alt Hacker NewsA peek into Reddit's anti-spam internals
Comments
Reddit must have some mechanism specifically for non-spamming bots that isn’t covered in this article. I wonder how it works. I imagine the mechanisms are more complex and opaque than anti-spam (with various levels being exposed to the hierarchies of Reddit and government backdoors). These days, I’ve noticed an almost forcing-function that operates to put the minimum spin needed on posts and comments to turn signal to noise. It seems smart enough to not only generate noisy comments but create comments to amplify existing organic noisy comments. I’m sure these systems are decentralized, emergent, and split across numerous nation-states and actors. I’m also fairly certain what we have now is a tenuous balance that has emerged from all these actors and Reddit policing actions as well.
I imagine Reddit has a high-level of insight into this and a certain level of permissibility it grants, both to inflate user counts and to steer public discourse and insight into less productive mean (or productive to certain interest groups at the expense of the people). I think is also an effect that Reddit has become more global and consensus of the USA people is very antagonistic to the consensus of the people of the world so that doesn’t help (+ access to LLMs to make English writing no longer a barrier to entry).
Neat rabbit hole. Reminds me of having to deal with email spam - it was a similar deal with rule-based filters, ML scores, domain bans,IP filtering, browse fingerprinting etc and mishmash of ever evolving scripts surviving org and personnel changes. Glad i dont deal with it anymore as the frontier seems to be 2 fronts now with human and agentic spam.
Based on the current status of my shadowbanned account (I suspect a competitor in our space retaliating), it looks like `banall` only flags posts from the last 6 years.
Of course, nobody can view my profile anymore anyway (I'm waiting on appeal), but on my account, only posts from the last 6 years have the "Sorry this post was removed by reddit filters" message.
Can't you just append ".json" to the end of any Reddit link and read all sorts of these fields?
Damn, maybe I can finally find out why my 10+ year account was globally (and retroactively) shadowbanned, even though the appeal was allegedly granted.
In the past, those post removals didn't even exist in the moderation log, so perhaps a reason could give me a clue... On the other hand, I'm taking a kind of emotional damage just remembering.
[flagged]
My friend got shadowbanned for posting a youtube link, part of a interview with Sascha Riley (the one where the explains the thing with the tent peg):
I swear I read this article 2 or 3 days ago and the comments on that post were also same as this post. Am I missing something here?
I cannot emphasize strongly enough just how deeply pervasive the spam is at Reddit. I'm a mod at the ecommerce subreddit, and I've only caught some of the AI-powered marketing operations because in one particular campaign that was making fictional claims about things I had direct knowledge about. Once I looked into the post history, and started to untangle the web of accounts that formed a self-supporting community of posters and commenters, just subtle enough to get genuine engagement, but specific enough to make the kind of posts that the LLMs will siphon up and regurgitate.
It's not just shady little operations. I'm speaking specifically about the SCAYLE ecommerce platform, in my example. They've got Zalando money to play with, and as a German platform that's trying to break into the North American market, it appears they've made a bet on indirectly spamming the LLMs with fictional tales of commerce replatforming horror stories. At first, they're some of the more interesting topics in a sea of really useless posts, with contributions from people who seem to have some real experience with enterprise ecommerce. I was a little suspicious, but these interaction campaigns were spread out enough that I didn't put the pieces together for months. Of course, to go back on what I said at the top of the paragraph, maybe SCAYLE is shady, and I'm giving them too much credit.
The good news is, some of the AI powered tools that mods have access to are getting better at surfacing suspicious patterns of behavior. However, I still find I have to manually address these campaigns.
In the cat-and-mouse game with these marketing jerks, I'm always reluctant to surface what's working and what isn't. This is an interesting post, but it's going to make things worse. Ah well.