+1

What's a CP spam?

why can't we just have an AI automod

cp is short for child pornography.

Agreed.

Agreed,Get rid of the CP.

Cp spam be gone.

You don't even need ans AI.

It's always the same couple of pictures. Hash the pictures and whenever someone tries to post a picture with a similar hash autoban him. Then whenever ans mod deletes ans post for ans cp add that picture to your hashlist.

 

Since the poster is just ans bot it will most likely not adapt its strategy to use different pictures.

 

sadly that is not how hashes work

It's about equality retardo. You just prevent the bot to post the same picture against.

The pics do be changing each round of spam thoughever

Igor don't pretend these incidents aren't the sole reason you come here

and then the bot gets modified so it changes a random pixel every time it posts

that's why I said >>300220 we need exactly ans AI

 

I've been using that bing dall-e portal thingy & it's kinda funny

it detects naughty prompts and doesn't let you generate that

but it also blocks images which it thinks are naughty from getting across to you

and what is funny, apparently dall-e has consumed quite a lot of porn

so when you ask for foids, chances are much higher than zero, that it will generate her naked

for example, if I specify her clothes, but forget to mention she's wearing pants, I will occasionally get only one pic through in which she's clearly not wearing pants but in a way that doesn't expose anything (the other three obviously being in more revealing poses that can't get through the filter)

and if I generate beasts rather than humans, well... you can get quite a lot through without triggering it, as long as it doesn't look too human

 

that's something like what every board would need

and we have the technology now

so what are we waiting for

guise ur overthinking why not just introduce few captchas per ip before posting

Yeah, something needs to be done about that cp

Meh

I doubt that you can get anything useful for offline use and it is a really bad idea to send cp images to some online service to find out if it is cp.

 

I doubt the owner of the bot monitors this board and will react to it. Either the bot already changes the images for every post a little and my method will not work at all or it will protect us reasonably well for very little effort.

you said "similar hash", as if similar pictures evaluated to similar hashes

 

but yeah let's proceed to just insult me lmao

 

 

what we need is clearly an AI classifier, we don't even need to analyze the pictures, training a network to detect the cp posts text based would be enough

 

 

actually i've already been collecting pedo post texts in a discord channel, eventually i will post a javascript function to detect them but i don't have enough pedo post texts yet because usually when i see them i don't even think twice and just delete lul

I can see why you misunderstood my post but I'm not sure if it was actually incorrect since similar also be used as a synonym for equal. What grinds my gear the most about your post was the arrogance to just say that something is wrong while being too lazy to write why it is wrong.

 

this is probably a way better idea though. Why not ban all posts that contains a url to some url-shortener and maybe captcha all posts that contain a url.

cp posters always want to lure users to one of these urls. When the bot tries to hide the url in the image you can try to find urls in the image using ocr. No fancy AIs needed.

learning a binary classifier isn't really fancy and you could keep posting whatever is not a pedopost

removing a true negative post with the reason "pedo" could be automatically added to the training data to readjust the weights

 

it all really depends on the results though, if you get too many false positives, simply banning short urls might just be the way to go

Okay. Here's all the wisdom you need:

1. Captcha won't stop it, Endchan has captcha, other IBs have it, still cp spam appear. Duckduckgo/searx/etc. 'OCR' It is possible implement captcha that OCR fails, but this is just an arms race. Endchan got flooded in 2022 autumn for like three months (not by cp spam), where the spammer in the end used a human-powered captcha solver service (i think it was 2captcha), so in the end spammers can just employ real human beans to solve that shit.

2. Hashbanning images is useless. They just use other pic next time. If you consistently hashban images, then you teach the spammer to pre-edit thousands of images so every time there is a new one available. (The aforementioned spammer did just that).

3. You could implement perceptual hashbans, we had that on 9chan. It works great. Except it does result in false positives and will catch images (and with that innocent users) it shouldn't have. So then you have try to fuck with unbanning people and images, or fine-tune your system. And then the aforementioned spammer used tens of thousands of pre-generated images that probably would gone through the perceptual hash (unique filenames, unique filesizes, qr code on the image and whatnot - they were all the same image beyond btw) anyway.

So in the end the only thing one can really do is to catch the spam in time. This is the least bother and time consuming approach.

This: >>300220 could solve the problem. Except in the end there will be some AI spammer tool to counter that. It's an arms race.

Notice to janny: when noncery appears on /int/ he made the same thread on /de/ and /b/ as well

don't leave your wörk 1/3 done

M8, you have open source stuff which is pretty competitive.

e.g. Mistral is fully open.

3 chan got flooded in the late summer of 2022 and has since been taken down because of cp spam. And probably other reasons. 3chan.co

Mistral and llama.cpp

this website runs on cloudflare, shouldn't its illegal content filter catch the spam? if it's turned on then maybe you should consider contacting them for help