I heard about the whole AI moderation thing and I’m trying to move to a new fedispace.
Might wanna consider the admins response first:
- https://lem.lemmy.blahaj.zone/post/42252997/20359964
- https://lem.lemmy.blahaj.zone/post/42212552/20347120
tl;dr: There is no AI moderation, the mod team built a tool to more conveniently get a user history in case a user needed some investigating based off reports they got. They tried a couple times to ingest what the tool returned into a local LLM for a summary but found in overall lacking. In one instance the local LLM’s finding matched what the mods had already found manually and seeing how the summary was concise just posted that.
This is not a problem of the lemmy software, this could as well happened on any fediverse instance. Your posts, your DMs, everything you do on the fediverse is federated and can therefore be seen by the instances it’s federated with.
But why even move, this is the fediverse, make accounts wherever you want and use them at the same time? I use this account for wider posting and have a hexbear as well as lemmy.ml account for local posting (although I sometimes confuse this account with the .ml account).
I can’t open the first link for some reason. But I also found this post to be useful: https://discuss.tchncs.de/post/59687224
Yeah they lost the plot and just deleted the thread…
Here is the comment from db0 originally
Another hit piece from rimu against /0. Must be a day that ends with -y…
EDIT: Here’s rimu admitting that this is all based on a single modlog and him just making assumptions based on that and that alone.
Here’s the series of events as I understand it
- Rimu sees a two-week-old public modlog entry from unruffled banning a zionist as per our instance rules.
- Rimu visits the public evidence linked and discovers that it’s a dump of the comments of said zionist, and it includes an LLM summary of those
- Rimu sees the FAF logo on the pastebin. Rimu assumes that even though it’s clearly not a FAF domain, since it has the FAF logo, it’s an official FAF tool. This is false.
- Rimu sees the LLM summmary and assumes that the banning happened due to the LLM summary. This is false
- Rimu assumes that the tool that gathered these comments is automated. This is false
- Rimu assumes that because it looks like an official tool, and despite the fact that he can only find a single modlog entry, all of the instance admins are regularly using it. This is false.
- Rimu concludes that the instance is using an automated LLM judgement mechanism for instance moderation. This is egregiously false
- Rimu does not reach out to any of our instance admins for clarification, even though we exist in the same instance-cooperation matrix channels. Even though our instance admins explicitly reached out to him earlier this month and asked to contact them for any questions/cooperation in order to avoid further drama and escalations. Note that there’s a reason why journalists ask for comments before publishing information.
- Rimu doesn’t do any other due diligence.
- Rimu rushes to publish with a cross-post to mastodon, because he knows the knee-jerk GenAI-hate there is much stronger than on threadiverse.
- When called on any of the false and misleading statements he’s made by multiple people in the comments, Rimu completely ignores everyone countering him and all arguments, not to mention being so confused he doesn’t even know if he’s talking about mods or admins.
I’ll leave the fact of why Rimu was randomly going through 2-week old modlogs of banned zionists as an interesting open-ended question.
Nothing against Mbin, but how would it help with the AI moderation issue? From what I understand about the AI moderation, it was a group of mods that sent a user’s history into a model for analysis. That will still be possible with Mbin, and anywhere else
I have an Mbin account I don’t really use, because I prefer Piefed features (personal feed similar to multireddits, comments consolidation) compared to Mbin (microblogging support).
It’s really a personal decision.
Back then I registered on KBin instead of Lemmy before I moved here on MBin because the KBin instance had issues. Didn’t care for the UI but the reason why I did that is because of whatever the hell Lemmy devs are heading to
Staff-wise yes but user count-wise no
I’ll give you my number one reason for why I stuck with kbin and mbin since the beginning. They have the best hot sort. It emulates Reddit much better than the other ways of sorting. I have fedia set up with pages, no unlimited doomscrolling, and large thumbnails so I can quickly peruse anime porn.
People don’t talk about this enough, Lemmy’s “Hot” sort is kinda garbage lol. It leans too heavily towards new posts. Same thing with sorting comments by Hot. “Active” is more in-line with what Reddit users expect, or something like Top 6 Hours. I haven’t tested Mbin’s “Hot” much.
I remember a user on Reddit saying that Lemmy is dead because they looked at All+Hot and saw very few comments. On Lemmy right now I go to All+Hot and the front page posts mostly have 0 to 3 comments. All+Active is showing mostly posts with 50 to 200 comments. All+Top 6 Hours mostly has 5 to ~30 comments.
Do you think it would make sense renaming
ActivetoHotand vice versa?I think maybe it would be better to just tweak the Hot algorithm slightly. Maybe either adjust the curve, or use
max(age, 1 hour)in the calculation so that new content is not favored quite as hard, aka new content is scored as if it were an hour old.But you could make a dedicated discussion post asking about this, so more people can see
I asked @[email protected] about this as he is more familiar with the ranking algos:
Maybe not renaming active, but possibly renaming hot… but IMO they’re fine. Its just different from reddit, and more like the hackernews algorithm: join-lemmy.org/docs/contributors/07-ranking-algo.html#hacker-news . It’s more new-focused to prevent the main problem of reddits which is rewarding first posts / comments, regardless of their quality
I also have an mbin account. But l’m enjoying Piefed much more.
I heard about the whole AI moderation thing
Can you fill me in please?
Long Story Short: A bot got created in Lemmy and Piefed to moderate around.
Ahhh no no it’s not a bot - the moderator is using OpenAI to check if a user has said anything that makes them a promoter of a political ideology that they consider to be bannable.
The human mod still presses the ban button themselves, manually.
Didn’t they say it was a local model?
Yeah also the whole “they’re trying to police wrongthink” is funny coming from the dev that implemented this:

Imagine the outcry if someone had a warning label: “🚨 Promotes zionist perspective”see e.g. https://piefed.social/post/2035412apparently it’s not on a user basis but for certain sites
Whoa, I didn’t know it! I hate this.
Try it and see.
It takes all of about a minute to sign up for an instance and it doesn’t cost anything, so there’s no reason not to do it. If you like it, great and if you don’t, you’re not out anything.
For the record I actually prefer KBin/MBin - they just have a better interface than Lemmy. But KBin was a one-man effort and when the guy running it burnt out, it soet of fell by the wayside. MBin is a fork that was meant to be a more open project, but the guy who was mostly responsible for it is a bit of a twat and mostly used it to pimp Monero, so it never quite caught on like it should have.
You might also sign up for an instance or two on piefed. The guy who’s mostly responsible for that is also a bit of a twat, but he’s had some good help and some people pushing back against his tyrannically censorious tendencies, so it’s pretty decent.
The only way you’re ever really going to know what you think about any of them is to try them out for yourself. And there’s literally no reason not to.
You might also sign up for an instance or two on piefed. The guy who’s mostly responsible for that is also a bit of a twat
If you mean Rimu, I haven’t seen any examples of that. Explain?
Piefed by default auto-subscribes accounts to a curated set of communities, auto-hides downvoted content and downvoted posters, attaches icons to accounts to mark those that have beyond a particular threshold of downvotes and entirely blocks links to a curated set of sites because Rimu deemed them to be unacceptable.
The only saving grace is that at this point those (anti-)features are generally toggleable at least at the admin level, but that’s thanks to wjs018.
Piefed by default auto-subscribes accounts to a curated set of communities
When I first signed up, I was given a ‘checkbox system’ to subscribe to /c’s of interest. IIRC I also had the option of signing up for none. That said, if PF auto-subscribes to some core communities (like META, help, Fediverse, etc), then I’m not really seeing a problem?
auto-hides downvoted content and downvoted posters
I mean… I see plenty of downvoted content & posters in my subscribed feed. Maybe that’s not the case with ALL, tho? TBH, that sounds like SOP in terms of social media.
attaches icons to accounts to mark those that have beyond a particular threshold of downvotes
I like it, as that generally helps identify low-effort & toxic posters. I guess I can see the point though that it might accidentally sweep up others, simply because they’re challenging or whatnot. On the whole though, I certainly don’t have a problem with that little bit of extra info, which I’m free to do with as I please.
entirely blocks links to a curated set of sites because Rimu deemed them to be unacceptable.
Yeah. Personally I would have liked to have some access to various NSFW content, as I’m not a child, but then again, PF.S is Rimu’s home instance, and as the dev and head admin, I absolutely do not have a problem with him doing that. I mean, if I really wanted to work around that, there are plenty of other PF instances as a base…
So at the end of the day, you seem to be calling someone a “bit of a twat” based on policy decisions. Someone who’s donated loads of his personal time, effort and brainwaves to a community project for loads of others to enjoy. Can’t say that sounds particularly cool to me, mate.














