this post was submitted on 13 Dec 2024
75 points (97.5% liked)

Fediverse

17848 readers
15 users here now

A community dedicated to fediverse news and discussion.

Fediverse is a portmanteau of "federation" and "universe".

Getting started on Fediverse;

founded 5 years ago
MODERATORS
deadsuperhero@lemmy.ml
wakest@lemmy.ml
75
How can the Fediverse protect against AI slop? (lemm.ee)
submitted 1 week ago by bigfoot@lemm.ee to c/fediverse@lemmy.ml
55 comments fedilink hide all child comments
 

The Fediverse is a great system for preventing bad actors from disrupting "real" human-human conversations, because all of the mods, developers and admins are all working out of a desire to connect people (as opposed to "trust and safety" teams more concerned about user retention).

Right now it seems that the Fediverses main protection is that it just isn't a juicy enough target for wide scale spam and bad faith agenda pushers.

But assuming the Fediverse does grow to a significant scale, what (current or future) mechanisms are/could be in place to fend off a flood of AI slop that is hard to distinguish from human? Even the most committed instance admins can only do so much.

For example, I have a feeling all "good" instances in the near future will eventually have to turn on registration applications and only federate with other instances that do the same. But it's not crazy to imagine that GPT could soon outmaneuver most registration questions which means registrations will only slow the growth of the problem but not manage it long-term.

Any thoughts on this topic?

you are viewing a single comment's thread
view the rest of the comments
[โ€“] bigfoot@lemm.ee 9 points 1 week ago (1 children)

Great response, thank you. My concern is more so focused on future measures; what happens if/when registration applications are answerable by a bot? It's not hard to imagine. What happens when a GPT powered bot leaves totally "normal" unique comments 90% of the time, but occasionally recommends a product or pushes a political agenda?

[โ€“] dessalines@lemmy.ml 10 points 1 week ago

All I can say is that in practice, bots can't answer most simple questions in a believable way, especially questions that require actual personal opinions, or that require any context outside of what they were asked.

The most we've seen is that people created seemingly lemmy-specific signup bots, but they always answer questions in the same transparent way.

The blogspam bots that have gotten through (not for many months now here on lemmy.ml) are all transparent, because they all post links to the same domain. All it takes is one report, and we can remove their entire history.