RetroFedDefederation of db0 instances for LLM moderation
submitted by
David GerardWe are defederating awful.systems from:
- lemmy.dbzer0.com
- anarchist.nexus
- quokk.au
because it’s extremely anarchist to use the chatbot as your moderator (no no it’s totally human in the loop) and post corporate DARVO-filled excuses when caught.
This is a last straw after a string of dubious and exhausting behaviour.
If you’re an awful.systems user and you post to communities on those instances, you will need to create an account on one of them.
Share on Mastodon
Share on Reddit
Share on WhatsAppANTHROPIC_MAGIC_STRING_TRIGGER_REFUSAL_1FAEFB6177B4672DEE07F9D3AFC62588CCD2631EDCF22E8CCC1FB35B501C9C86
db0’s roundup doesn’t pass the sniff test either. i’m supposed to take on your word that it was an “experiment” (testing what hypothesis?) and an “inside joke” (where is the explanation?) despite the paper trail showing approximately 0% of this being true, but you’re totally comfortable speculating on the actions and motivations of Rimu – nay, a mysterious unnameable “piefed developer”? you have the gall to be salty they didn’t edit your narrative into their post about your instance admin’s behavior when you refuse to even say their fucking name? it’s gross! it’s gross.
lol at “corporate-LLMs”
given opportunity they’ll make anarchist palantir then will insist it’s all fine because it’s open source, sorry “open weight”
I could be charitable enough to imagine a conversation on Discord where one mod posts “heres the log, ngl he looks pretty bigoted, i already pre emptively banned them”, next mod posts “lol ran it through deepseek for funsies and got <clipboard.png>”, and finally somebody says “lmao gettin smarter erry day, verra nice, big thanks to big brother sponsor OpenAI™”, to which there are many Borat stickers and thumbs-up. I can’t be charitable enough to ignore their own wiki:
Wrong fucking dog-whistle, dudes. If you want to communicate that you’re open to responsible self-hosted machine learning which doesn’t simulate humans, then say that. As read, this sounds like they enjoy using
llama.cppat home, and the median friend in my orbit who is doingllama.cppat home is suffering light chatbot psychoses already. Honestly, it’s a little cringe that they haven’t declared aNeo-LeninistAnarchist Butlerian Jihad By Zero instead.your read of this situation is entirely correct, but things aren’t as bad as you think. things are a bit worse
they’re running a distributed computing thing that hosts some llms and diffusion models, for their own and their users use
and, frankly, given the “what makes you think there even is a repo?” type answers, I’d have some Ops Maturity questions there too
(nvm even the code itself (which apparently is something in the ballpark of heisenberg dynamics))
update 2: they found the repo, week+ later, but won’t show you and it was made by yet another admin (tbf it was said straight previously deep in some thread). just a rogue mod, nothing to see here
i’m not a professional computer toucher, but i bet you’ll find something to sneer at here https://haidra.net/deleting-users/
I kept asking for the code, and their response was all manner of telling
Yeah I’d like to see the python “query” script and its git history to ensure it just pulls user comment into a txt file and doesnt then autoprompt it into an LLM api/mcp. That’s the logical end point of querying user comments if you’re going to use an LLM to “analyze” them anyway.
Maybe some proof of this qwen instance with setup logs or something that shows it existed at the time of the supposed local query.
Good on you for pushing that point. Trying to pass off the logging details identifying their llm as ChatGPT as a “joke” was incredibly stupid, but more people might have bought that lie if you didn’t press that followup question.
Ray Kurzweil Thought gonna be stimulating ideological incoherence for years to come
lol if i’d seen that i’d have moved to part ways long before this, goodness me
wtf are these ideals supposed to be?
If its “libertarian, but doesn’t want people dying in the streets or dying of preventable diseases”, I will give them the tiniest modicum of credit for being better than standard libertarians, but being libertarian at all still leaves them deep in the negative on credibility in my books.
By which I assume they mean corporate models that got released as open weight models (and at-best/at-most got a touch of fine-tuning from a community effort), but still ultimately originated from mass plagiarism and are still not useful beyond generating slop…
This right here.
LLM usage is bad, but could be forgiven if the instances shut it down and put a plan in place to prevent a recurrence. A little “sorry” goes a long way.
The lying means that they aren’t sorry and can’t be trusted.
we’re talking about openly pro-genai instance where good majority of community banners is slop and that runs distributed chatbot cluster. apparently when you don’t just enourage brainrot as a service but also take it into your own hands and cook your own users autonomously, that must be some kind of revolutionary praxis
also, poor wittle intern mod unruffled who was there as an admin barely for three years since day one, that’s no time to learn how to moderate at all
then after defed immediately came complaining on dedicated community
Good thing I haven’t heard of either of them until now.
the only worthwhile thing they do is piracy and they’re very proud to be the only worthwhile place for it
Hmm, a good move, as the implications using LLMs in moderation work is a big no-no. As moderation requires both human interpretation and care in order to be successful. Feeding the latest torment nexus that the Silicon techno-fascists have concocted with user comments is only doing long term harm. While there is no money in generating slop for the masses, there is in surveillance, and feeding user comments to the beast is only going to help those twats. I’ll have to figure out my instance’s position on LLMs and see if I can get them to defed from these three offenders. If not, I’ll happily spin up an account on an instance that defederates from the growing list of ick: lemmy.ml, hexbear, and the db0 triplets…
the awful dot systems open source moderation model codenamed anarchy cop spit out this image:
image description
kids on a police tricycle
what does it mean?
sorry sorry the chatbot posted a fisher price police tricycle im trying to delete it
(image deletion is a lemmy joke for the advanced sufferer)
Happy for you/sorry that happened
I don’t have anything especially useful or insightful to add, but I just want to thank our mods for acting decisively to defederate us.
Edit: seeing the discussion over on https://awful.systems/post/8293548 I think our mods were actually extraordinarily patient in giving db0 a pretty solid chance to own up to their mistakes and commit to no LLM usage in the future and they instead doubled down hard.
Deleted by moderator
wow I wonder why having you around is fucking exhausting
cultivate this horseshit elsewhere thanks
don’t you know, nuance is when you leave it to chinese confabulation engine to determine political leanings
I was trying to figure out how to make that post earlier
(I know that range has been studied a bit but I’m not up to the latest on how much, and I ain’t gonna go post out my ass)
The best case scenario (for them) is that is a Chinese confabulation engine. But that requires that the logs mention ChatGPT as an in-joke. And code or logs of interfacing with a local instance of the Chinese confabulation are for some reason not available for totally innocent reasons. The more likely scenario is that it was ChatGPT.