William Pietri on Nostr: David Gerard just pointed to some #bluesky code showing that they use a proprietary ...
David Gerard just pointed to some #bluesky code showing that they use a proprietary vendor, Hive, for ML of content analysis to deal with porn, spam, hate, and the like: https://circumstances.run/@davidgerard/110303972798576262
I tried their demo out with some basics and quickly got false positives and false negatives. At first glance, I expect they are not suitable for an adversarial situation like a popular social network. Which is true of all the APIs like this I've looked at.
And then there's the cost. Want to use Hive for your own Bluesky instance? You'll have to talk to a salesperson, meaning that it's expensive. Just at a guess, paying for AI moderation could easily double the cost of running a node. And let's hope your users are ok with you giving all their posts, DMs included, to a third party.
I tried their demo out with some basics and quickly got false positives and false negatives. At first glance, I expect they are not suitable for an adversarial situation like a popular social network. Which is true of all the APIs like this I've looked at.
And then there's the cost. Want to use Hive for your own Bluesky instance? You'll have to talk to a salesperson, meaning that it's expensive. Just at a guess, paying for AI moderation could easily double the cost of running a node. And let's hope your users are ok with you giving all their posts, DMs included, to a third party.