Salta al contenuto principale


"I don’t know how to run a community forum in this future. I do not have the time or emotional energy to screen out regular attacks by Large Language Models, with the knowledge that making the wrong decision costs a real human being their connection to a niche community."

aphyr.com/posts/389-the-future…

Marcos Dione reshared this.

in reply to École des Bro-Arts

sexual content

Sensitive content

Questa voce è stata modificata (2 mesi fa)
in reply to École des Bro-Arts

Charge $10 membership fee, refundable after 3 months of good standing and positive participation.
in reply to Raphael Lullis

@raphael Yeah, the SA approach. We might end up going that direction, but payment gating introduces other complexities--like direct dependence on payment platforms which are not kind to queer smut.
in reply to École des Bro-Arts

Putting a paywall doesn't have to be the *only* solution, you can couple it with any form of social proof: invite/referral codes, sign-ups conditional to the vouching of an existing member, etc. If you have people at the bars and clubs that are willing to help, they can set up a QR code generator so that only those that really were at a party could join for free, etc...
in reply to Raphael Lullis

Yeah. As I mentioned in the post, that has other drawbacks--those physical spaces have been dying out for the last couple decades, and it isolates people in rural areas, repressive countries, etc. Might be where things head, but...
Questa voce è stata modificata (2 mesi fa)
in reply to École des Bro-Arts

Your final words on the matter are something weighing on my mind. I too have been withdrawing from the public internet and focusing on in-person groups. This segues into friends in marketing and adjacent fields all terrified as conversion and engagement rates are plummeting across most channels all while the cost per impression continues to go up. This feeds back into leveraging LLMs to try to juice the numbers.
in reply to École des Bro-Arts

I have started noticing these accounts too. I am really glad to see you addressed the issue right away.
in reply to École des Bro-Arts

Oh noes! Got me thinking... Dog's recent post about my camera was real and not a secret ad!

But it's so true, in this day and age with LLM BS, how do y'all know I'm a real person trying to make space to chat as my Dog side, and not some new secret ad bot?

In person video screenings for new users? Donations? It's gonna get so tough I agree.

in reply to École des Bro-Arts

would be interesting to run an experiment using an LLM to assess whether some text is from another LLM or from a real human being... I wonder what kind of accuracy we could get from this... I guess we'd need a decently labeled ground truth set to begin with...
in reply to École des Bro-Arts

I don't know about that... there are subtle patterns the AI would be good to spot... I'd say AIs speak a dialect of languages that is "LLMese" in a way...
in reply to :verified_gay: Addickted

@Addickted LLMs by definition minimize their loss against a presumed human corpus--any LLM capable of reliably detecting variance from human text would be capable of rejecting that text in favor of more plausible outputs--defeating itself. This is why LLM detection technologies are notoriously inaccurate.
in reply to École des Bro-Arts

this shit is ruining all my spaces on the Internet. Fuck. I've even seen plenty of what appear to be AI bots on hotcigarmen but I can't confirm
in reply to École des Bro-Arts

Sensitive content

in reply to École des Bro-Arts

thank you for this, and for all the work you and my friends do!
(I finally managed to bypass the UK blockade to your blog using 12ft 🌷).
in reply to École des Bro-Arts

I have had a similar kind of attack recently against @cloudlab , the NSF academic testbed I'm PI on and handle applications for. It was similarly both sophisticated and highly naive. An interesting twist in this case is that they managed to worm their way into getting at least two email addresses at US universities.
in reply to École des Bro-Arts

it's SO fucking hard. This shit is rampant everywhere, and every measure that's even a little effective is guaranteed to leave someone behind.
in reply to wuest

Lotta folks out there insisting "Oh, I would have known that was LLM text", and if it weren't a privacy issue I'd make a quiz where one has to review real applications and determine if they are from an LLM or not. The quiz takes your email address, and for each human post you reject, you get an email, days later, asking what's wrong with the sign up process and if registrations are still open or not. 🫩
Questa voce è stata modificata (2 mesi fa)
Unknown parent

École des Bro-Arts
@cbleslie That's actually the best idea I've heard so far haha. I'm pretty sure that would stop 80% of real humans too though. :-/
in reply to École des Bro-Arts

My day job is trying to insert LLM-generated text _into_ the social surface of our product, and I am constantly left wondering (either aloud or otherwise) whether we've lost track of the product's role in facilitating conversation.
in reply to Owen

@owen yeah I think that would be a signal for me to push back or start planning an exit
@Owen
in reply to École des Bro-Arts

@École des Bro-Arts @C.B.Leslie there are a number of countries where it would be a significant hurdle, because sending a letter is significantly expensive (especially compared to the local wages) or just not a service that is provided.

but it would be *so*cool*

Unknown parent

École des Bro-Arts
@cbleslie @valhalla Yeah we'd almost certainly get LLM-generated images there haha
in reply to École des Bro-Arts

That... is scary. Tools have always existed for the betterment of society, but we always find ways to exploit them. And for what? Absolutely nothing, except misery to the rest of us.
in reply to École des Bro-Arts

This is going to become a dissappointingly frequent occurance isn't it? 😞

Also, sorry that LLMs allow sucky people to crap on good things with minimal effort.

Questa voce è stata modificata (2 mesi fa)
Unknown parent

École des Bro-Arts
@risottobias few of our signups would have this
in reply to École des Bro-Arts

I had an interaction with someone the other day and am still not sure if they are a bot or not. And that is just sad.
in reply to École des Bro-Arts

Thanks so much for writing this!! I was talking about this subject in hashtag MastoAdmin today and your fantastic blog post came up a lot.

I also found the same account on my instance, though it had no bio/pic/name and never posted anything. Thought you might find it interesting to add to your catalog.

in reply to École des Bro-Arts

To follow up on this: you can call Market Research Future at +1 (855) 661-4441, or +44 1720 412 167. I've just checked with them again: the person I reached was unaware of their ongoing spam campaign. You can also email info@marketresearchfuture.com to ask that they stop.

I've also started collecting IP addresses of the bots, which I've added to the post. For other admins, you can look up the IPs of the bots on your servers and reach out to their DNS abuse contact.

Questa voce è stata modificata (2 mesi fa)
in reply to École des Bro-Arts

What we need is anonymous proof of unique human.

It's a hard unsolved problem but of great value.

Easier on an anonymous #AutonomiDweb, but still unsolved. I think there are ways forward there which while not 'proof', could at least make it harder or more costly to operate fake humans at scale.

Questo sito utilizza cookie per riconosce gli utenti loggati e quelli che tornano a visitare. Proseguendo la navigazione su questo sito, accetti l'utilizzo di questi cookie.