Mastodawn

Why reactive moderation isn't going to cut it, aka, "The Sucker-punch Problem".

Imagine you invite your friend—let's call him Mark—to a club with you. It's open-door, which is cool, because you like when a lot of folx show up. Sure, it might get a little rowdy, but they have a bouncer, and you've never seen things getting out of hand.

So, you're busy dancing when a new guy walks in wearing a "I Hate Mark" shirt and promptly sucker-punches Mark. You didn't see it happen, but Mark is upset and tells the bouncer, who kicks the guy out.

A few minutes later, the same guy walks back in and sucker-punches Mark again. Same result. Some people in the club say they'll tell the bouncer if they see him come in again.

Mark wants to leave, but you tell him it's not that bad—after all, you've never been punched, and you didn't see Mark get punched, so maybe he's just being sensitive.

A different guy walks in wearing a "I Plan On Punching Mark" shirt. No one tells the bouncer, because they've never seen *this* guy punch Mark.

He sucker-punches Mark. At this point, Mark is pissed and yelling about being punched.

The club members talk about putting up a "No Punching Mark" sign, but the owner is worried it'll hurt his club's growth.

Another Mark in the club proposes they turn away anyone wearing an anti-Mark shirt or espousing anti-Mark rhetoric at the door, but this gets shot down for the same reason as the sign idea—then someone sucker-punches him.

By the end of the night, your friend Mark is beat to fuck and says he'll never come to this club again. In fact, he's going to tell anyone named Mark to stay clear of this place.

The next time you go to the club, half the folx there are wearing "I Kill Marks" shirts, but there aren't any Marks there, so it doesn't come up.

I've been sucker-punched every day, for the last three days in a row by some of the most vile hate-speech and imagery. The accounts are using open registration servers and signing up with variations on the username "heilhitler1488". I fully expect it'll continue as long as we have open registration servers.

And no, username pattern blocking alone won't fix this, it'll help a little, but mostly it'll just make them wear a different shirt while they sucker-punch us.

#OpenRegistrationHurts

Show thread

Netscape Navigator Nov 6

@alice

So, you’ve actually just supported an argument I made a while back — that the Fediverse needs AI. I first made this point during the major spam attack that hit the Fediverse last year.

Here’s why: if the goal is to grow the Fediverse, registration can’t be invitation-only. Right now, Fediverse software isn’t as user-friendly as mainstream platforms like Reddit, Twitter, or Facebook. Making it even more exclusive would only slow growth further, or even stop it internally.

I say this from experience. I remember how online forums tried to handle spam, bots, and trolls — by making registration harder. Some required manual review or moderator approval. That worked for a while, but it also made those communities less welcoming and more difficult to join. Over time, most forums faded away.

At the same time, keeping registration completely open invites bad actors, trolls, and spam. It’s a tough balance — too open, and the space gets toxic; too closed, and it stops growing. That’s exactly where AI could help: by automatically handling spam, filtering bad behavior, and letting real users in without creating unnecessary barriers.

Would a non-ai solution be feesable? I am personally for human review of all AI decisions as sort of mandatory, especially when discussing moderation activity.

Bot wise at least why can't we integrate Spamhaus and other currently active services?

Show thread

Netscape Navigator

@CWSmith @alice

> Would a non-ai solution be feesable?

I do not personally think so, and I am speaking from experience.

You need instant approval or rejection so as not to hurt the growth and overall flow of your site. Furthermore, it has not been my experience that humans are good at distinguishing between good people and bad-faith actors on first contact — which is exactly what this system would require (deciding based on a random comment made before approval).

I believe that if you’re going to gatekeep who can and cannot join, A.I. tends to do a better job once properly trained.

That's the key there though. Properly trained.

I have seen far too many people get screwed by a system just because it thought they were doing something when they weren't. You see people complain about it from sites like YouTube and Twitch all the time.

It just takes one bad piece of code to unintentionally cause a lot of harm.

Frankly I don't feel that any moderation enforcement actions should be done until an actual act that violates a forums rules occurs. I saw Minority Report.

As it relates to bots and spam, the Spamhaus model has been pretty effective in its own right. Surely at least on that front we could simply retrofit that solution into our use case while maintaining a real human interaction.

You can maybe judge people based on their reputation... but as a man with two of the most common first and last names around I don't want to pay for others mistakes.