Post · bonfire.cafe

@ophiocephalic@kolektiva.social · last week

@cyno
My understanding of the situation is a little different. The dbzer0 crew don't characterize themselves as ancaps, but as real anarchists. They're interested in other anarcho-adjacent tech subjects like file-sharing "piracy". The last time I looked, they were taking the sort of contrarian view of AI that champions local models and open source - a figleaf if you ask me, and one they can't hide behind if they're now embracing shit like ChatGPT.

The Nivenly connection is all about a grant the admin took, to build... an AI-based moderation tool for the threadiverse

@violetmadder

db0

@db0@hachyderm.io · 10 hours ago

@ophiocephalic @cyno @violetmadder The amount of misinfo I see randomly proliferating freely on microblogging is staggering. AnCaps? FFS!

For the record, the instance has never "embraced" stuff like ChatGPT or any corporate GenAI content or clients.

There was no grant from Nivenly. Haidra (a completely different project ecosystem from the lemmy instance) attempted to utilize Nivenly as a fiscal host, until microblogging outrage and rampart disinfo made Nivenly capitulate and throw Haidra under the bus. We've never received anything from Nivenly. They just wasted our time and money before discarding us.

The FOSS anti-CSAM tool we developed **on our own**, from our experience working with the AI Horde, has been shared for free, with all fediverse owners, to help them guard against CSAM attacks. We have not received a single penny for this work other than random donations.

Stop spreading random bullshit. Be better

un nadie sin nombre (@cyno@novo-atlantis.null.media)

impoverished mentally ill migrant worker formerly @mook @possum.city and @big_louse @todon.eu but i can't keep my mouth shut probably really depressing please don't go out of your way to person...

ophiocephalic 🐍

@ophiocephalic@kolektiva.social · 3 hours ago

@db0
I know little about the childish scene drama on the threadiverse, but I can tell you that the response to this on microblog-fedi has nothing to do with "discrediting anarchist instances", not least since I am an anarchist tooting to you from another anarchist instance.

On the contrary, it's *because* I'm an anarchist that I saw a need to take responsibility and denounce the slop-based moderation tactics you're implementing over there in the name of anarchism. As I will repeat from another post in this thread, the only conversation anarchists should be having about "AI" is how to most effectively sabotage and destroy it.

Your logs reference a currently active GPT model. This was dismissed in your lemmy comments as "funny". Why is that funny? It's not. It's not remotely funny that your logs indicate you're sending user post histories to OpenAI. Whether it's true or just "funny", it's utterly negligent moderation practice.

The fig leaf your clique hides behind for justifying your use of "AI" in moderation is that your models are "local" and "open". Open what? Source? Weight? Neither really? Whatever, but let's unpack the "local" excuse.

The notion of "local models" among the soft-AI apologists is that an individual is downloading a model and running it on their own machine at home. You guys calling your serverside "ideology analyzer" local will be cold comfort to the users whose post history you've extracted and fed into it. In fact, from their perspective, there's little difference at all - their accounts were hosted on different instances than yours. Why is it less violating to them if dbzer0 is doing it as opposed to OpenAI? You're becoming the extractive techbros that all of this was supposed to provide an alternative to.

As I noted above, this is copthink. You're doing exactly what cops do - sourcing this exact kind of tech and weaponizing it against "the enemy" on ideological patrol. Anarchism doesn't need its own version of Palantir.

Finally, I'm aware the user account behind all of this is a zionist troll. I know that because I spend 30 seconds examining the user posts, exactly what every fedi mod does when confronted with a report. Why do you even need all of this computationally expensive "ideology analysis" anarcho-solutionism when it's clearly obvious from a quick glance the account is reactionary? Block them from your instance and move on, no "horde" of energy-sucking LLMs required

@cyno @violetmadder

un nadie sin nombre (@cyno@novo-atlantis.null.media)

impoverished mentally ill migrant worker formerly @mook @possum.city and @big_louse @todon.eu but i can't keep my mouth shut probably really depressing please don't go out of your way to person...

db0

@db0@hachyderm.io · 3 hours ago

@ophiocephalic @cyno @violetmadder Here is our actual stance on this matter: https://lemmy.dbzer0.com/post/68757445 and the history of the disinfo behind it.

You keep repeating the disinfo as if it was true. **We do not use LLMs, local or not in moderation.** Never had. Never will. So the rest of the post is yes, discrediting to continue spreading disinfo when we explicitly explain it's. not.true.

What's funny about the inside joke? You have to ask the people who wrote it as I don't find it particularly funny either.

un nadie sin nombre (@cyno@novo-atlantis.null.media)

impoverished mentally ill migrant worker formerly @mook @possum.city and @big_louse @todon.eu but i can't keep my mouth shut probably really depressing please don't go out of your way to person...

ophiocephalic 🐍

@ophiocephalic@kolektiva.social · 3 hours ago

@db0
From your linked post:

> As an experiment, the admin pass the user’s comment history through a locally-run open-weights LLM (Qwen) to see the summarized output. It happens to match their own decision.

In fact, your mod *did* use an LLM for moderation. The user's posts *were* ingested by an LLM. No sympathy for zionist trolls, but your own post confirms this

@cyno @violetmadder

un nadie sin nombre (@cyno@novo-atlantis.null.media)

impoverished mentally ill migrant worker formerly @mook @possum.city and @big_louse @todon.eu but i can't keep my mouth shut probably really depressing please don't go out of your way to person...

~9 more replies

db0

@db0@hachyderm.io · 3 hours ago

@ophiocephalic @cyno @violetmadder No, they did NOT use it for moderation.The moderation decision had already been taken. It was NOT used to confirm that decision either. It was an experiment to see if the LLM summary would be a good way to enhance the modlogs.

un nadie sin nombre (@cyno@novo-atlantis.null.media)

impoverished mentally ill migrant worker formerly @mook @possum.city and @big_louse @todon.eu but i can't keep my mouth shut probably really depressing please don't go out of your way to person...

ophiocephalic 🐍

@ophiocephalic@kolektiva.social · 3 hours ago

@db0
That's hair-splitting. But question for you. Since you're feeding user post histories into it, have you audited Qwen's network activity? And is the server it sits on itself defended from *other* AI scrapers?

@cyno @violetmadder

un nadie sin nombre (@cyno@novo-atlantis.null.media)

impoverished mentally ill migrant worker formerly @mook @possum.city and @big_louse @todon.eu but i can't keep my mouth shut probably really depressing please don't go out of your way to person...

db0

@db0@hachyderm.io · 3 hours ago

@ophiocephalic @cyno @violetmadder I don't think it's splitting hairs at all. It's a completely different purpose. We are NOT feeding user post histories into it. From what I understand, it was one time while testing the summary feature. Qwen is a local LLM model, it has no network activity (it cannot). The whole fediverse is scraped by LLM scrapers constantly and we posted the zionist's receipts publicly anyway. Do we need to publicly audit our whole server setup because one admin used an LLM once? This is really becoming unreasonable.

un nadie sin nombre (@cyno@novo-atlantis.null.media)

impoverished mentally ill migrant worker formerly @mook @possum.city and @big_louse @todon.eu but i can't keep my mouth shut probably really depressing please don't go out of your way to person...

ophiocephalic 🐍

@ophiocephalic@kolektiva.social · 2 hours ago

@db0
I don't find these to be unreasonable questions. In fact, many of these "local" models are quite chatty if you actually take a look at what they're doing. They can and do connect. They're not being built by people who give a shit about anyone's privacy. LLMs, "local" or otherwise, are surveillance technology. They exist entirely as an outgrowth and extension of surveillance capitalism. It's good praxis to interrogate their possible introduction into fedi moderation tooling (spoiler: it shouldn't happen at all).

FWIW, I still think FediSeer was the best approach to horizontal blocklist compilation, by far. Far superior to all the centralized IFTAS-type shit we've endured since then. But the beauty of that approach is that it distributes the agency of **human** judgement in moderation decisions. We don't need solutionism to solve our moderation problems on the fediverse

@cyno @violetmadder

un nadie sin nombre (@cyno@novo-atlantis.null.media)

impoverished mentally ill migrant worker formerly @mook @possum.city and @big_louse @todon.eu but i can't keep my mouth shut probably really depressing please don't go out of your way to person...

db0

@db0@hachyderm.io · 2 hours ago

@ophiocephalic @cyno @violetmadder Depends on how you're using them. If you know what you're doing and download the .safetensor file and use FOSS software to run them, they are literally incapable of connecting to anything. And I really know what I'm talking about here.

I completely agree with you about moderation and solutionism! I am just very frustrated that I repeat again and again that we do not use LLMs for moderation and people just go like "I roll to disbelieve" and there's literally nothing I can say.

un nadie sin nombre (@cyno@novo-atlantis.null.media)

impoverished mentally ill migrant worker formerly @mook @possum.city and @big_louse @todon.eu but i can't keep my mouth shut probably really depressing please don't go out of your way to person...