This doesn’t make sense to me. Their automated content moderation is so good that it’s unable to detect “almost certainly disturbing shit”? What kind of amazing automation only works with subtleties but not certainties?
So the things that are flagged for human review would be on the boundary, but trend more towards disturbing than legitimate, on the grounds that the human in the loop is there to try and publish as many posts as possible, which means sifting through a lot of disturbing stuff that the AI is not sure about.
There’s also the question of training the models - the classifiers may need labelled disturbing data. But possibly not these days.
However, yes, I expect the absolute most disturbing shit to never be seen by a human.
—
Again, literally no experience, just a guy on the internet pressing buttons on a keyboard.
They'd prefer that as many posts are published, but they probably also don't mind some posts being removed if it meant saving a buck. When canada and australia implemented a "link tax", they were happy to ban all news content to avoid paying it.
This probably means fewer human reviewers reviewing a firehose, not the same number of human reviewers reviewing content at a slower rate.