Mod repository should have more explicit AI tagging, ideally with two tier tags, one for human made mods supported with AI and one for purely vibe coded stuff. I'm not against using AI when creating content, at least as long as you use it as a tool, not as a replacement of your brain (controversial, I know). But the number of low effort vibe coded apps with preview images made with ChatGPT has been growing fast in the recent months and inarguably many of them count as slop. It would be nice to be able to filter them out.
I heavily agree...17/20 uploads on the first page of the UI apps section are clearly AI-generated or AI-assisted. If this type of mod permitted and approved, it should at least be marked.
yeah, i used to check out the repo every day and its getting tiring, its all just blatantly AI slop, at this point modland is gonna be the higher quality repository...
mods approved today: https://www.beamng.com/resources/rl...th-rls-career-overhaul-v2-7-0_hotfix_2.39059/ probably AI, it has a very AI looking UI on the images, and the description is 100% ai https://www.beamng.com/resources/hrc_télémétry_v1.39140/ this UI looks pretty AI, the description maybe, theres no reason to include install steps so i think thats ai https://www.beamng.com/resources/hrc-drift-v1.39142/ "made" by the same guy, description is very similar, ui too, probably also AI https://www.beamng.com/resources/wip-kagoni’s-tyre-simulation.39079/ AI generated thumbnail, AI generated images, absurdly long description, and apparently, under the hood its just sloppified luuks tire thermals mod https://www.beamng.com/resources/rls-stock-exchange.39139/ AI thumbnail, AI images, ui also looks AI, description also looks AI generated https://www.beamng.com/resources/cop-pursuit-controller.38762/ AI thumbnail, ui also looks AI https://www.beamng.com/resources/rainfx.39126/ Apparently doesnt even work, the most AI looking description ever https://www.beamng.com/resources/tire-wear-and-thermals-respin-—-compat-tires.39083/ https://www.beamng.com/resources/tire-wear-and-thermals-respin.39082/ AI thumbnail, ai description and apparently, under the hood its just sloppified luuks tire thermals mod out of 26 mods uploaded today, 9 of them are Obviously AI, and some others im pretty sure are AI but theres no obvious tells --- Post updated --- the reason theres no "ai vehicle mods" is because it cant, but beamng uses lua and cef/web development for UI/code/vehicle systems... and ai can do that, which is why theres a gazillion ai generated UI apps popping up every day now
Also, for some of these mods; Doesn't GPT primarily use EN-US and not EN-UK? You'd be seeing tyre less, no?
depends on the user, ai model and a bunch of other things --- Post updated --- yeah, most modders especially new ones that are starting out always make a "please help im dying" post lol, or at least have a pfp and use the forums to comment on something once in a while
https://www.beamng.com/resources/roadwave-—-in-car-music-player-with-3d-audio.39115/ this one even has the fucking EM dash on the title :sob: and its not even correctly formatted, the description is using **text** to make text bold, which is what markdown uses, not what the beamng forums uses... they didnt even check if the ai description was correct.........
Akchually, you can get Claude Code to do jbeam quite reliably, just the effort is not worth it. The catch is that you still need to know what you're doing. Good idea, if you can't make a single preview image for your mod then it should not go on the repo lol. Spoiler: How LLMs work and why it's so easy to make UI apps with them I've been working with AI since I got first access to GPT-3 (around 2021), and the way LLMs work hasn't changed much since then. To make an LLM, basically, you take a huge corpus of cleaned up texts and then let the machine figure out its way around them with little to no direction at first. This is why ChatGPT can write in virtually any language – all you need is to throw enough examples and it will figure out that B comes after A. This is also why LLMs are so good at common coding languages. HTML, python, lua, Java, etc. are very common, and more importantly, formulaic, so as long as you point it in the right direction and give it enough context to work on your LLM will spit out something useful. Ask it to write a mission script for GTA San Andreas using its proprietary format and it will probably generate gibberish that will crash your game. There's another thing that current tools are a bit more sophisticated, i.e. Claude Code can write automated tests so that it can check the lua code it writes is correct, or ChatGPT can remember user's preferences so it uses British English rather than American.
Fair callout. Give me some time to clean up things. Seriously, not being sarcastic about this. This is the first time I have done modding at all, and I like the blunt feedback. It let me know what needs to be corrected. Thanks again.
It sucks to see that most new BeamNG mods are low-effort and lazy AI slop who don’t even put effort on their mod description. I honestly feel like BeamNG mods that heavily relied on AI need to be removed from the repo entirely at some point…
I definitely agree that there should be some kind of AI tag, especially for heavily AI-generated/vibe-coded mods. But I don't think AI assistance itself should automatically make a mod “slop.” I've been using AI to help with BeamNG modding, mainly for things like Lua/JBeam development, and I've found it can be genuinely useful. The important part is that you still need to understand what it's generating, test it in-game, fix the problems it creates, and actually put your own work into the project. IMO the better distinction would be something like “AI-assisted” vs. “AI-generated/vibe-coded,” rather than treating everything that used AI as the same thing. That would also let people filter out the stuff they don't want without discouraging people who are legitimately using AI as a development tool.
it's funny how AI slop mods get approved with little to no resistance, yet if i upload any of my mods it's getting rejected
Of course, that's why I suggested two tags there. Just using AI as a tool doesn't make a mod bad, having AI make mods for you does. If you research something and have AI just write lua for you then it's not ideal, but isn't hurting anyone either. If you ask ChatGPT to download a tire physics mod and do something better, then you deserve an AI slop badge. I can imagine there are people who want to religiously avoid AI, the other tag is for them.
well if your mod doesnt meet the few rules there are for the repo then it doesnt get approved, ive been told by a repo mod that people submit tons ai generated mods that just dont work, as in, the ai hallucinates jbeam and the person submitting it doesnt even check if the mod works
See, I’m a tiny bit conflicted on my views on what to do with usage of ai as a whole, I’m generally against AI, but I’m absolutely in agreement that an AI tag needs to be made. I feel like there are some extremely rare situations where if used properly as a TOOL, and not taking the bulk of the work (95% human made, maybe assisted a bit) then there’s are use, HOWEVER, AND I CANNOT STRESS THIS ENOUGH, we as a society have also found that AI is has quickly become a net negative for us. Data centers are just awful, and are bad for our environment, most of what generative ai puts out is soulless slop, and in cases like beam, it just takes the life out of a game which is extremely human by nature. I gee like those who want to avoid ai at any costs should have that option to do so, I certainly would like it. There’s just something about AI mods for beam that feels wrong to me. I’m not sure what it is, but it takes away that magic. Anyways, that’s my two cents on this
