Re: salaries, don't forget that there's also salaries working on the content moderation problem - taking the huge datasets that the content moderators are reviewing and applying machine learning and content ID and I have no clue what else more to automatically filter out a lot of the content, ideally with such a high certainty rate that the moderators don't see it in the first place.

I want to believe the content moderators only end up seeing the 1% of content that is dubious, or brand new content (not published and hashed earlier). I don't understand in that regard why the one video mentioned in the article (about the iguana) wasn't marked as "ok" in the systems, or marked as "on hold for law enforcement" so the moderators didn't have to see it again.