Based on all existing viewpoints, a simple understanding of LLM is: although LLM may be helpful from various practical perspectives, its generated results are not suitable for human reading.
Whether it’s code, documents, or ideas, LLM’s output is harmful from the perspective of readability for humans.
On forums, you can make use of LLM’s assistance, but do not publish LLM’s content, because LLM’s output is unreadable. If you attempt to display LLM-based projects, remind reviewers which parts come from LLM—because those parts are not suitable to read.
I use AI for work daily and I do not enjoy it one bit. The fun part of the job I delegate to agents, then I get stuck with the part I do not like (reviewing their code). I don’t like AI, I don’t like AI CEO’s and all they stand for. I legit come to this community to escape all of that after my 9-5. However, I understand that some people are bullish about this and are part of this community, and that’s fine. I personally would like some kind of disclosure so I can avoid it as I am not interested in it. But the right to participate by them is fair and some kind of topic for it would be ideal.
20 Likes
I don’t think that “flagging” content made with LLMs is that important, because it is not a real problem for the reader, but it is for the one writing the code.
I have tried coding with agents for quite some time and they can be helpful by generating
stuff that “works”. This is awesome at first, but it is a slot (or should I say slop lol) machine.
I’ve felt my brain rewire in a really bad way, it’s like TikTok but 10x worse.
You start to delegate every kind of thinking to the LLM and your brain basically goes numb.
I always felt fatigue during this testing period and I wasn’t able to do anything in my codebase without assistance, even if I’m an experienced programmer.
I think that this kind of coding can become a real addiction after some extensive usage, this scares me a lot and I decided to not use LLMs.
This is my own experience but I think I’m not alone feeling this. I encourage people to not use LLMs, not for the craft or passion for coding (they’re also important, don’t get me wrong) but for their mental health and ability to think without a machine saying that they’re absolutely right.
4 Likes
It’s an emotional topic for me, so my opinion may be seen as extreme, but I’ve reached a point that I’ll avoid any product or service that involves the funding of the race towards AGI. In the past couple of years I’ve dropped Android, Google, Apple, HN, and down to just a single (TODO) spam account with MS. Yes, it makes my life harder, but these freedoms we enjoy didnt come cheap.
LLMs are currently funding that race with obscene amounts of money whilst the planet suffers and burns. So despite the fact that we seem to have crossed a threshold, and suddenly find some of these tools quite powerful, I’d much prefer to live in a world without them.
I compare it to the artistry of creating a piece of wooden furniture. I can lovingly carve it by hand via lots of hard work, or maybe use a lathe to help, but once I can say “Sudo, make me a chair”, then what is the point? Where is the human value in that product, the artistry, the love? It’s a piece of slop furniture.
So for me the term slop is naming something machine generated with minimal human input.
My vote would be that submitting to showcase includes a few questions regarding use of LLMs, such that can I filter out that content.
I’d go as far as to advocate the licensing of Zig for use by humans only. What a statement to the BigTech world that would be 
25 Likes
That’s fine. It’s not about advertising slop, it’s about disclosing some reasonable amount of AI influence so that some can turn on a discourse filter and feel like they’re not “wasting their time”. We don’t want to discourage such people from participating at all - they have a perfectly valid, arguable point, and can take their time elsewhere, but appreciate when some flag for courtesy.
Not a rule. A courtesy.
I disagree. I think some readers think it’s a problem having to constantly have their AI radar on, and would prefer the courtesy of flagging.
3 Likes
My vote would be that submitting to showcase includes a few questions regarding use of LLMs, such that can I filter out that content.
Yes. I’d love if including a “no llm” or “llm” tag was required at least for the “Showcase” and “Media” categories was a must-have.
I’d go as far as to advocate the licensing of Zig for use by humans only. What a statement to the BigTech world that would be
That would be fun actually. There aren’t that many licensing problems (in theory, Zig can just relicense from MIT to anywhere as long as the copyright is kept AFAIK).
4 Likes
Interesting, and surely attractive to many, but obviously impossible and impractical. Now, if there were some reason to need this at some point for some reason, it could be nice if it were possible and practical, but…..
I think that it’s really easy to spot LLM generated code, so I do not mind flagging.
But I get your pov, making sure that something is not LLM generated can be annoying for someone.
I really like this, I think it’s more scalable than a hard rule.
2 Likes
Our inability to concretely articulate the dividing line between “this is an acceptable use of an LLM” and “this is slop” should not mean we don’t at least attempt to create some distinctions (i.e. Loki’s Wager /continuum fallacy).
If there is a spectrum, what about disclosing it on a spectrum? Maybe it could be reduced down to a short “I used LLM for the following…” checklist that gets included in showcases instead of a binary tag. This could provide a general overview for people to manage their expectations.
Though I would still prefer for it be disclosed, I do not personally care if someone used an LLM to do some refactoring, asked it to check their code for bugs, generate some tests, and other trivial(-ish) tasks.
I do care if they vibe-coded a project.
Surely we can find a way to differentiate these without lumping them all together under the same generic tag/category? I don’t think it takes an oracle to predict the negative effect this will have on the posts that use such a tag/category and the attention that they get.
This would then likely lead people rationalizing to themselves that their use of an LLM didn’t reach this subjective threshold of deserving the tag. We will then be right back here philosophizing on when the tag should be used and when it shouldn’t, but never coming to a concrete decision.
I would prefer that users just disclose themselves generally what they used an LLM or whether they did at all. I trust we are capable of discerning what is acceptable and what is not on a personal level. They can be more detailed if they want, but a general explanation alone would be helpful.
If it is decided to enforce the use of an LLM classifications, it might be worth considering adding additional tags/categories that are more specific than just a yes/no. Generating a README is not equivalent to vibe-coding, and it doesn’t feel fair to classify them together, even if I personally wish people would just write their own README.
3 Likes
It sounds like most of the issues people have are from Showcases coming from low-trust users: users who are unwilling or unable to understand their own code, drive-by submissions that over promise.
Perhaps the Showcase category can be restricted to higher trust score members using one of Discourse’s builtin mechanisms?
11 Likes
Hi everyone, I want to say that I’m the one that suggested that an AI tag, would be a good idea. The reason is that this is a community, and we ought to respect each others, now some people (including me) are fine with AI usage, some are not and don’t want to invest time in something that wasn’t fully made by another human.
Nobody is right or wrong here, it’s 2 different approach. So i think it would be a fair compromise if people who really don’t like AI usage, refrained themselves from calling everything that uses AI “Slop” and if people who are using AI were willing to disclose the usage voluntarily.
As @mnemnion said Ziggit is for humans first, we need to be considerate of the diversity of opinions, experiences, and practice, I think disclosing AI usage is the right thing to do, and I don’t think it would stain the effort that went into making something. As long as one is willing to stand by what he/she made than I don’t think this would bear any kind of judgment or negativity.
And as many have said it would also help those that definitely don’t want to see any Ai related code or topic and keep ziggit a welcoming place for them too.
7 Likes
You definitely got my vote, although its unrealistic 
Indeed, Zigs move to Codeberg, literally giving Github/MS including Copilot the finger, was the main reason for me switching to it as my go-to language (beside all the beatuy/readability etc)
1 Like
This is a great point. There is a notable increase in the:
This is the first time RandomNewUser has posted — let’s welcome them to our community!
Zinux
I just started using Zig yesterday, but wanted to introduce my port of the Linux kernel that is 300% faster!
…but on the other hand, I have seen new users showcase high-quality and hand-crafted projects, and simply wanted to show them off and share with the Zigit community.
3 Likes
Yes, the slop issue. For us it’s primary, that’s why we decided to start this thread. We’re going to do something about it.
This was considered, but a Showcase is a very popular debut post. If it’s a good one, it’s a great way to show up. We don’t want to close the door on that.
9 Likes
Definitely! Mostly I just want to thank the mods for their nice work and for talking about stuff like this both internally and with us. I think the forum is in a pretty good place because of you 
5 Likes
If a program I use is created by generating code with an LLM, that’s fine, but at that point I don’t care about the code, the architecture, or what language the code is written in. I’m here for Zig code discussions, and I don’t want to spend extra time figuring out if the code is worth looking at (based on my own subjective values). I want LLM usage disclosures so that I can make my own judgements about whether I’m interested in the code - from “README/documentation created largely by an LLM” to “I used LLM autocomplete while working on this code” to “Much of this code was created by directing LLM agents.”
Without a disclosure, people who don’t want to waste time on LLM-generated code feel tricked when we get part-way into reviewing a project and find that it is likely LLM output. A disclosure is the best way to enable users to choose what projects they want to engage with, without causing conflicts and negativity.
14 Likes
A lot of good discussion here, all of which we’re reading.
This is a good touch-point for where I’m at with labeling, in general:
You’re right, and this is not a good position to be in as a mod.
We’re not signing up to rigorously inspect every Showcase for traces of LLM use. For a codebase, like Zig? Rigorous inspection is table stakes, and code is selective: by default, it doesn’t get in. We don’t have the ability, or the resources, or the motive, to confirm that Showcases follow some disclosure policy: that is, that whatever level of LLM use they admit to, is exactly the amount they engaged in.
Deciding Slop-or-not, we can do that. It won’t be perfect, it won’t satisfy those who mean “any LLM output” when they use the word, but it’ll be nicer for all of us.
Disclosure as policy is asking us to do something we actually can’t do. It is difficult to merely encourage something in official voice and not have it read as policy.
I think you have low false positives and high false negatives.
There’s also a moving target problem: this is, at best, like saying “it’s easy to spot generative art, they get the fingers wrong”. True until it wasn’t.
I wonder if this is the wrong angle. Say you have a flea market, and some of the customers strongly prefer hand-sown clothing. So a regulation is passed: all factory-made clothing has to set up a sign which says FACTORY MADE.
Now when the customers who are looking for the handmade clothes come to the flea market, it’s just: FACTORY MADE, FACTORY MADE, all over the place. Finally spot a table which doesn’t have the sign, but it still kinda looks factory made, and they ask and “oh I edited the post, sorry”. This is not a fun experience.
Meanwhile there are a couple unlabeled tables, with their hand sewn clothing, waiting for their customers to figure that out.
This is obviously backward, you want the signs to say HANDMADE. Even customers who care more about quality than means of production don’t find hand-sown kit objectionable, and may even prefer it in some ways. But they don’t care much about the signs to begin with: and the ones who do care, are getting the signs slapped on the product they do not want.
I inserted an analogy into a post which didn’t really need it, because I was concerned that if I just came out and said “what if we encourage disclosure of non-LLM assisted code” you might feel trolled. “Why should we have to, when they’re the ones—”
There’s a mixture going on in this conversation, of a sincere desire to know what’s going on, not having to guess or be exposed to something you don’t like: and making the heretics wear the funny hat.
Well, we don’t really want to make the heretics wear the funny hat, and if that’s the main thing you want out of this, I feel you will end up disappointed.
I think if we were to encourage the use of a handmade tag on LLM-free Showcases, we’d get good compliance, basically no fakers, and you’d know what you want to know.
It’s also easy to define: a human wrote all the code. all of it. Consulting a chatbot is allowed but, when handmade is used, must be disclosed.
To contextualize this post, I’m not coming out of a mod huddle here: everyone is seeing this at the same time, it’s a notion I had which seemed worth introducing into the mix.
17 Likes
It was mitchellh’s post that encouraged me to finally install an llm and give it a shot. From there it was a short hop to writing my first few lines of zig. Personally I’m enjoying the llm era so far, at least from a coding perspective. I’m using it for both work and hobby projects.
Unfortunately, the coding communities on reddit are struggling right now. They are completely inundated with slop and people are frustrated. It’s actually a double whammy - anyone who posts has to suffer through both the negativity bias of reddit and then a second wave of people who are convinced everything is slop. HN has the same problem, even antirez was accused of being “ai slop” a couple of days ago. Comical.
Ziggit could be nicely insulated from this sort of thing. Vibe coded projects will largely be quarantined to Showcase, right? Where else is slop a problem?
FYI on my flight yesterday I built a zig library to detect ansi16/256/16m support for the current terminal. This is a non-trivial problem. The first thing I did was feed a dozen existing go/rust/ts libraries into the llm and ask it for analysis. Then quesionts like “does anyone still use that TERM” or “when was that version of windows released”. Does this count as vide coding? These tools are supremely useful in the hands of a senior person with domain expertise and decent taste.
3 Likes
This is the juice of it. AI code will possibly improve, and make everything heartless. We’ll even regret the times when AI code was crap. Already people write music and novels with AI, but who’s going to listen/read/love them? Nobody. Already I don’t trust anything anymore that appears on a screen, it could be 100% fake. The world we live in took millions of years to evolve, and we as a species hundreds of thousands of those, still we’re being thrown against our will in a future where everything must move at the speed of light or not exist. As a result, it will do both things. There’s no need for such an acceleration, it will destroy our faith in the value of doing anything. So why are we doing it? Is anything or anybody threatening our existence if we don’t move this fast? Instead, moving this fast threatens it, and the rest of this world with us.
At this point I think all code should be considered partly or entirely AI-written, unless there is a very clear disclaimer that it isn’t, not even in a small part, and only if that claim can be actually believed. And how can it be believed?
For my part, I did my share of vibe coding the last months (privately) but I enjoyed more the time I spent clearing out a field and planting potatoes.
4 Likes
The problem is that everyone who has ever written a “Hello, world” thinks they are this person. We can’t all be a Mitchell Hashimoto. He is an exception, not the rule.
6 Likes