Wow, I'm impressed! This is pretty close to how I imagined it, and it also seems simple enough for everyone to understand.
Essentially, by default you only see content recommended by someone you care about (i.e. in long term you care about the people you follow; and in short term you care about the person whose article you are reading right now). So people cannot insert themselves into debates forcefully.
I'm trying to imagine how Facebook would look like if they switched to this system (using the existing "like" button as the sign of approval). So when you post something on your wall, the comments you "liked" are displayed to all readers; the comments to didn't like are displayed only to friends of the person who posted them, and you are not allowed to remove any comment.
Sounds reasonable, assuming there is a visible difference between "the comments I didn't approve because I don't want to approve them" (e.g. the "hide" button), and "the comments I haven't approved because I haven't seen them yet".
The only possible form of "spamming" here is to annoy someone by posting many replies to their articles, and even then you are only annoying them privately. (There should be a way to block a user, that is "auto-hide" all their replies, so the only possible way of "spamming" would be posting many replies with many sockpuppets. This would take the usual attacker much more time than the attacked person.)
Maybe the disadvantage is that it kills the "linear debate of trivial comments"; the type of discussion where everyone only types a line or two, which best resembles how people chat, but maybe that's good. People who want to chat without writing an article-length reply might miss this feature.
So I guess my perfect system would be a combination of the Medium way, plus old-style linear discussion below the article, where all replies are invisible until approved by the author (optionally, the author could switch it to "auto-approve" with possibility to delete anything afterwards). Or, to make it more unified, every reply would start as a comment below the article, but you would have the checkbox "also show this reply on my homepage as an article". All approved replies would be displayed below the article, but replies longer than three lines (that includes full articles) would be shortened until you click to expand them.
A few months ago, Vaniver wrote a really long post speculating about potential futures for Less Wrong, with a focus on the idea that the spread of the Less Wrong diaspora has left the site weak and fragmented. I wasn't here for our high water mark, so I don't really have an informed opinion on what has socially changed since then. But a number of complaints are technical, and as an IT person, I thought I had some useful things to say.
I argued at the time that many of the technical challenges of the diaspora were solved problems, and that the solution was NNTP -- an ancient, yet still extant, discussion protocol. I am something of a crank on the subject and didn't expect much of a reception. I was pleasantly surprised by the 18 karma it generated, and tried to write up a full post arguing the point.
I failed. I was trying to write a manifesto, didn't really know how to do it right, and kept running into a vast inferential distance I couldn't seem to cross. I'm a product of a prior age of the Internet, from before the http prefix assumed its imperial crown; I kept wanting to say things that I knew would make no sense to anyone who came of age this millennium. I got bogged down in irrelevant technical minutia about how to implement features X, Y, and Z. Eventually I decided I was attacking the wrong problem; I was thinking about 'how do I promote NNTP', when really I should have been going after 'what would an ideal discussion platform look like and how does NNTP get us there, if it does?'
So I'm going to go after that first, and work on the inferential distance problem, and then I'm going to talk about NNTP, and see where that goes and what could be done better. I still believe it's the closest thing to a good, available technological schelling point, but it's going to take a lot of words to get there from here, and I might change my mind under persuasive argument. We'll see.
Fortunately, this is Less Wrong, and sequences are a thing here. This is the first post in an intended sequence on mechanisms of discussion. I know it's a bit off the beaten track of Less Wrong subject matter. I posit that it's both relevant to our difficulties and probably more useful and/or interesting than most of what comes through these days. I just took the 2016 survey and it has a couple of sections on the effects of the diaspora, so I'm guessing it's on topic for meta purposes if not for site-subject purposes.
Less Than Ideal Discussion
To solve a problem you must first define it. Looking at the LessWrong 2.0 post, I see the following technical problems, at a minimum; I'll edit this with suggestions from comments.
I see these meta-technical problems:
Slightly Less Horrible Discussion
"Solving" community maintenance is a hard problem, but to the extent that pieces of it can be solved technologically, the solution might include these ultra-high-level elements:
As with the previous, I'll update this from the comments if necessary.
Getting There From Here
As I said at the start, I feel on firmer ground talking about technical issues than social ones. But I have to acknowledge one strong social opinion: I believe the greatest factor in Less Wrong's decline is the departure of our best authors for personal blogs. Any plan for revitalization has to provide an improved substitute for a personal blog, because that's where everyone seems to end up going. You need something that looks and behaves like a blog to the author or casual readers, but integrates seamlessly into a community discussion gateway.
I argue that this can be achieved. I argue that the technical challenges are solvable and the inherent coordination problem is also solvable, provided the people involved still have an interest in solving it.
And I argue that it can be done -- and done better than what we have now -- using technology that has existed since the '90s.
I don't argue that this actually will be achieved in anything like the way I think it ought to be. As mentioned up top, I am a crank, and I have no access whatsoever to anybody with any community pull. My odds of pushing through this agenda are basically nil. But we're all about crazy thought experiments, right?
This topic is something I've wanted to write about for a long time. Since it's not typical Less Wrong fare, I'll take the karma on this post as a referendum on whether the community would like to see it here.
Assuming there's interest, the sequence will look something like this (subject to reorganization as I go along, since I'm pulling this from some lengthy but horribly disorganized notes; in particular I might swap subsequences 2 and 3):
(Meta-meta: This post was written in Markdown, converted to HTML for posting using Pandoc, and took around four hours to write. I can often be found lurking on #lesswrong or #slatestarcodex on workday afternoons if anyone wants to discuss it, but I don't promise to answer quickly because, well, workday)
[Edited to add: At +10/92% karma I figure continuing is probably worth it. After reading comments I'm going to try to slim it down a lot from the outline above, though. I still want to hit all those points but they probably don't all need a full post's space. Note that I'm not Scott or Eliezer, I write like I bleed, so what I do post will likely be spaced out]