I believe it means handling moderation yourself rather than relying on platform providers to do it for you. You have to choose your peers that you receive content from carefully rather than rely on platform moderators to filter content.
About the only thing considered universally intolerable and subject to administrative removal, at least on USENET's alt.* hierarchy, was spam. But spammers could spam faster than cancellers could cancel, so it eventually overwhelmed the platform.
Essentially, each user is just writing their own diary. If you follow someone, your client periodically asks around for a copy of that person's diary.
No-one's gonna ask for the spammers' diaries.
I guess you could design a group so at first only a few get hit with junk from a new member and as long as nobody stop listening, others would eventually accept the new member as well.
Just thinking on this use case as USENET was discussed elsewhere here as an analogy, maybe it's not relevant at all.
Scuttlebutt clients typically request the feeds of anyone you follow, plus anyone _they_ follow. (This is called “2 hops”.) If you only follow Alice, and Alice follows Bob, you'll also see Bob's messages. Then if you choose to follow Bob, you may also see Charlie's messages; and so on.
It builds a social network in the original sense — neither a bubble with sharply-defined edges, nor a global free-for-all.
I see messages from people I don't really agree with (because I follow someone who follows them), but they're not strangers with wholly-incompatible worldviews.
And because they're not strangers, and because messages tend to be slow and considered, I'm more likely to engage constructively with their different opinion (even if that's just by listening). It's anti-divisive.
The downside may be that, presumably, this puts everyone into a bubble even more than Facebook did? At least there is no algorithmic meddling or centralized content control.