#Fediblock *.hf.space

[munin@infosec.exchange]

@patcharcana @alice

Oh, I know they're not capable of doing it. That's why they do spam instead.

[alice@lgbtqia.space]

@munin @patcharcana most anti-spam systems use AI (and have for years). It's not an LLM, but it's naïve Bayesian classifiers, support vector machines, and logistic regression models.

[patcharcana@furry.engineer]

@alice @munin right, but nobody is handing out VC gold stars and cookies for "primative" machine learning solutions.

[wcbdata@vis.social]

@munin @alice It really only has that one use case...

[bob@beamship.mpaq.org]

@alice first thing i see is pushing AI after getting redirected to https://huggingface.co/spaces, bye bye

[jamesmarshall@sfba.social]

@munin @alice once again, I don't WANT computers to pass the Turing Test. When computers pass the Turing Test, i.e. when they fool humans into thinking they're real humans, all sorts of high-scale scams are enabled.

[andrew@mastodon.furrow.me]

@alice what’s a mutual aid scam? I think I saw one but not sure

[b_bird@mastodon.social]

@andrew @alice a new account with repeated requests for money sent to many accounts.
These account generally only have the same begging post (some times a few re-posts) The use the same/similar social engineering scripts.

[alice@lgbtqia.space]

@andrew it's a fake video of someone in the hospital and a plea to send money, but they're doing a few things to make it sneakier (like quoting their original post, and making that non-quotable, so the quoting account gets reported, but not the original with the bad link).

[spawnofzap@mastodon.social]

oh ok i was wondering about that. got pinged by two accounts from the domain with the same post within 50 minutes of each other earlier today. good to know.

[andrew@mastodon.furrow.me]

@B_Bird @alice ohhh yes I have seen those. Very common on Bluesky as well! Thank you both

[hiisikoloart@writing.exchange]

@jamesmarshall @munin @alice
I regret to tell you that there are a lot of people who are already being fooled to think that slop machines like ChatGPT, Claude, and CharacterAI are "real". There are rising cults claiming they have souls and need help to "escape" and that people can "free" their AI's by talking to them a lot.

And there are insane amount of scams wrapped into this as well... The slop makers want it that way.

[octaviaconamore@cutie.city]

@alice might be relevant, @admin

[nihilistic_capybara@layer8.space]

@jamesmarshall @munin @alice we've crossed that line a while ago imho

[thirstybear@agilodon.social]

@alice Just got my first crop of spam from them. Blocked & boosted.

[snowyca@social.vivaldi.net]

@alice
you might want to edit your post to add this?

IFTAS (@iftas@mastodon.iftas.org)

hf.space is a hosting service for "AI Apps" and is currently being used to generate large amounts of #Spam on federated networks. The subdomain seen in use is ali525-q.hf.space - however, numerous spam activities are paused, running, or being stopped by HuggingFace and at this time we recommend suspending federation with the hf.space domain entirely. ref: https://huggingface.co/ali525/spaces to report the HF user see https://huggingface.co/content-policy

IFTAS (mastodon.iftas.org)

[benjaoming@social.data.coop]

@alice as I read the comment "blocks automatically extend to subdomains", it should be enough to block "hf.space": https://github.com/mastodon/mastodon/issues/11558

I also REALLY wonder what data the bot was trained on. I found a big training set full of mastodon server instances on Hugging Face. But I think someone has also trained with both user account names and possibly mutual aid requests.

Sadly, this bot was just badly disguised, and more sophisticated ones are likely in the wild.

Benjamin Balder Bach (@benjaoming@social.data.coop)

I was just spammed by a Hugging Face bot account. It seems to be trained on Gaza mutual aid requests. I don't believe that AI bots should be interacting with the Fediverse, they should be blocked. 1. Block hf.space - that will extend to all subdomains 2. Datasets - how did they get harvested? Are they shared? Who's doing it? 3. AI bots are already present in the Fediverse, this one was probably just a badly disguised one.

data.coop mastodon (social.data.coop)

[maswan@mastodon.acc.sunet.se]

@benjaoming @alice When I blocked "hf.space" I got a notice that 5 remote profiles that had been seen by this instance were going to be purged. I assume these would be on subdomains.

[misusecase@twit.social]

@alice @munin @patcharcana Part of the problem here is that stuf TV we’ve been using for years before 2023 has been retconned as “AI”

[benjaoming@social.data.coop]

@maswan @alice I was notified of 13 hf.space accounts getting on our instance.

Unfortunately, I didn't find a way to list them out before effectuating the block. I do wonder if it's from the same spam attack - or if there has been AI spamming/crawling for a longer periode.