返回资讯中心
外部精选
软件工程
#开发者实时流

Lobster moderator's updates to site guidelines and AI/LLM tags

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 comp…

Hacker News 最新droidjj2026/10/19 · 2026/10/1930 分钟阅读

以下正文同步自 Hacker News 最新,版权归原站所有,已转换为易读排版。

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:

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.)

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!

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.

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.

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.

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.

No, see my first link. There are many issues with flagging slop as simply "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,

正文由 探How 从来源站点 RSS 同步,内容未经改写;遇到排版缺失或需要图片、视频时请以原文为准。

打开原文
去社区讨论 →