@Safety
There needs to be more granularity than "sensitive" -- different levels of violence, sexuality, ability to specify these levels for specific topics, and so on.
User should be able to set a "profile vector" which we can dot-product with a "content vector", and set a threshold