I have been on reddit for just about 12 years now. Something I’ve noticed over time is just how hateful the place has become. A complete outrage machine. Every single sub became filled with it. I’ve filtered so many subreddits over the last few years, it’s insane. I don’t know enough about this place to be sure, but I do hope it doesn’t become the same type of echo chamber of anger.
I’m telling myself it was terf and troll sock puppet accounts.
I’m very keenly waiting for captcha to be fixed, I hope most instances decide to make you fill out a captcha for EVERY post. That and paid instances will make a huge difference compared to Reddit. It will become way harder to spam and astroturf discussions, but I’m not sure how to handle legitimate bots.
AI can solve captcha easily, captcha only causes inconvenience for authentic users, it does nothing to prevent bots
That’s provably false across many social media sites. No anti-bot solution will ever be perfect and it will always be a cat and mouse game. Captchas have a measurable effect on limiting registrations and comments from bots.
We don’t say “deadbolt locks only cause inconvenience for homeowners, they do nothing to stop burglars breaking a window”. We defense in depth. We use the deadbolt as one part of the security/defense plan.
Captcha is one part of the many actions and systems that would make up effective protections.
Yeah I know we have an automated service at my work that automatically solves the captcha off some government site and then scrapes some data off of it every day (it’s public data). The sucess rate is near 100% I believe.
Magic happens when you only require captchas that a language model told you were inciting, hateful, or plain troll feeding. It even makes sense to make part of the score thread-global, as in “Someone already made a Hitler comparison, better throttle this thing”. The worse the score, the more often claim that the user failed to solve the captcha.
Accusations of censorship will fall flat because you don’t prevent anyone from posting, troll feeders won’t bother posting because they don’t care enough to bother, trolls get bored, trolls leave.
I’ve heard of PageRank based solutions, what sort of ai models should we be looking at?
As simple as SpamAssassin, that is a naive Bayes classifier, to running a full-fledged LLM which is very likely complete overkill. It really doesn’t need to be particularly sophisticated as being inaccurate isn’t really a problem, the whole scheme relies on the statistical impact it has on the whole forum, not the impact it has on a single post.
PageRank really only applies to analysing links.