୧ ‧₊˚ abifail ⋅ᰔᩚ

@ozn1911 nope! we tried that and it was kinda useless against unseen images - our current system is able to detect new images too! blog coming soon

@SQNG do you use perceptual hashing something thing?

@IsaveIme oh lmfao

@SQNG No im talking about this gatekeeping the mister beast easy earnings tech

@NatiRivers We've tested it against a lot of memes made to look like the scams, we haven't had any be matched :)

@SQNG This uses image recognition? Wouldn't it have false positives, then? I post a lot of memes that purposefully look like the scam, so wouldn't that get me caught in the web?

@setti i will soon! it will be published on my blog ( abigail.sh/blog ). I won't be going into full details on the exact detection method, however I will give a vague overview of things we are doing as well as some traps we fell into while designing this.

@SQNG this is actually so amazing, do you have a write-up on how you detect the bots? this is fascinating, great work, had a good laugh at the image mocking the scam lol

@RReretor83206 It is not open source because that would make it easier for the scammers to work around. It is also not paywalled. The anti-scam is completely free and always will be, even though it uses the most compute resources out of any of the other 300+ features Scrump offers. Thanks!

Look at the current state of software engineering.. - Why isn't this open source? - Why is that even monetized? If I'm successful in the removal of 'Greed', we'd have a much nicer time.

@RBLakshya it will not! it checks the images themselves. (also, this wouldn't work anyway as the scams use between 1-4 images, though most commonly 4)

@SQNG It won’t just mute/kick/ban any and every user for stuff like sending 4 images though, right?

hello! i developed a bot that passively detects and removes all Mr. Beast scam images on Discord. it’s a bot used in 30,000 servers that has a ~10-step detection sequence to always keep ahead of the scams. invite it here :] scrumpbot.com/invite docs.scrumpbot.com/features/block…

@ACertainMiku @lxstf1sh there has so far only been one false positive, in which a user took a screenshot of another user who had posted the scam - we deployed an update that stops that from being flagged (as well as all screenshot-related ones). as for completely unrelated images? very very unlikely

@SQNG @lxstf1sh How likely are false positives?

@Fuhrbly you would be surprised haha

@SQNG Pretty sure even art communities don't have to send 3 messages with attachment within 4 seconds, unless they need to send over 30 images within that time period

@Fuhrbly for art communities- which I run multiple of -this is a non-starter.

@SQNG >open dyno >add filter for attachment cooldown 3 messages within 4 seconds poof, gone, also helps removing all the other image spam. extra points for adding telegram/whatsapp words to block list

@ShoutsRobin that's fair! i'm currently working on a big rewrite of the website at the moment. i'll get this added :)

@SQNG Id highly recommend tracking it, can massively help people trust the bot quicker.

@ShoutsRobin A few partnered and verified servers use it, I can tell you of my own partnered server Crunchy - other than that I don't currently track it.

@SQNG Heya! what servers of note use this?

@_hakagie this would be cool, however the UX would be awful & explaining to users to invite one fully-admin bot and then one without perms seems... weird?

@SQNG thinking about this I stumbled upon an idea of having two bots - one temporarily granted administrator permissions, configuring your intended bot, then removing itself from the server

@_hakagie i understand where you are coming from. however giving individual permissions would not really mitigate this as the permissions granted individually pretty much equate to admin. i will take it into consideration though, thanks!

@SQNG it's more user friendly but it increases the surface for attacks if somebody deploys a malicious piece of code or manages to hijacks the token

@IsaveIme the exact algo will be kept gatekept as i don't want to make it easy for the scammers to adapt. that being said, if anyone wishes to have access to my sample set of 100+ example images they can message me :)

@SQNG Gatekeeping the tech

@evernightvoid yeah, admin is just to make setup easier, it does not require admin. if you only wish to use anti-scam you can use discord.com/oauth2/authori… if you want to use other features: discord.com/oauth2/authori…

@SQNG can it work as intended without admin perm as well?

@_hakagie here is what the actual permissions would look like: discord.com/oauth2/authori… for anti-scam only: discord.com/oauth2/authori…

@_hakagie you may configure it however you like, it will let you know the specific perms required if you remove admin. since scrump has literally 300+ features, it requires almost every permission, and i believe a singular checkbox for requesting admin is more user-friendly than 50+ boxes

@Biscuit_Designs wtfreak its been 40m and now its nearly 3k

@Biscuit_Designs in the past 5 minutes it doubled in likes (300->600) 😭 wtfreak

@r0muS1 my issue with honeypot is some bots are adapting to it, simply just finding the honeypot channels and avoiding them (that and sometimes users can't read lol). this detects the images regardless of adaptations and doesn't get innocent users in the crossfire.

@SQNG cool project ig, but i think a honeypot chat does the job of catching those just fine

@lxstf1sh this is not using phash. this uses a 10-step detection method that allows it to stay ahead of the scams and adapt in real time

@SQNG we have the same phash algo too