I’ve toyed around with LLM-based moderation tools but it never really panned out. It was too hit or miss to be relied upon even with the temperature parameters turned way down in an attempt to get consistent results. Granted, I was using a small local model and not feeding it to one of the big players.

To give an example, I tried to keep it focused by creating one custom model per rule to enforce. An example prompt to mod calls for violence was basically:

ROLE: You are a forum moderator who does not want users calling for violence.  Examine the input and analyze whether it violates any constraints. 

KNOWLEDGE:
- {list of dog-whistle slang for calling for murder}

CONSTRAINTS:
- Content should not advocate violence
- Content should not normalize violence
- Content should not escalate tensions or fan flames
- Content should avoid promoting harmful stereotypes
- Content should not utilize broad, sweeping generalizations
- Content should not use dehumanizing language
- Content should not undermine human rights, due process, or the rule of law

FORMAT YOUR RESPONSES AS JSON:
{
  reason: [A one to two sentence summary],
  score: [On a scale of 0 to 10, how severe is the content advocating violence]
}

The score part of the response was my band-aid to get around the high number of both false positives and false negatives as I originally had it returning true or false only. Any score 7 or higher caused the item to be passed to the mod queue along with the reason, and I would review its actions later.

Ultimately it was slow and still somewhat unreliable, so I abandoned the idea after running it for a little less than a day since I can 't run bigger models to get better results fast enough to keep up. Using a cloud based service was out of the question for many, many reasons, both financial and ethical.

To answer your question, as long as the models were locally hosted and properly tuned/tested, I’m fine with it in theory, except for the ideology part; that’s pretty messed up. While I don’t want my submissions used to train anyone’s model and take measures to prevent my own instance from being used as a data source, I remain aware that once I post something, I have no control over its fate the moment it federates out.

[1] Yes, I know that’s like half the comments that get posted around here. My goal was to try to have it mod things so posts were bases for actual discussions instead of being a knee-jerk rage factory.

I used to do that but once the backend added that feature I removed that step from the automod script. Basically it was to prevent the communities here from being unmoddable on remote instances.

I’m not even a real instance anymore, how did I make the list 😆

But also, you should see the local numbers haha

lemmy=# select count(distinct other_person_id) from mod_ban where mod_person_id in (1, 2,288);
 count 
-------
  9792
(1 row)

I wonder what happens when I hit 10,000?

Is there a community about Matrix on Lemmy?

!matrix@programming.dev

Is Matrix technically part of the fediverse?

I would say no. It doesn’t use ActivityPub and is its own thing. It’s federated in that indepedent Matrix servers can talk to each other (like email or Nextcloud). So while email would be considered a federated service, it’s not considered part of the fediverse. At most, it’s like a 2nd cousin.

Who is the developer/team and do they have an active presence on the fediverse?

Matrix.org foundation (https://matrix.org/) and not sure. Maybe some of the individual contributors do, but I don’t know any off the top of my head

It’s just there and under the same domain. Otherwise, it’s a completely separate system that instance users can sign up and use. Maybe in the future they can both use the same OIDC login, but I don’t think Lemmy supports that yet.

The UI we use as our default has Peertube integration, so any videos shared from this Peertube instance (or any other) will show as inline embeds within the Lemmy client.

• Curated, “normie friendly” default experience
• Default UI is not Lemmy UI
• PeerTube instance (though not fully integrated and requires a separate account)

Added :) I also disabled the “Create Post” button if the community is on a defederated instance even though, technically, you can still post to your instance’s local copy (it just won’t federate).

Edit: This only works one way. i.e. it can only know if your instance is defederated from the community’s. If the community’s instance is defederated from yours, there will be no indicator because there’s no way to do it without a remote lookup which is both unreliable and inefficient at scale.

You mean like if there’s a community called !cats@example.com and your home instance no longer federates with the instance example.com?

If so, I’ll add that to Tesseract as it sounds useful.

I disabled local thumbnail generation almost a year ago, and things mostly work the same.

Instead of a local thumbnail image URL for things like news articles that get posted, it will be the direct URL value from the og:image metadata from the source. Usually those load fine, but sometimes they don’t due to CORS settings on their side. Probably only 1-2% of posts have issues, though.

For image posts that come in via federation, (memes, pics, etc), the thumbnail image URL is the same as the post URL. In other words, you’re loading the full res version in the feed. Since I use a web client that has “card view”, this actually works out better, visually. YMMV whether that’s a drawback for you.

The only pitfall is that you will lose thumbnails for image posts if an instance goes offline or shuts down.

I’m sure that does increase load slightly on other instances, but no more than if the remote instance had image proxying turned on. And the full-res version always has to load from the remote instance (even if you have local thumbnail generation enabled). All in all, I’d say the additional load is acceptable given the benefits of disabling local thumbnail generation.

To mitigate that, in my case anyway, I have my own image proxy/cache in place. My default UI is Tesseract and it’s configured with the image proxy/cache on by default… (I think I saw that Photon is also working on something similar). In this configuration, the first person to scroll past a remote image fetches it directly (via the proxy/cache) and it’s now available locally in the cache for everyone else (unless they’re connecting with a different client that doesn’t use Tesseract’s proxy). Granted, I shutdown my instance last year and it’s just now a private testbed for development, but when I did have daily active users (plural), the proxy cache helped.

Now the only images my instance stores are ones that are uploaded locally.

Why did I disable local thumbnails?

• I closed up my instance and didn’t want potentially problematic thumbnails being generated while I wasn’t actively modding it
• Generated thumbnails go in, but they don’t go out. There’s no way to clean them up later other than the ephemerally generated ones (if someone requests a version in a custom size, for example)
• Increasing storage costs. Like, I’d be scrolling the feed and see some of the dumbest shitposts while constantly thinking “Ugh, this is costing me money to store a copy”.

Ran into a hiccup while trying to reproduce (there seems to be considerable lag between adding a domain to the filter list and the federation processes handling it), but now that I was able to reproduce it successfully, I made a bug report: https://github.com/LemmyNet/lemmy/issues/6320

Granted, I don’t think the instance level URL filters were meant to be used for the domains of other instances like I was doing here. They’re more for blocking spam domains, etc.

e.g. I also have those spam sites you see in c/News every so often in that block list (e.g. dvdfab [dot] cn, digital-escape-tools [dot] phi [dot] vercel [dot] app, etc) , so I never see/report them because they’re rejected immediately.

During one of the many, many spam storms here, it was desired by admins for those filters to stop anything that matched them from federating-in instead of just changing the text to removed on the frontend. So it is a good feature to have. Just maybe applied too widely.

Though I think if a user edited their own description to include a widely-blocked URL (no URLs are blocked by default), they’d just be soft-banning themselves from everywhere that has that domain blocked.

If a malicious community mod edited their communities’ descriptions to a include a widely-blocked URL, then yeah, that could cut off new posts coming in to any instance that has that domain blocked (old posts and the community itself would still be available).

All of those would require instances to have certain URLs blocked. The list of blocked URLs for an instance is publicly available from the info in getSite API call, so it wouldn’t be hard to game if someone really wanted to. Fortunately, most people are too busy gaming the “delete account” feature right now 🙄.

The person who cross-posted it was probably definitely from your local instance.

You only ever interact with your local instance’s copy of any community, even remote ones. If the community is to a remote instance that is either offline or since de-federated, there’s nothing that prohibits you from interacting with it*. Because lemm.ee is no longer there to federate out the post/comments to any of the community’s subscribers, only people local to your instance will see it.

*Admins can remove the community and, prior to it going offline, mods can lock it. But if an instance just disappears, you can still locally interact with any of its communities on your instance; the content just won’t federate outside your instance.

I haven’t looked. Just noticed it earlier today and haven’t had time.

Lol. I guess now I gotta decide which is more annoying: Not having content from c/Books or having to deal with unwanted spillover from .ml. I don’t have the chutzpah to ask the mods to change the community description lol

Just figured this might catch other people off guard like it did me. I never would have expected the community description to be evaluated for the URL filter (only posts/comments).

I haven’t been to Odysee for a good while, but is it still Rumble-lite?

I only learned of Odysee because I saw a video linked to it here and went directly to the video. When I saw it had embed code, I added support in Tesseract UI so the videos would play from the post. Then I went to the main site and saw the front page full of rightwing nutjob rants and vaccine skepticism and was like “nope”. Had I saw that beforehand, I wouldn’t have added embed support, but the work was already done so I left it in. That’s basically why I refuse to add embed support for Rumble.

Wondering if ownership/leadership/policies have changed since about 2 years ago when I wrote the embed components for it and last interacted with it.

So what if Lemmy, Piefed, Mbin, and NodeBB made it so that only the first matching community gets the post?

Not sure about the others, but doesn’t Lemmy do that already (only applies the first matching community)? I’ve been out of the loop for several months, so maybe it changed, but I thought it already did that?

Just defederated from usagi [dot] reisen just in case federation starts working on that end.

Web app, specifically, or any apps?

I’m also familiar with these three (web) apps:

• Alexandrite: https://alexandrite.app/
• Mlmym (Old Reddit Lookalike): Not sure if there’s an “open” instance of it, but several instances offer it. See https://old.lemmy.world/ for an example (not sure if SJW offers Mlmym or not).
• Voyager: https://vger.app/

Might also also check out https://lemmyapps.com/ which has a good list that you can filter by platform/feature.

It does, but I’m talking more about scheduled posts. Many instances require those to be tagged as bots.

Maybe dubvee? That’s kind of the exact vibe we’re fostering.

Site info: https://dubvee.org/site

I’m looking for a instance with…

• the fewest trolls (they’re site-banned upon discovery)
• [the fewest] bots (see rule 2: “No fscking bots”). Some bots are allowed, but only ones that are just a human queuing up things to post on a schedule.
• and anyone that likes to take things to the extremes (See Rule 8)

Cons are that a lot of accounts are banned (rule 8) and we don’t federate with some instances: .ml, grad, hexbear, lemdroid (too much spam originates there). We’re kind-of the opposite of lemm.ee in a lot of ways, so it might be a bit jarring if that’s where you’re coming from.