My suggestion of a llm category is based on my comment from above:
User @g41797 raised the concern that embarrassment regarding LLM disclosure in showcase posts could hinder further discussing over topics regarding Zig and LLMs. Thats why I suggested a whole category, so people could also discuss other aspects of LLM usage, while other users could still mute this. (However, I might understood the worries from g41797 wrong, since I’m no English speaker and didn’t use translation tools)
This might not be the perfect solution, it was just my thoughts. It can also stay at it is…
Ah, so meta category, with the tag still existing for showcase posts. I’m not sure I like it. I use AI agents fairly regularly, but I wouldn’t want to poison Ziggit with these kinds of topics. It should stay about Zig, in my opinion. And there is the one thread where people can post, if they really want meta discussion.
I’m not sure I like it either . As I stated several times, personally, I’m very anti-LLM/AI (due to its environmental and political impacts, not the technology in general; and not against the users as individuals). However, there should still be a place for LLM-users to discuss those things here. In the end, its up to the mods to decide. I only made this suggestion as an alternative (good or bad) and will stay away from further comments on that
I agree that LLM-users should have a place, but I don’t like the binary “LLM-users” vs “non-users”. One of my biggest frustrations when putting the LLM tag on my post is I’d promised myself not to use AI and still did. Internally, this was not a leap of going from being non-LLM user to LLM user, it was failing to be a non-user. As of now, I’ve now rewritten I think all of the code AI touched in my codebase, but I still think the LLM tag is correct because the git history, if nothing else, still has AI, but I don’t consider myself an AI user presently.
If one thing does bother me about the tag, it’s that people will look at projects through this binary lense–though that’s not the tag’s fault.
I’ve also seen examples of LLM disclosures where it wasn’t even code or help with debugging, just translating or document writing for the benefit of people that do check out their repo despite the LLM tag. I don’t consider such people “LLM users”, but people with strengths and weaknesses and out of a desire to enable themselves to share their work, used the tool that was available.We can debate over whether it was necessary or good for them to do that, but I think “AI users” is too broad a category if it can contain both vibe coders and someone unconfident in their communication skills generating a README.md.
I get what you mean and of course if you consider all the aspects you mentioned you’re right. “LLM-user” is a over-generalization, as very likely is “non-LLM-users”. As always with such ideal example types, almost none matches it exactly, its more for visualzing patterns[1].
What would fit the type of programmers I had in mind better might be “regular/heavy-LLM-users”. But that term too is up for discussion…
On the one hand, I’m disappointed that people like @rockorager - from whom have seen some real contributions - that people like him are feeling beat up just for talking about using AI.
On the other hand, everywhere I go on the internet (excluding Mastodon) there are *constant* *floods* of these stories: “I made a brand new thing that I couldn’t possibly have made without AI - I haven’t tested it or finished it but I am amazed at how quickly I started it!”.
I just want to mute the flood.
I believe that the bubble will burst in time, we’ll find out it costs *lots* more than we believed, and after that we’ll find out which parts are worth paying for and which parts were nothing more than social-media advertisements.
I don’t believe we can change anybody’s minds, though. There are a lot of people who are *in* and they’re not getting out until somebody hands them a bill they won’t pay. From my side, I will not be buying in on the value side until after the cost side is visible.
I don’t enjoy endless debates about it, I don’t want to beat anybody up, I don’t want an alternate version of the tag for “LLM with high standards” that I would just also mute, I just want to mute the flood. The tag is working, and it’s not been painful to use the Mute options I’ve seen in this thread and elsewhere. I don’t feel alone in saying that if we had fewer tools I’d leave y’all to it and I’d settle for discussing Zig on the fediverse. I’m not sure what more tools would look like, though, other than dividing whole categories, and I see why that wasn’t chosen (and to be honest, I think most people who needed *more* tools have already left).
Edit: if you had another tag for ‘llm-with-high-standards’ (I don’t have a clue how you’d define it but maybe that isn’t a problem?) I can say it wouldn’t bother me in the slightest once I found the mute button.
You can dig your head in the sand, you can treat LLMs as a completely morally neutral thing that exists in a vacuum without any externalities or consequences, you can pretend like the reason people are against them is that they like, dislike that you code faster with it or something, but I beg of you to not pretend others should do that as well. In the real world LLMs are very very obviously not comparable to “using another editor” or “the IDE showing you a list of the functions that start with foo”.
I think it’s fine to call a project LLM-free if there is no LLM code in it currently, assuming you have no intentions to add more LLM code down the line. It’s not like using AI taints you like you’re some kind of zombie forever, you are allowed to have changed your mind.
It is a binary for most anti-LLM folks though, because again, the problem is not that LLMs make the code worse, even if they didn’t, I still would prefer to not have anything to do with a project that’s okay with stuff like that. Nowadays it’s impossible to do that with everything, but I’m not gonna stop trying.
I agree with this, I think there is also some sort of nuance to be had between straight up vibecoding and code generation, and using LLMs for research or things on the side like that.
Having one LLM tag puts the vibecoding people in the same boat as the research/“LLMs as search engine” people
I oversimplified before. I considered making the code LLM-free before uploading, but felt that it wouldn’t have changed the need to disclosure to using AI for debugging, planning. and research, which I now consider even that to be too much AI. I was having dopamine abuse issues with AI before and I feel like I still had dopamine-chasing habits while working on the project.
What if using AI to find bugs discouraged me from writing more unit tests? I know I used some deprecated functions because AI informed me about them. AI doesn’t have to write code to influence a project negatively, there are subtler ways.
So I’ll remove the tag after I’ve been working on it for a while, totally without AI.
And then it remains the responsibility of the author of any such posts to be a good (or bad, or indifferent) ambassador for whatever tools they are using - whether that be LLMs, or Zig, or Vim/Emacs … etc. Thats entirely sensible.
This was one of my suggestions when the policy was first being discussed: utilizing multiple tags that could provide some rough detail on how an LLM was used. Sure, the disclosure section makes this easy to explain, but this isn’t helpful as it currently is when many users simply filter every post with that tag.
For example, I personally don’t care if someone uses an LLM for planning, finding bugs, or verbose boilerplate, etc, and explain that within the post. I don’t filter the tag for this reason, even though I am generally adversarial to LLM use, because I don’t want to miss out on a cool project just because someone used an AI to generate the rough structure of their README or something else I consider inconsequential.
It’s fine to use LLMs to answer questions, analyze, distill, refine, check, suggest, review. But not to create.
If you did not used LLM to create, I don’t think it warrant the LLM tag.
OTOH even if the code is straight clanked slop, the Readme is really the one thing you should write yourself. If you can’t write a Readme, how can you write a prompt?
. As I stated several times, personally, I’m very anti-LLM/AI (due to its environmental and political impacts, not the technology in general; and not against the users as individuals). However, there should still be a place for LLM-users to discuss those things here. In the end, its up to the mods to decide. I only made this suggestion as an alternative (good or bad) and will stay away from further comments on that