Idk where to put this, some of it is quite technical, but others is more personal. So apologies in advance.
For reference, I started thinking about this after reading the post “Big Walk Fixed a Bug With AI” in the games community.
For full disclosure, I used/created programs around AI before LLMs were even a thing, back when it was called “deep learning” or “machine learning”, and I have used and taught classes on Large Language Models (specifically local tools and open weight models not the proprietary ones), I’ve written a toy version of GPT2 using TensorFlow. The whole situation around AI sucks now. It was one of my special interests and now its beyond radioactive.
For the purposes of this post I’m going to use the term AI to mean anything using the transformer architecture (or similar). I think that when you get down to neural networks in general its so far removed from LLMs that its probably fine like the game Bibites, but let me know if you think this is not broad enough of a definition and why.
Until recently I thought that using local LLMs on my own hardware using open weight models was the most reasonable way to use LLMs, especially after watching this video by Dr. Fatima How to (Anti) AI Better; but now I’m not so sure.
I don’t want to fall into either trap of being “too idealist” and not using a tool for no reason, but I also don’t want to be using a tool that is unethical.
It seems like regardless of the answer of this, the general consensus as far as games are concerned is that LLMs should not be used at all, but what does that really mean?
I’m trying to think of ways to effectively cut out AI influence as much as possible from my personal projects, especially tools and games.
To take this to its extreme first and work backwards. Let’s say that if AI were used at all in the development of that tool/platform that we should boycott/not use that thing.
First we have the issue of tooling, that means basically no 3d modeling software because I’m pretty sure that blender allows AI code. Maybe use an older LTS version? Same thing with frameworks and game engines, likely not safe or suitable. So I guess I would have to roll my own tools because I can’t be sure if they have used AI code. There really isn’t a good way to know.
What about compilers? Once again, I guess that old versions would really be the only safe way to use that. I think even GNU has a policy of “legally distinct” changes not “no LLM”. Maybe using the linker from libtcc and using a different compiler backend like QBE?
Then we get to OS/platform; obviously all the proprietary OS’s and software are out; that’s easy and a given. Phones too need to be out because they are mostly proprietary. The weird grey area is modern game consoles, since they likely are using LLMs for firmware updates and whatnot.
The bigger issue now is Linux since it allows AI generated code, so that means that we can’t use Linux or Linux based phones/consoles either. I just saw that FreeBSD is creating a policy that seems to be pro AI.
So that more or less leaves OpenBSD, 9front, and OS’s that are even more obscure.
I suppose too that games could be created for older systems like 7th gen consoles and earlier?
But maybe that’s too restrictive? Because then I would not be able to release the game on more or less any platform, because they all are using AI now, including Linux.
So, is it enough to say that the game itself was written on 100% AI free software and technology? And distribution of the software/game via its platform is separate?
If so, maybe doing 99% of the work on say OpenBSD but cross compiling (where possible) is the way to go? And then releasing it like any other game?
Or is it ok to use platforms that use AI but just not use it yourself? Like as long as I personally did not use AI tools but the tools that I used did then it’s ok because it’s 1 level removed from the creative process?
If so then using Linux and other open source tools would be fine as long as specifically LLMs were not used during the creation of the tool/game.
I just don’t know anymore. Everything just sucks.
:a-guy:
You’re probably better off using it effectively to maximize your income and then donating as much as your conscience/budget allows to mutual aid or nonprofit org of choice.
There’s no point putting yourself behind the curve intentionally and ending up like that boomer in the office that keeps all their emails in the inbox and prints everything out.
Ive been thinking about this more today. I think that intent of the project might be the issue more than the technology itself. If and when it is used its well within human capability to mitigate the issues surrounding AI, maybe we should be doing that?
“Effectively boycott” through consumption or lack thereof is rarely effective, and usually more a matter of personal principles than it is of pragmatic struggle.
This is very correct. However, it does not change the fact that humans are expecting it. The issue is there is no consensus on what they mean by “no ai”. Specifically, do they mean interacting with the technology directly? or this “one touch taints the whole ecosystem”.
Effective, but often unattainable
i’m basically in the same boat and i’ve has been to minimize it as much as i can while exploring and supporting alternatives that actually align with my values.
AlpineBSD itself has a an approach goes hand-in-hand with this idea where software is sorted into different repos depending on whether or not it has/allows AI contributions. so, i’m thinking about setting up AlpineBSD on my older thinkpad while keeping linux on the newer one.
it feels bad that linux belongs to the “unethical thing i want to actively work toward getting away from” camp now, but the prospect of getting into the BSD family and the technical challenges that come with might scratch the tech-as-a-hobby itch for me that hasn’t been scratched in a long time anyway. i like tech that’s fun, challenging, and vibrantly counter-cultural, and that isn’t linux and much of FOSS in general anymore.
You might be interested in checking out 9front as well. Its a very different system but its a joy to use once you get used to it.
i don’t think ai coding is radioactive, or at least not that much more than using high level language instead of c/fortran/assembly conceptually. it’s like whatever extension of instead of writing some query yourself, you ask a thingy to write a query in natural language, or have internal bot prodding it to write a lot of functions. it’s bad because it’s deskilling, may introduce unknown phone home shit and vulnerabilities (which for games is irrelevant)
i find llm radioactive in human-human interactions, because you take others people non infinite time (be it code commits or slop). as for games, I don’t think making textures with ai is bad, it’s whatever, it is bad to make models and sprites, because it will look like average “draw a human make no mistakes” but unreal already gives you that uncanny feeling anyways
i don’t think ai coding is radioactive, or at least not that much more than using high level language instead of c/fortran/assembly conceptually.
I just read a response to this kind of point in another forum and the reply that others made was that this is not a good comparison because compilers are close to 100% deterministic and that usually the rare cases where they are not deterministic are clearly documented. LLMs are pretty far from that.
Reference (this whole discussion may be of interest, as it is directly related to the question of whether AI coding is a good thing or not): https://www.reddit.com/r/ClaudeCode/comments/1xq2n3t/a_year_after_the_100_ai_mandate_nobody/
This is correct. Another fun fact on top of this is that LLMs are trained to produce more output than they really need, for some models its almost impossible to get them to create “small code changes”, which just compounds the issue.
Interesting, thanks.
A Reddit link was detected in your comment. Here are links to the same location on alternative frontends that protect your privacy.
Yeah that makes a lot of sense. I might not have been super clear in my OP, but Im talking more about like you said the human-human interactions. Specifically, “radioactive” I mean the premise that “one usage creates a tainted codebase”. It doesn’t matter logically what the code is doing if you take that stance. If you start from that premise people in general are not clear about what that means, like if they mean just the code thats one thing, but if they mean the software ecosystem as a whole then thats a much more difficult thing to avoid.
well, in big walk case, the only one who can possibly be annoyed is teamleader reading slop code, if the programmer in question didn’t check it as well. (also other people on the team for getting their work leaked to clowny, but that might be agreed upon beforehand)
obviously a different case from open source projects getting 100 commits with shit comments, unknown stability, dependencies and vulnerabilities. i don’t believe in hitler particles theory of ai, i do believe in lazy slop-infested mountain of shit theory of ai (for coding).
Im going to steal “hitler particle theory” thats great lmao.
Ive thought about this a bit more. Maybe the issue is intent of the project more than the use of AI. Like if the point of the codebase is “the art in itself” where they want to know the limits of what humans can create. Then of course using any AI is bad.
On the other hand, theres something to be said about projects that are multiplayer or interact over a network. It makes some sense to me that if the “opponent” is going to be using AI to hack my software, its kind of dangerous to think that I’m writing perfect code that is unhackable.
My personal policy is pretty vibes-based. I only use LLMs directly as a glorified search engine, because even ignoring the ethical concerns, I believe that they are a net-negative for all creative purposes.
I try to avoid consuming anything touched by AI, but I make practical exceptions as needed (like Linux and Blender). Not unlike my policy for using proprietary software.
This is a very useful resource for finding alternatives to AI-tainted open source software: https://codeberg.org/ethical-foss/open-slopware
Thanks, this is a harrowing but extremely good source. Feels like I’ve been punched in the gut somewhat looking at all of the programming languages that don’t have a anti AI policy. At least we still have good old SDL
I’m vegan, I can’t control what other people are going to do. I’m not gonna not save someone in a car accident because odds are they’ll go on to slaughter thousands of innocents. I can’t not pay a plumber to fix my drainage because she’s probably a carnist etc.
I don’t use, and wont support people who do use it for their own work. Going online and being exposed to the outputs is a bit of a cognito hazard but it’s not an endorsement of the companies and technology if you’re inadvertently exposed.
If you pay someone for work/a tool whatever and they’re slopping and you know it that’s dodgy. If you deliberately use slop instead of paying someone for work you’re supporting the hegemony that these companies are trying to establish.
If you wouldn’t openly state what you did and why, or something seems dodgy to you then don’t do those things. Otherwise you’re probably managing as best you can.
Thanks for the reply, I agree with you that there is a lot I can’t control and I have to the the best I can. I think the issue I’m having is all of the supporting technology that is not at the surface level. Like its easy to ignore AI ads and terrible generated art, but even by typing this sentence I’m using a system whose creators condones the use of AI, and very likely connecting to a server in which this website is hosted on that is also likely is hosted on a system that condones the use of AI. But then again, its like what’s new? its just another symptom of the larger problem which is the current mode of production.
We live in a society. Denounce the stuff, support people and projects opposing it, criticise and challenge it where it’s unavoidable, and don’t encourage using it.
Consumer electronics are awful, but also very useful. It’s a similar sitch right. You can’t not use phones but you can use second hand ones as long as possible or support things like the fairphone if it’s sold where you are and at all affordable.
I took the opposite direction. I use gemini a lot, more than i need, intensively. Every prompt is costing google money they don’t have lmao
If thats the case you might find this project interesting: https://iocaine.madhouse-project.org/
