Discussion
I should make a list of FOSS apps, projects, and services that have Fediverse accounts and don't code with AI, then strike them from the list with a big ELIMINATED notice if they do start drinking the slop kool-aid.
EDIT: Made one https://fedidevs.com/s/MTA2OA/
Honorary mentions
https://infosec.exchange/@catsalad/117341552145650963
@catsalad just a list of slop code in general, hand written or otherwise, would be nice. OpenSSL wasnβt AI coded and that didnβt turn out so well, for example.
@catsalad Great list?
Maybe to add to the list: @littlefedi_official
@scott @littlefedi_official Bookmarked for when the source code is published :3
@catsalad I would differentiate between using AI vs using what these losers call βAgentic AIβ which is nothing more than neglecting your own agency.
I use multiline autocomplete, a chat tab as a search on steroids and targeted medium-sized function/block generation. All powered by local models.
And I donβt think Iβm using βAIβ in a way that wasnβt possible 10 years ago already. And this didnβt hurt software development nor the planet nor my cognitive skills.
A doubt: Iβm developing a Grammarly alternative that uses rules + local LLMs. But with the aforementioned workflow, do I qualify for the list?
@iscarlosmolero
You're skipping the legal and environmental aspects of using LLMs.
@catsalad
@iscarlosmolero yeh, sorry even local, open weights etc are a problem.
Where do they come from, does knowing you use #localAI help normalise it, encourage other uses (with harmful effects).
I won't argue because even if you are the one person who's #LLM use genuinely isn't doing direct harm, do you want to risk secondary harm of helping tech fascism and the inevitable genocide that will flow from it? Have you read how it is inherently fascist tech?
Food for you to chew on.
@happyborg @0x0 @catsalad I donβt think youβre understanding me.
I trained my own 1-3b models to do autocomplete and some for web browsing and retrieval.
In any case, Iβm open to discuss the possible dangers even of this approach.
@iscarlosmolero I don't have time to get into the detail but I'm not sure it matters...
And maybe, as I said, you are unique IDK, but even saying "I use local LLM models (I trained)" can normalise and encourage other -harmful- LLM use because few will understand any difference with what you do.
@happyborg @0x0 @catsalad I got you perfectly, the LLM acronym and the normalization of its use is encouraging people to use them irresponsibly.
But this only happens when the reader is tech-illiterate (by no means as an insult). Amongst engineers, the difference between these approaches should be more than clear.
AFAIK, 10 years ago we were using so-called AI already on our IDEs. So called because itβs just predictive text. As for the chat tab I use, for me is like having a convenient search engine.
But again, I totally get you and I understand the concern.
> But this only happens when the reader is tech-illiterate (by no means as an insult). Amongst engineers, the difference between these approaches should be more than clear.
That's a conjecture not fact. If it were true I do not think many engineers would be using this stuff in any of its forms.
Anyway, I've already said I do not wish to debate this. By all means do you, but don't seek my approval.
Suggestion for the list: @OBSProject
(They have an anti-AI coding policy, see https://mastodon.gamedev.place/@duke_of_germany/117316316603616893 )
@catsalad well, I looked into Forgejo as a potential base for my project, but there are over 2k open issues in their repo, and some of them are over two years old. Honestly, the only way for them to keep up would be employing a clanker.
My project would require new features in it, but there's no chance they would be implemented within Forgejo, and myself I'm bad at golang.
I'll probably implement it from scratch, using Claude as a coder, as I don't have a budget to hire a full time developer (it's not even a commercial project).
@catsalad good list, thanks!
@catsalad Honest question. Is there a way a company could redeem itself, and would there be a process for granting them a place in the list? Not that I am at all aware of or nominating any candidates, just wondering if it had been considered in addition to requirements for removal.
@roko_basilisk An AI code contribution ban would redeem an account, and mere lack of AI code will get them on the list.
@catsalad Just found this: Fork of KPXC without LLM commits: https://codeberg.org/ChiPass/ChiPass
Also, this is the opposite of what you asked for, but many entries have alternatives.
@mttaggart Yes, and I really appreciate all the work @whitequark put into this!
There is no dedicated ChiPass fediverse account, but it would gladly be on my list if it was.
@catsalad @whitequark Ah geez, I didn't read carefully enough, sorry!
@catsalad @Kien They just rebranded, here's the new account: https://social.treehouse.systems/@nura/117343503396319233
It doesn't have any βeliminatedβ notices, butβ¦ https://codeberg.org/brib/slopfree-software-index
@argv_minus_one @catsalad no AI/LLM slop in #GIMP code.
@GIMP @argv_minus_one And we APPRECIATE it so much!! :3
The first item does have a section on retractions, by the way. codeberg.org/brib/slopfree-sofβ¦
@freeplay@wetdry.world @catsalad@infosec.exchange
No, our point isn't to shame projects that use AI, and rather to highlight those which don't.
See https://starlightnet.work/posts/the-no-ai-list.html which goes over why I started the list and what the approach/philosophy behind it is
@freeplay@wetdry.world @catsalad@infosec.exchange
Historically, the projects that got removed from the list fell into one of two categories
other cases are rare, moreso exceptions than the norm
@alexia
Does https://projects.torsion.org/borgmatic-collective/borgmatic/src/branch/main/AGENTS.md warrant an update of their entry?
@freeplay @catsalad
@alexia
On the other hand, suggested addition to the list: https://mastodon.online/@readeck
@freeplay @catsalad
@viq@hackerspace.pl @freeplay@wetdry.world @catsalad@infosec.exchange
RE: borgmatic
don't think so! this still clearly says Please do not use AI agents to modify this codebase.
Weirdly enough it then...still lists out interesting parts of the codebase?? but I think I can write that off as an oddity more than anything
RE: readeck
sure why not π
if ya can show me where their policy is ofc
Hi, I saw my project was on the list. I just want to clarify that novelWriter does not accept generated content, for much the same reasons as others (licensing) and also because it's usually crap code and we don't want it. But our policy isn't strictly No-AI in that we don't say you cannot use AI assistants in your code editor. On narrow scope edits these tools mostly just replicate/mirror existing code (hard to check), and they also have non-coding use cases.
Install bonfire.cafe
Get the full app experience
@catsalad wow @elementary has such a beautiful banner !
I'll consider switching to it next time I break my Debian