.
Fourth, it was a marker for content which was generated using AI/LLM tools.
Is there no blessed way to mark posts as containing content that was wholly or partially generated by an LLM? This is how I used the "vibecoding" tag. It was useful 75% of the time. Not perfect, but useful.
I find the experience of reading unlabeled, LLM-generated English* to be frustrating. It looks reasonable at first. Complete sentences. Maybe a chirpy tone, but that's how some people write. When someone takes the time to share something they've written themselves, I want to "round up." Maybe English is their second language. Maybe they don't have a CS education and are using non-standard jargon. This style of writing can take a little longer to read, but I want to reflect the effort that the author put into writing it.
When I read something that was LLM-generated, it often leaves me with more questions. There are linguistic quirks, but these quirks often don't form a pattern in the shape of a human. It takes time and effort to read the LLM-generated text before I realise what I'm dealing with. At this point, I can choose whether to stop reading or to continue. If I continue, I will usually skim. Reading LLM-generated English is closer to reading LLM-generated code than reading English written by humans. I want to put less effort into reading LLM-generated English because there was less effort put into producing it. That's fair, isn't it?
The "vibecoding" tag collectivised the effort of discerning whether something was LLM-generated. It let me decide upfront whether and how I want to engage with the post.
* Or any human language, of course.
Fifth, it was being used in competition to the spam flag, for really low-quality content which was generated via AI/LLM tools but without apparent significant human authorship.
I'm not talking about spam.
Spam is the name of the flag for things that were significantly generated by AI. It's considered spam here even if it's not technically spam in the general sense.
Spam is the name of the flag for things that were significantly generated by AI.
No, it's "for really low-quality content which was generated via AI/LLM tools without apparent significant human authorship."
So that excludes:
Those categories of posts should also be tagged "llm-assisted". In my experience, LLM-generated English is difficult to scrutinise regardless of how good the content is or whether it's sandwiched between some manually written paragraphs.
No matter what, there will be people trying to pass off LLM-generated writing as... not LLM-generated. In the subset of cases where we can say "This was LLM-generated because the author admits to it," we should disseminate that information as a signal for people to read or not read or skim.
The actual text on the about page is:
Spam which should be used for content that either is designed to promote a commercial service or for content that is created without meaningful human authorship.
I'd like to request a clarification regarding the ai-tooling tag, if I may. The tag's description is
The design, ux, development of AI/LLM-powered tools themselves
Does this cover discussion of the usage of AI coding tools for purposes other than development of AI tools themselves?
As an example, this new post with the tag isn't about the design or development of AI tools, but includes a section about their usage.
My super-original and unique feedback:
Pending an upcoming deploy, we no longer allow submissions which are code repositories
I'm disappointed by this. I didn't find spending the two minutes looking at a repo to decide if I'm interested more of a burden than looking at an article to decide if I'm interested.
This comment will probably not be seen, but I want to thank you all for your hard work on this site; not just its existence but for the work you put in to keep the "vibe" (see what I did there?)
My interpretation of the new ai-tooling tag is that it should cover 1/2, and what should happen to 3/4/5/6 is unspecified. or I guess, implicitly allowed without tags unless another rule (spam) is explicitly violated.
This is disappointing to me; I'm ok with talking about LLMs, and actively disinterested in talking with LLMs (or engaging with the code they write), and leaving the door open to 3/4 seems like it can only lead to me doing more of the talking with LLMs. (5 falls under the spam rule, and i totally understand not wanting 6)
kay, that's all I have, thanks for your time~
Note also that these changes included expanding the ai tag to include development of LLMs. Hopefully this will be obvious and infrequent enough to ignore such posts quickly.
As I understood it it did before too, and this is just a clarification of the description. At least my impression is posts about development/research of LLM already mostly is tagged ai. It's just that such posts are not that common.
Yes, but the change did explicitly occur, and the redirect to vibecoding (which now should be directed to ai-tooling, or, in other cases, has no specific labelling anymore) was removed.
This is very very welcome. I hope this helps in achieving the goals. Do you think there's a chance of enforcing the sites definition of spam, the overload of me too comments and such things?
I know this is lots of work but I wonder whether the front page on certain days filled with such things leads to people having a wrong assumptions about the goals you mentioned and basically lead to more such submissions increasing the moderation workload even more.
Another idea that might be a bit off topic would be to only allow links or text since the placeholder text seems to often be ignored. This could also allow eg upvoting the story and down voting or even removing a summary comment.
I thought that maybe a content flag tag (among video, pdf etc.) like ai-output would be useful, but I guess with presented reasoning it wouldn't fit.
EDIT: yup, content tag not content flag as gerikson noted
These should be flagged as spam AFAIK.
If it is a purely ai-output sure, but people filtering out vibecoding don't want to see AI assisted stuff as well. Such a tag would be for anything that includes AI output.
Nitpick: "video", "audio" and "pdf" are tags, not flags.
Tags define topicality and can be filtered.
Flags are for off-topic submissions.
Having an "ai-output" tag would enable filtering, but would also signal that such content is topical. I don't think that's what the majority of users want.
This is an impossible situation. Thanks to the mods for trying to respond to it in a way that is measured; it does seem like it is impossible to reconcile the opposing views, and that is a bummer.
It’s very sad to see people whose comments and ideas I’ve valued (whether or not I agree with them, they have enriched my understanding of things) say they’re going to leave or deactivate their accounts. I don’t know how this is gonna play out here, but the garden party was a really great thing to be at, and thanks to all who brought things to share.
I think it is a change for the good. But "vibecoding" tag would still have its use, which are the contents about projects that were mostly coded by LLMs. This doesn't fit to "ai-tooling" tag, yet some people would like to be able to filter this kind of posts. In my experience most of such posts are about vibecoding an application and they don't contain that much interesting technical content. Therefore, I think we need another tag specifically for projects mainly vibecoded, together with the newly introduced "ai-tooling" tag.
Thanks!
Other people can't be your punching bags and Lobsters can't be your boxing ring. That's not acceptable.
Are you going to start removing derailing and personal attacks directed at AI users?
vibecoding tag is now marked as inactive
It covered a lot. For a site focused on the software craft, the gripes about low-effort content and projects are legitimate, and it's important that quality bar exists. There's a lot of blogs I've learned about here who excel at technical depth, and that's exactly what I'm here. Many of those authors even participate here!
I'll probably always be trying to balance LLM adoption to pay the bills and learn quickly, and the fun and confidence of figuring things out and being inspired by people teaching me.
new tag
ai-toolingcovers "The design, ux, development of AI/LLM-powered tools themselves"
This tag is great since I've been trying to get up to speed and learn this.
careers
It's a little sad to see this go. I haven't found anywhere to be a good place to learn about career progression.
I agree with the reasoning but disagree on the methodology. The new ai-tooling tag is too general, and encompasses a number of projects that fall under the umbrella of "a human putting in hard work with their fingers," rather than what people actually want to filter out, which is "content that is not sufficiently technical or thoughtful because of its relationship to generative AI."
For example, I'm working on a machine learning math compiler from scratch in C, but I'm writing it with my own two hands without assistance from AI tooling (my reasoning for this practice is not relevant). Does that fall under the umbrella of ai-tooling? Hopefully not, given the definition of content under the ai tag is "Developing artificial intelligence, machine learning, LLMs," but much of the community is not aware of that definition.
If you attempt to rid the site of a "tag of shame" by lumping in things that AI-cynical people may actually want to read, all you're doing is hurting interesting content that relates to AI in subject matter but not in quality. What you (moderators) are trying to do is separate the quality of the content from its subject matter, but what you are accomplishing in practice is removing the proxy by which we may quickly judge the quality of the content, as unintended and emergent as that mechanism may be.
I see the solution to be to judge content not on its topical nature, but by the things I wrote above: technicality and thoughtfulness. Allow us to publicly judge the quality of the content multi-dimensionally rather than with up/downvotes and "tags of shame." What people actually want is to know before they click on a link whether the thing behind it is worth their time. Technicality and thoughtfulness are much better proxies for that than the "is it related to generative AI" mechanism.
This said, I do not have a constructive perscription as to an effective solution because I am not a qualified community moderator of any kind.
I rather doubt this is the end of the story, but mayyyyybe it's a step in a useful direction?
I'm ready to try this scheme for a couple weeks and then reflect. If you're thinking of leaving, I invite you to try delaying that decision and see whether you do in fact have a bad time here. Granted I have been reading the tags before I click rather than filtering things out of view, so it's not an immediate shock to me, but I'd rather you all stay so I can keep learning from you.