Samuel
WHAT YOU NEED TO KNOW ABOUT THE UPCOMING MODERATION SERVICES FEATURE The Bluesky team is working hard on a new feature called Moderation Services, to help with community moderation. Like how custom feeds are advanced versions of curation lists, mod services are like advanced versions of mod lists.
arrow pointing from "Curation lists" to "Custom feed generators" arrow pointing from "Moderation lists" to "Moderation services" (which is labelled as "soon" 1 39 75
Samuel
Mod services, like custom feeds, are services that you host independently, and then can subscribe to in the app. They basically provide the same sort of options as the current "content filters" in the app settings - except people can make their own ones! They can also recieve reports!
1 2 19
Samuel
In fact, Bluesky's default moderation system will move to this system (although you won't be able to unsubscribe from it, for obvious reasons). But if you wanted to take a much harsher stance on say, hate speech, you could host a mod service and label hateful posts yourself - and get reports too!
1 6 22
Samuel
Paul shared a WIP screenshot of what this is going to look like. From a user point of view, you can find a moderation service, and then hide/warn/show the posts and profiles that it labels.
"Alice" moderation service. there is a Subscribe button, a description, a like button, and a bunch of hide/warn/show settings 1 3 22
Samuel
@samuel.fm
Moderation services apply "labels" to both posts and profiles. Your bluesky client will then read these labels and show/warn/hide these posts/profiles accordingly. Note - your subscriptions are PRIVATE. (although the mod service might know, it's unclear)
2 1 15
Samuel
However, whereas with custom feeds where the team provided an API and an example template and said "go ham", this time they're open sourcing their internal moderation tooling, named "Ozone", which is a prebuild tool to run your own moderation service
1 3 16
Samuel
This means there should be very little technical impediment to running your own moderation service, beyond hosting it (and I'm sure community members will figure that out too)
3 15
Samuel
NOTE - while they are open sourcing Ozone, their internal tool, you don't necessarily have to use it! You can just use the API directly if you wanted to make custom tooling, and apply labels using whatever methodology that you like!
1 1 6
Nafnlaus 🇮🇸🇪🇺🇺🇦
A thought: 1) Person A makes a "No Nazis" mod layer 2) Person B subscribes to the list. 3) B writes a normal post. 4) Nazi replies with Nazi shit to B 5) B doesn't see it. Good! But... 6) Everyone else still sees Nazi shit on Person B's thread. Maybe a whole Nazi subthread. B would be mad!
2 1
Samuel
hmm, good point. although same applies to normally muting someone I suppose?
1
Nafnlaus 🇮🇸🇪🇺🇺🇦
In short, it seems easy to make mod lists act as mutelists, but is it possible to make them act as *blocklists*, to make it so that: If person P subscribes to list L, and a reply to one of P's comments is flagged by list L, then the reply doesn't show up to *anyone* unless it's visited directly?
1 1 3
Samuel
you can block everyone in a modlist, yes. but for moderation services, I suppose you can't?
1 1
Nafnlaus 🇮🇸🇪🇺🇺🇦
Correct, so it doesn't solve the issue. True moderation would normally prevent a Nazi subthread from forming under a post of someone who actively is trying to avoid Nazis. It's good for "I'm tired of reading about X topic", but not "I strongly dislike and want nothing to do with X."
2 1
1 more reply
Nafnlaus 🇮🇸🇪🇺🇺🇦
But my suggestion above - having moderation layers be able to apply to the visibility of immediate replies to your posts - should fix it. Pseudocode like: render(parent) for reply in all_replies: for list in post.moderated_by : if not parent.subscribes_to(list): render_reply(reply)
1 1
Nafnlaus 🇮🇸🇪🇺🇺🇦
This design sounds so much better than moderation lists. Right now people create lists of "users" to mute / block, usually missing a ton of bad-content and wrongly including a ton of unrelated-content. This should beeline in on the bad content, and be good for catching it all. :)
3
Samuel
it's up to the moderation service! like custom feeds, they can do anything they like to decide what labels to apply. some might use humans, others might use machine learning, it's completely down to the moderation service how to implement things
1 6
Samuel
the API is open for developers to make any sort of tooling here. however, they'll be providing their internal solution for us to use if we want to
Samuel@samuel.fm· 16 Jan 2024
However, whereas with custom feeds where the team provided an API and an example template and said "go ham", this time they're open sourcing their internal moderation tooling, named "Ozone", which is a prebuild tool to run your own moderation service
1 5
Samuel
Ozone isn't available yet but I'd imagine it's probably a combo of automated tooling, human intervention, and report processing
1 2
𝚒𝚗𝚝𝚎𝚛𝚗𝚎𝚝 𝚏𝚛𝚘𝚐
Is that what they're calling it? Ozone? If it was something like a GPT implementation I would imagine you could feed it a bunch of training data. Maybe like a database of hate symbols and terminology for example. Sounds like loads of work, tbh. I'd rather them hire human moderators.
1 3
Samuel
Ozone is their internal moderation tool. You don't need to necessarily use it though, you could just use the API to implement some sort of AI-powered one if you wanted to, it's totally up to you
1 3
Martin Laurent
I think both will be possible. (And if I understand it well the moderation service of Bluesky currently works mainly with AI for image and via human intervention for text.)
1 3