It sounds like, perhaps, the moderators only label the content. Then it’s up to your own client (and how you configure it) to filter the content, based on those labels.
If I’ve got that right, then a client could be created that, e.g., displays labels from different moderators rather than filter the content. In fact, I’d guess most clients will have that mode.
That's my understanding, too. And since it's underpinned by ATProto, rather than being coupled with Bluesky, "moderator score" apps could be built that independently track how 'useful' the labels are (and, by extension, the labeling services), subjective to each individual app's preferences. Then users could rely on moderation rankings from their favorite moderation ranking app to determine which moderators to use and when to switch if the quality tanks.
Yes! This is why I’m so bullish on atproto as a platform. This stuff operates at an infrastructure level, meaning you can build other applications to use it too.
I need labels on labels and labels on labellers. I also need labellers for labellers. With that, I can create a network of labellers which can keep each other honest with enough distribution; think DNS root servers but which constantly check if every other root server is still reasonably trustworthy to be authoritative.
Then I need users who (hopefully) vote on/rate/report labels, which is its own problem.
Practically, labels are probabilistic. Different people who are trained on how to label will label most things the same way but will disagree about some things. I know my judgement in the morning might not be the same as the afternoon. If you had a lot of people making judgements you could say that "75% of reviewers think this is a scam".
But "lots of reviewers" could be tough. Look at the "Spider Shield" example: if Spider Shield is going to block 95% of spider images, they're going to have to look at 95% of the content that I see, before I see it. This is a big ask if the people doing the labeling hate spiders! (Someone who values a clean feed might want to have a time-delayed feed)
It seems also that the labels themselves would become a thing for people to argue about, particularly if they get attached at the 50% point of the visibility of a post as opposed the first or last 2%.
Something based on machine learning is a more realistic strategy in 2024. Today anti-spiders could make a pretty good anti-spider model with 5000 or spider images. The tools would look a bit like what Bluesky is offering but instead of attaching public tags to images, you would publish a model. You could use standardized embeddings for images and text and let people publish classical ML models out of a library, I am looking at one of my old recommender models right now, it is 1kb serialized, a better model might be 5kb. Maybe every two years they update the embeddings and you retrain.
If I’ve got that right, then a client could be created that, e.g., displays labels from different moderators rather than filter the content. In fact, I’d guess most clients will have that mode.