I was shocked to see this:
Are we now leaving an AI mute and censor people?! Or is that hopefully a feature that's been overlooked and should be shut down?
5 Likes
The current spam filter on Discourse is LLM-based, and better at catching most spam posts [yet a bit overly eager at times], especially nowadays where the spam can also be LLM-generated.
It only applies to new users[1] and every case of a message being filtered and a user being blocked involves a flag to moderation to review and either reject it, unblocking them[2] or approve the flag[3]. So you’re generally either seeing this on a user with an not yet reviewed automatic spam flag, or with a manually accepted spam flag.[4]
If the messaging is concerning, I can check if it can be customized – “Discourse AI” is probably not a term easily understandable in isolation.
Edit: It appears to be customizable text, but not applying retroactively.
not 100% sure on the exact conditions, but that description I’ve linked reads like trust level 1 or 0, and in particular the first 4 posts ↩︎
which unhides their message, and unsilences the user ↩︎
with additional options, like deleting the post entirely; or in clear cases even delete the whole user account immediately ↩︎
as mentioned in the previous footnote, deleting a spam account is also possible; leaving them “just” silenced instead is useful in less clear cases, where moderation may want to reconsider a decision later, and/or allow the user to still be able to PM mods ↩︎
15 Likes
Interesting... I have been posting on here nonstop since I joined last month O.O I never noticed it before.
Most cases of spam involve newly created accounts who make a new topic thread for their spam post. In which case, if it’s the OP of a topic – the whole thread also getting unlisted automatically.
In this case don’t really have any reason to discoverthat user account anymore to begin with, and thus won’t be clicking on their username either to see that “silenced” status.
2 Likes
Thanks for clarifying, @steffahn. 
I know moderating is hard and unrewarding work (I've been there, long ago), but I just found a little disconcerting that an LLM could be allowed to do those things rather than simply raising a flag. Seems like a "shoot first, ask questions later" strategy—with a madman holding the gun. Thankfully these bullets don't hurt too much. 
2 Likes
I think it's not surprising this forum has a spam filter.
10 Likes
Can't really blame "AI".
"AI" is just a marketing slogan.
It's the humans who write the program who include their biases into the computer program.