What is your opinion on submission quality since the `vibecoding` tag was removed?
17 points by adriano
17 points by adriano
Were concerns about its removal real, or has its absence moderated content to be higher quality?
Were there any unexpected consequences?
Is anyone willing to admit that their hope for its removal, or fear of its removal was misplaced?
Is lobsters better or worse without it?
It's too early to tell, but personally I've been annoyed a few times opening obviously LLM-assisted articles (even with disclosures in the article) that don't fit the current spam criteria, and I find it unfortunate to see that this change has caused many active contributors to deactive their accounts, but I understand why they did it.
The second post on the front page right now is someone who prompted Codex to jailbreak a Switch, and the article essentially boils down to "Codex said A, Codex then did B", lacking any meaningful details. This is not at all interesting to me, and I would like to filter this out. Of course this wasn't clear from the title of the post.
I like the idea that some people suggested of adding an "llm-assisted" content warning, but ultimately that'd serve the same purpose as the vibecoding tag, so I don't see that happening.
Oh that one is spam, and I marked it as such. (from https://lobste.rs/about)
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
✅ No meaningful human authorship
✅ promote a commercial service (codex)
I fully agree, but 18 hours and 33 flags later it seems the moderators don't.
It did gain an ai-tooling tag, but the post isn't about "the design, ux, development of AI/LLM-powered tools", so I don't think that fits.
Given the quote, the naming of this flag is ... "suboptimal". Let's treat this as a simple ui/ux issue and improve it: https://github.com/lobsters/lobsters/pull/2260
All of the ones I have seen have met the off-topic criteria and I have been flagging them accordingly.
I dunno, feels like nearly everything was tagged with vibecoding before. It didn't really add much in that regard.
Yep, and every entry started with an enormous bike shed thread about whether something was vibecoded or not or whether the tag was correctly applied. I don’t miss that noise so far.
I agree with the others that it's too early to tell, but based on these few days, here's my initial impression, just to start a conversation:
Pro:
Contra:
vibecoding tag) for "hey, this is entirely handmade, come appreciate the art" was a good thing (maybe we could actually have a handmade tag? Name pending, but the idea would be to tag with positive connotations, only for stuff that was made with strictly zero-AI. Of course it'd be a pinky-promise thing, but lying on the internet for fake points is so lame.)[0]: I hope I'm not sounding too callous here. I understand, that to many this is a matter of potentially losing employment and/or a cause for complete burnout, and I sympathize with that. On the other hand, talk about LLMs kept spilling out even into threads that had nothing to do with them and dragging down the mood there too. I'm all for raising awareness and as much as we can resisting the negative externalities of LLMs... but god, I also just want to talk to cool people here and read insightful comments and I was strongly considering taking a break from this site.
While its implementation was non-ideal, having a signal (the absence of the vibecoding tag) for "hey, this is entirely handmade, come appreciate the art" was a good thing
Perhaps it was, but having an insult-coded word for a tag was very definitely not a good thing. I'm glad it's gone, notwithstanding anything else.
(maybe we could actually have a handmade tag? Name pending, but the idea would be to tag with positive connotations, only for stuff that was made with strictly zero-AI. Of course it'd be a pinky-promise thing, but lying on the internet for fake points is so lame.)
I like this idea, but it would have to be special-cased in the codebase somehow. Right now suggested tags mechanism uses the intersection of all suggestions, which means you won't be able to reliably suggest this tag as long as not everyone cares about it.
A lot of cool people left in protest and many who didn't feel disheartened. That sucks and casts a long shadow on all the positive outcomes.
I haven't been paying super close attention so I don't know what the scale of people leaving in protest was, but I'll note that even in the original post there was evidence of cool people that were leaving because of the way the vibecoding tag worked before the announcement as well. (And others in this thread describing how they came close to leaving)
This is true and frankly, I don't think there was ever a chance of a solution that'd make everyone happy. That being said, there were other, less radical options to try (renaming the tag, enforcing topicality more, even splitting the tag into multiple variants) that would've acceptable to a wider range of people and (at least such is my assumption) would have made less of them leave.
The moderators' hopeful suggestion that flagging LLM garbage with spam or off-topic in my opinion not going to work. jailbreaking the switch 2 is currently at +62, 8 off-topic, 15 spam and tagged with ai-tooling even though it has nothing to do with ai-tooling. No amount of tagging LLM garbage with spam will cancel out 62 upvotes keeping it on the front page for days. The vibecoding tag was fine and the shame of being vibecoding-tagged was a feature, not a bug.
I am not certain if that one should be flagged under new rules or not. I checked it with pangram, and it says 100% human authored (it does feel like it, too). Obviously the actual discovery was made with AI but that's a lot of stuff nowadays, for better or worse... I'm mixed on that one.
The new rules say that the flags are for articles with no valuable human contribution, and I am not certain if that's true for this one or not, as it is a human written article talking about an AI-discovered thing (but then, so was the NIST curves article that triggered the last set of discussion about the tag!).
I do agree that the ai-tooling flag is wrong for it though, as the article is more about the usage of AI tooling than the development, which the tag description explicitly says it is not for.
Obviously the actual discovery was made with AI but that's a lot of stuff nowadays
For me this is almost the point, the vibecoding tag meant you could filter Lobsters into non-LLM-related and LLM-related articles. There is now no way to meaningfully avoid content about LLMs, and even the lowest hanging generative fruit that someone's written up about can take up space on your front page.
That post does seem to have been automatically hidden for me, so the flags do seem to have worked for keeping it off the front page.
It's too early to tell but the modlog seems very promising. I think the right stuff is being deleted, both for submissions and comments.
Correct, I'm seeing aggressive deletion of genuine slop. Malicious actors getting the boot (early and often.) And if things slip through, the spam flag was already well-defined "for content that is created without meaningful human authorship."
As others have said: way too early to be sure, but I’m cautiously optimistic.
I was very annoyed with the modteam’s lack of action regarding the discussions around the tag and got pretty close to leaving myself after the nth “the vibecoding tag is broken” thread that felt like it went basically ignored (even though mods said “don’t worry guys, we’re reading all of this!”).
I’ll just straight up say I also didn’t like the announcement post. It didn’t make me feel good. It felt unfocused and I really wasn’t sure where the site was going or exactly what the mods planned to do. But, seeing the recent activity in the modlog has made me feel better about the actual stance of the mods and the kind of activity and attitudes they’re trying to foster on Lobsters.
Overall the site feels like it went from an angry beehive back to a calm one. I don’t feel (as) annoyed when I visit the site anymore because a lot of the junk is being removed by mods and a lot of the annoying rehashing of the same arguments by the same people is (mostly) gone. I also feel like they’re much more quickly removing bad actors and that’s been nice to see.
I have not read a single link from lobsters since.
I simply cannot trust what is in there anymore to have actual information, so... I read the thing, read some comments and then my head reminds me "is it so interesting I should take the risk to lose multiple minutes for mothing?". And I move on.
We will see how much the spam stuff works out. But at this point, right now, Lobsters lost its utility.
Yeah, same. I'm still looking at the lobsters front page but i can't bring myself to click on most links. "Am i going to find out half way through the article that it boils down to 'i asked the robot to build it for me'?"
I think it's still too early to tell.
I'm observing some interesting dynamics but I don't know if they'll hold.
As someone that didn't think removing the tag would do anything, I'm cautiously optimistic. The front page is full of interesting articles and the comments on them seem chill!
I suspect lots of people are steering clear of submitting anything that even tangentially could have gotten the old vibecoding tag. This might also be why there aren't more articles on php2shell and the flood of other security events that went down this week.
Am I using lobste.rs wrong? I don't filter by tags at all, and expect most users don't.
I think, overall, it's better; we gotta move on with the times, it became a way of tainting content as opposed to bringing valuable conversation.
Like others, cautiously optimistic based on the modlog. The most interesting remaining questions are "how much does a post need to use LLMs to warrant a spam flag?" and "what's the line at which mods will actually remove a spam-flagged story?". The switch jailbreak post is a current test case. Even though I'm generally positive about LLMs, IMO that story is pretty dull reading because of its LLM use so I flagged it. Whatever differences remain, I'm hopeful we can find some common ground in "I'm not here to read boring stuff thanks".
What's your/everyone's criteria for flagging vs ignoring?
E.g., I felt the Switch 2 jailbreak article had much more technical detail than the Docker Is Weird article, which currently sits at #1 for some reason, despite being the most cursory overview of long-known Docker details.
Despite that, I would never have flagged the docker article myself, just ignored it, even though I found it a bigger waste of time to read.
I hide submissions when the comments start overflowing into irrelevance. See for example the recent passkeys submission.
I don't have hard and fast flag rules. I do flag corporate press releases and product announcements, especially if they're for a paid service.
I’m confused about why some stories were removed in the mod log for being “about (not) using llms” (e.g. Why I’m No Longer Updating this Repository). Is the idea that non-technical posts hand-wringing about the impact of llms on the craft are no longer considered on topic?
I can understand why Lobste.rs might prefer to leave these stories out, but I’ve read the announcement post on the “ai-tooling” tag a couple times and I don’t see where this policy is mentioned. The topicality section on the about page also doesn’t mention this policy. Am I missing something?
I have noticed several links that mention the existence of LLMs being flagged as spam or off-topic, when I could see no indication that the post was AI generated. Most recently https://lobste.rs/s/nxvfjx/bridge_agent_scale_git_infrastructure
Time will say or heal. I'd keep an eye on the Recent Stories to check the pulse of the new moderation rules.
But, like I noted earlier, stories would get stuck in flag wars, and inevitably, people would end up complaining about it in the comment section.
Currently I still feel that a benignly named tag would've done the job better, and would've lightened the burden of moderation. The new rules could've still been put in force alongside the tag.
I didn't know that it had been removed.
Please, what did I miss? Was there a single (long) conversation before removal, or were there multiple conversations?
A quick search for vibecoding found things such as:
Update: https://lobste.rs/t/meta led to what I wanted.
Tag changes are covered in the modlog:
https://lobste.rs/moderations?moderator=%28All%29&what%5Btags%5D=tags
Announced here: https://lobste.rs/s/h0eate/updates_site_guidelines_ai_llm_tags
I think it's a whole lot better. Now I hope that all the other rules are followed more and breaking them sanctioned as well. To be fair that surprisingly looks a lot better right now as well.