For what it’s worth, I don’t think OP was calling you out, but trying to use it as a reason for why forcing disclosure of AI is wrong.
Let’s keep this on topic. So far there hasn’t been any constructive feedback on what better would look like.
4 Likes
I did not get that from that. If only I had an assistant for that. But I see now, I have supposed that the request for disclosure conveys a sense of hostility from the community which causes one to throw the guards up. There’s really a lot more that can be said, like I know that AI isn’t a real thing and that would come under the purview of philosophy. And for what it’s worth I can program by putting the screws into a bit of wood and wiring it to the power of a lemon. Huh? Tweak on that!
But lets unwind to the first conundrum. I believe it can be restated perhaps more clearly:
We’re pretending AI-assisted programming needs to be confessed or justified, even though we’re clearly going to use it. An experienced programmer can compensate for AI’s weaknesses through decades of knowledge, but we haven’t figured out how beginners can develop that knowledge when AI can do so much of the actual coding for them.
A lot of the remarks here seem concerned with concepts such as learning and becoming better programmers. That isn’t the same goal for everyone. Some of us have reached career pinnacles and have nothing left to prove for passion projects. My goal is to teach people about the universe and learn to appreciate art. Strange. I don’t want to build out a CV. An intermediate goal is to make it easier to do the things I’m destined to do; and that’s where the software knowledge comes in. So take a step back and see that as a community member it isn’t your responsibility to foster a new generation of developers. Also see that the problems we face today we may not face tomorrow.
Let me build my billion dollar Zig machine, and maybe you’ll see that someone already has all of this figured out. WHAT! The wisdom is in letting go and seeing how it plays out. You don’t need to put the screws in yourself. Enjoy the ride and you will find the power of the lemon was never really in the lemon at all. And it was not really a lemon.
continue
1 Like
Honestly, and I’m so sorry for playing the same ‘you’re late’ card that I see all over the internet, but I think this is where the community was in 2025.
I would say that at least in my version of 2026 “we’re all clearly going to use it” is much less clear. I see lots of people who are all over the spectrum both on current use, future use and ideal use. Also about whether AI is a “just at work” tool or not.
I think we’ve seen that experienced programmers can compensate for some of AI’s weaknesses, but that some of the weaknesses are pretty tough to overcome. For example, the syncophantic tone of an agent is starting to grate on some of my most AI gung ho friends and acquaintances.
We also seem (in my professional space as an educator) to be settling on a realization of the value of doing the work yourself as a way of building up the knowledge that experienced programmers (and experts in other AI-exposed fields) have gained.
I agree that the “confessing my LLM sins” aspect sucks. It’s bad vibes no matter how you feel about LLMs. And! at the same time, as a reader of things online I appreciate up-front transparency about something that still is a contributing factor for me about whether something is worth my time. I’d rather the author tell me about their AI use than have to make a guess.
18 Likes
Sure, we haven’t solved “how to use AI such that it is beneficial to beginners”
But we certainly know plenty of ways it shouldn’t be used as we know they are detrimental to beginners. Which has been explained in more detail by others in this thread.
1 Like
One thing I don’t get is the duality of some LLM users.
Both being embarrassed to disclose LLM usage and also that it’s the best way to code.
I don’t mind using LLM for certain things. But also I don’t want to lose time reading LLM written code. If your project is LLM written, fine, I’ll juge it based on its output and performance. If it’s not it’s a different story, I’ll peek inside to learn new ways to write Zig code.
Just be upfront about it. But weirdly even under LLM tag you have people deceiving themselves and others about their LLM usage
So if the policy could be improved it would be by improving its wording to make LLM users less likely to lie or minimize their usage. But I’m not sure how to proceed. Maybe flattery ? “Tell us all about the great models you used, your harness and prompt framework”
18 Likes
I don’t think anyone wants to stop you.
But by the same token, let us just have our forum where we can filter out projects that were written by an LLM.
8 Likes
I agree that the “confessing my LLM sins” aspect sucks
Mine literally was a confession, but it wasn’t a confession that I used AI, but how.
As I wrote my post, I was remembering the moment when I decided to ask claude to write the floodfill. I thought to myself that perlin noise would be difficult to get right so I spent time getting claude to do it properly. After it’d implemented the perlin noise, I didn’t know how to interpret the result so I let the AI generate basically the rest of the program. It wasn’t what I wanted, but it still functioned as a load test so I needed it committed.
So while I’m writing my post I started playing this story back and feeling very strongly that I should have just written it by hand. I’d already have had the algorithm I wanted, I wouldn’t be uploading a library with slop code with the exception of some unit tests, and I’d be smarter. It’s a bit unlucky that the moment I processed frustration over that when I was writing about my project, but that’s life.
And it turned out the feeling was right, when I looked into it perlin was not hard to get into and figuring out how I wanted to express the version of floodfill I had in mind in zig was really fun. Having confidence is something I think I could try more often.
I feel that references to my post in this thread have distorted that story by framing needing to apologize for AI usage as defensive rather than, in my case, self-reflective.I don’t know if the person I’m quoting necessarily intended to do that, or even the OP of this thread, but that’s what I feel it means to juxtapose my post against a call for changing the disclosure policy and I felt the need to set that straight. Nobody decided for me how to write that post.
8 Likes
This part was the key moment of your mistake. LLMs are actually a good way to get into some topic, they can explain from all different angles, they can create different version for you, you can repeat until it clicks without feeling like you are wasting someones time because you still don’t understand.
Obviously if you had a good human teacher, that’s better.
Just don’t submit to accepting LLM code you don’t fully understand and then you can use it just fine. Accepting code you don’t understand was never good, even before AI we were making fun of people copying StackOverflow answers into their projects without understanding, and other ways before that.
8 Likes
I want to echo this sentiment.
I use this forum to learn from real people and discuss and talk to real people. I’ve enjoyed and learned a lot from seeing other people’s projects and how they tackle problems in their way, often it allows me to re-evaluate my own approaches.
When reading AI generated code I don’t learn anything, it has it’s own fingerprint on how it writes code and I’ve seen it a billion times already. The problem might be new to me but how AI written code approaches it just isn’t.
Another reason I like the AI disclosure is that it’s a good indicator for the longevity of a project, not that us humans don’t drop projects left and right, but most AI projects don’t last very long which is where the most interesting part about writing software comes in (how to deal with wrong decisions from the early days, how to write readable and maintainable software).
I personally think all Showcase posts that don’t have the AI disclosure should be closed / locked / removed. Being embarrassed about your use of AI is not a sentiment you should push to your “readers”, being honest leads to better and more fulfilling interactions.
11 Likes
When I started programming, there was an incredible amount of information out on the internet, I could have learned about perlin noise then if I’d wanted to, but I’d never heard of it. There’s been 20 additional years worth of educational resources added to the internet since then.
The issue wasn’t failure to use AI to learn.
3 Likes
At the risk of being a little inflammatory, I think the problem with only requiring disclosure is that basically everyone is going to say “don’t worry, I used LLMs the right way” when disclosing. No one is going to say “yeah, my project is complete slop”.
If you live in this weird middle ground where you accept and recognize the vastly documented terrible consequences that LLMs have on you, the environment and the codebase, and yet conceive of a “right, responsible, ethical” way of using them, that way is always going to coincidentally be the way you use them already. Because deep down it’s all just cognitive dissonance.
To be completely honest, I don’t really understand the incessant need for every single programming space to tolerate LLMs. People into them have literally 99% of the internet to go to and be welcomed in. The rest of us could really appreciate a space where we don’t have to deal with it. Everyone always acts like excluding LLM users will mean the end of the world, but all of us using Zig clearly means it wont.
6 Likes
This has been opposite of my experience. In fact, I’ve mostly stopped frequenting the places I used to talk about my zig stuff because the zig I write nowadays is LLM generated and the places I used to show off projects don’t like that. IRC, ziggit, discord, mastodon. X is the only place where people don’t get flamed for using LLMs, IME.
It’s very disappointing, but I also get it that the influx of slop projects is disappointing on the other side.
10 Likes
Having to moderate and decide whether something is or isn’t LLM stuff all day is simply a huge pita, people on all sides have their opinion whether they are against or for allowing LLMs, or are agnostic to it (or what degrees they find acceptable).
But the moment you decide to take a hard stance you need to invest significant effort to enforce that (and btw. the people who are in favor of that hard enforcement could just as easily (if it was so easy) create their own pure LLM or pure no-LLM spaces, but often they don’t, because that means they would actually have to do the work of moderating these spaces and maintaining them).
My hypothesis is that tolerance with self-disclosure is simply less effort for all involved, everyone gets their space and possibility to participate without anyone having to shoulder the complete burden of creating their own space, moderators get less work through people posting in the expected space, people can select what parts they want to interact with and what they want to spent their time and attention on.
For me the wish you express for a completely LLM free experience in this forum, would simply mean that I would no longer want to be moderator here, because I have better things to do than to curate your completely LLM free reading experience all day, you couldn’t pay me to do that job (it is simpler to wish for an experience then having to be the one that creates it).
With the LLM tag the practical effect for us moderators has been, that the amount of posts where we have to make a value judgement has decreased, simply because enough people follow the rules and tag their post with the tag. This means we only have to look at the posts in detail that don’t follow the rules or don’t fit what is deemed on-topic for this forum, or when there are conflicts about people not agreeing on whether something follows the rules (basically if there is uncertainty about interpreting the rules).
Simply stated I don’t see how any of the extreme positions on the pro/anti LLM spectrum actually have a workable practical implementation that is rooted in reality, that doesn’t hinge on some poor sucker (which in this case would be me and other moderators) that then would have to do to much work to roll a ball up hill and keep it balanced on that point, so that the people with that exclusionary world view can stay in their polarized fairy tale world view (of pretending that LLMs were never invented). If you think you can build your purist space and have it actually remain a working healthy community then please do so and empirically prove me wrong, but don’t expect me to filter the world for you.
I have the opinion that it is better (and long-term more sustainable) to have some lenience, tolerance and openness, towards things you don’t agree with (until the tolerance starts to impede on other people negatively, so it is a balancing act), because otherwise it cuts up the community into ever smaller and increasingly polarized ideological spaces (which then have the risk of becoming culturally dead echo chambers, that no longer support people facing and overcoming their disagreements, or just gaining new perspectives), we shouldn’t define our communities by what separates us, but by what brings us together.
I don’t see the LLM tag as a dividing line between two groups, I see it as a park that has both a basketball court and an area where you are allowed to grill and the rule that people shouldn’t grill in the middle of a basketball game, or play basketball in the middle of a bunch of people grilling and picnicking. It is a way to hold space for people following different activities, such that they can have more positive interactions with another, without having basketballs fling grill forks around and impale somebody.
So I don’t think that a community space needs to solve the big societal problem of where our world should evolve towards, it just needs to serve that communities interest in being able to interact with another without constant high levels of conflict, so basically it needs rules that lead toward less-frequent conflicts. However only allowing one kind of thing, could lead towards stagnation and missing out on valuable new perspectives and personal growth through being open to others perspectives.
I also don’t think that we all should believe, explore and work the exact same way, it is better to have many people exploring and believing different things and being able to still talk to each other, respect and disagree with another, then to form a group that blindly follows some doctrine without questioning it or re-evaluating it from different perspectives. This also includes different community spaces making different tradeoffs between openness and defining their area of focus.
24 Likes
Having just chomped through a bunch of code with AI, I will say this: Form this point on I will have to confess to using AI on practically every project simply because I will 100% use AI on everything related to the infrastructure of the project. Life is too short for me to be bothered learning Yet Another Build System or Yet Another Continuous Integration. Sorry. AI can one-shot that code all day, every day and I mostly don’t give one iota of damn about it other than “Make it work.”
On the core code, I’m almost 100% diametrically opposite. I actually find that I am more productive without AI. I have been optimizing and any time the AI writes a bit too much code, I have to spend a ton of time figuring out the codebase again. I’m better off writing all the code myself so that the mental model in my head is correct and up-to-date. I will sometimes use AI on a spike for the core code to do a quick “Is this possibly feasible?”, but if I don’t immediately throw the code out, I regret it every single time. The AI is good at pinpointing bugs, but I’m much better at applying an architecturally correct fix. The problem is that if I don’t find the bug myself, I don’t have the mental model for what the fix should be. Basically, if I let AI into the core code, I’m about to have a bad time.
Finally, the AI is really good at keeping my unit tests and documentation current. That’s a bit of mixed bag as it’s covering up a personal failing. I really should be doing that work by hand, but I’m irredeemably lazy when it comes to those. So, I wind up tolerating creeping AI slop in those until it offends me enough that I take a hatchet to them and fix it properly.
9 Likes
This is really the crux of it.
Whether you are an experienced developer with 35 years experience and actually do 100% know how to use the AI as a powerful assistant, or you are a first year CS student who isn’t even aware of the the depth of the knowledge you lack, both fully believe that they aren’t one of those peasants who would use AI irresponsibly.
This leaves anyone being presented with the code with the same problem: which type of person is this? While a person who is known to the community and has already “proven” themselves and earned their stripes, it isn’t a problem to be unconcerned about the use of an LLM, but when someone who is not known or making a debut post to show off their project, the “trust me, bro” explanation has been abused to the point of being meaningless, which I am sure does not feel very welcoming.
2 Likes
I am sorry you’ve had a bad experience. For what I’ve been through though, only Mastodon really takes a hard-line stance against LLMs, and only some parts of it anyway. Reddit is full of LLM-accepting policies, and most Discord programming servers too. I haven’t been on IRC for a while, though.
I don’t fully see why being LLM-free requires such an intense moderation job. Much like with Zig and CodeBerg themselves, and much like for most other moderation rules, it’s more about drawing a clear line in the sand and signalling to those people that they should go somewhere else than strictly vetting every single thing that is presented.
If someone posts LLM-generated content without disclosure (which I should say is a hypothetical I only entertain when people sealion me, I’ve never seen it actually play in real life enough to be a problem), then either:
- Someone sees it, notices it’s obvious slop, reports it. Job done.
- No one sees it enough to report it, in which case, why bother if it’s a post no one cares about anyway?
- Hypothetically someone sees it, notices it’s slop, and doesn’t report it anyway. It’s either going to eventually be seen by someone who does, or it’s going to get buried.
- It’s slop, but it’s hidden so astronomically well that no one ever notices, and the author never mentions it at all, ever. For this to happen the codebase and thread would both need to have a very high level of quality, so I still consider this case a done job.
Furthermore, I am trying to say that the disclosure rule in theory requires the same work anyway, since people will always say their projects used it responsibly and awesomely, while they could very well be full slop in the background. You are just more willing to take people who confess to their LLM usage in good will rather than people who say they don’t use LLMs at all. You’re already biased towards them, and these things are what end up causing the “community split” you talk about in the end, because LLMs just inherently erode trust.
Communities are social spaces, and people having social spaces where they can relate to each other is important. There is no point in me being in a space where no one actually shares any values with me or even seems to care about the same things as me. Not everyone has to always come together, sometimes people are different, and need to be in different spaces, and that’s fine. My problem is, LLM-bros are the ones who can’t seem to accept this. They don’t want to concede a single inch. Everywhere has to welcome them, or it’s bad.
Why stop at LLMs then? I’m sure there’s lots of bigots, extremists and fascists that enjoy Zig all the same. This big-tent logic only sounds nice in theory, in practice it’s always been proven to be impossible. If the only criteria for someone to be in a Zig community is “they must like Zig, and also not be too much of an asshole sometimes” what ends up happening is a lot of vulnerable people leave silently because they are uncomfortable with the kinds of people that openly hate them but are allowed to stay in the community because they’re not “rude” about it. I’ve seen it happen in Odin. You are always taking a stance on these topics whether you’d like to or not, consciously or not.
My point wasn’t that Ziggit should “solve LLMs”, my point is that there is no reason to be so lenient towards them when they clearly go against the stated goals of both the community and the programming language it’s about. I repeat, they have 99% of the rest of the internet to turn to, they may feel many things but “oh no, I have nowhere to go now” wont be one of them. And in the meantime, a better community can foster.
2 Likes
I can only reenforce my suggestion to make a whole LLM/AI category. If the tag is working well for the mods, a category should be possible to handle very similar. With the exception that its not mainly limited to showcases like the tag.
TBH I would also welcome a decision to make Ziggit a fully LLM-free space, but don’t think thats going to happen. Thus, a category seems more clear than a tag. The latter can be omitted, but if you post in a category its automatically tagged. And posts that went into a different category as false negative can simply be moved to the right one.
Regarding this subtopic:
I think if someone feels embarrassed about the disclosure of LLM usage, they already know/feel that there is something wrong with using it (and there are many arguments why LLM’s, as they are today, are harmful and wrong in my eyes). There are many spaces also related to Zig which have no LLM-policy were anybody can showcase their stuff without that embarrassment.
Why not simply embrace that Ziggit could be a less-LLM-friendly space. There are thousands of LLM-welcoming spaces out there already. There might be some truth in this words by @zwynd:
However, since a totally LLM-free Ziggit is unlikely and also maybe not wanted by many members, I would prefer an explicit category over a tag.
1 Like
Here is one datapoint. This is how I use LLMs. I take c++ libraries that I like but do not understand well because they are written with the opaque c++ style. I ask the LLM to translate them in idiomatic zig, optimised for understanding. Now it compiles and runs faster and I can finaly understand it perfectly. I verified with millions of tests that there are no bugs and it has the same behavior as the original code. I love zig and I dont write code by hand anymore. My job is now to build systems and understand them. And try to invent new things. Zig is the easiest language to read for me, so this is where I landed. I know it can be all described as slop but I stopped being ashamed of it in the last few weeks. I have transpiled to zig raylib, imgui, pytorch, various compilers and interpreters, game engines, graphics, physics, robotics libraries, etc. In the past I published papers in physics and robotics, so I already know the subjects, and I like to think that I am an ok programmer. Now, students should absolutely code by hand for the first 15 years to also become an ok programmer. This was written by me, and I agree that posts should be written by humans. Online forums are about humans talking to each others. I am still racist against AI, which is also called humanism, which is philosophically sound. I think Andrew Kelley should continue what he is doing, with the hard stance against AI, because this works for him and he really needs to understand very deeply everything. Maybe I am wrong. But on my side I dont think I will write code by hand again and I made peace with this fact.
I’m curious, what would go into that category? Anything where the user touched LLMs during their work? To me, that seems like you wanted to categorize community by which editor they use for editing the source code, Helix users discuss things here, Vim users here, and VS Code users, we would rather not see them here, but if they have to be here, we will do a category for them as well. I’m sorry for being silly, but that’s the analogy I see.
When I started programming, people were questioning whether IDE with autocomplete is good for you and that maybe you should get into the habit of typing everything yourself. Later, in 2000 and something, people were amazed that I could solve problems by typing a few keywords into Google and actually get an answer. “Prompt engineering” before anybody even dreamed about LLMs. So while I see moral issues with how the AI companies are doing their business and pushing the product to everybody, I don’t see categorization by which tool did I use as relevant. I can see the value of the disclosure, to help the people who have the moral issues with it, and I will assist them, but I think the tag serves the purpose.
I think both AI tips and tricks and anti-AI politics, can be discussed elsewhere and Ziggit should stay a mainly technical forum.
3 Likes