a) people are more likely to have gumption to be negative in the absence of social cues of another person in their presence (global effect, whether or not user is mentally ill)

b) those motivated and possessing time to post are more likely to be suffering from issues (where those not mentally ill may just not engage because they have other things going on in life)

c) unlike the physical space where filters exist, you're often exposed to these people when navigating to entirely innocuous content (e.g. a kids video).

d) way way easier to find people with same or adjacent point of view on internet, reinforcing beliefs and potentially driving person further to extremes

Problem needs to be addressed from multiple fronts - as you say, more control for end users to tweak their experience (top down), alongside better platform-level filtering (bottom up), along with all the helpful designs therein like good defaults.