Got few interesting bits of information.

Facebook and Youtube now send all the comments to their multistage AI systems. Mostly to properly teach and adjust them for now.

Especially important are longer posts and comments.

Main output of first stage presently is:

• How positive posted content is towards "company goals" (fuzzy)
• How advertise friendly it is (it means approximately how much company can make on ads placed around)
• How tolerant it is towards LGBT and specific minorities
• How socially engaging it is (AI judges if it can bring additional comments or viewers)
• How much it will be touching and involving children (company want to see more children even near inappropriate content)
• Language level used (AI do not want to see too complex comments or posts, but too simple is also not good)

Before US elections they are activating next stage that will analyze your comment and adjust your profile accordingly.

Main output of new stage will be how your post matches mainstream views (as company see them). If it does not - it will be routed for further analyzing by additional NNs. Results will be used for comments sorting or even automatic flagging as spam.

If lot of your comments constantly do not match mainstream views - your profile will be flagged and additional level of AI will be analyzing 100% of new comments, moderators will check around 10-15% of most questionable comments.