Maybe I'm too optimistic, but I do feel like platforms could improve to better foster higher quality discussion, or at least better tailor users to see the type of content they enjoy, which is what I mentioned at the end of the article. It doesn't seem like there's been much innovation in that department, especially since Reddit virtually has a monopoly so there's not really much incentive for them to change anything.
So I don't disagree that techniques can improve quality. I just think the big bang for the buck is in the read layer.
Also, I believe that the most popular posts on Facebook are the ones like "Can you solve this math problem? 4x3-2=?" and stuff like that, so you overestimate how much quality matters to the bottom line which means that platforms are not incentivized to improve quality (not out of some evilness but out of pure economic incentives).
The only way these platforms can improve is by growing the moderation teams in ways that scale with the userbase and reinforce the neutrality you're looking for. It's a bureaucracy nightmare but you almost need mods for the mods level of checks and balances to ensure that you don't end up with a Reddit-like situation of 25 people pulling the strings of discourse.
FWIW HN shows the signs of this deterioration in the sense of a certain respect. I like to model forums by the percentage of their posts which lead to information gain. Ideally, I'd like to formulate this as some sort of K-L divergence with some sort of temporal decay on a binary variable. In a rough handwavey sense, I find that the information gain from many HN posts is high but HN comments is low, but it would be interesting to me if we could think up something more concrete.
The worst examples are posts on Google, Facebook, or Apple. Most commenters could be replaced by /r/subredditsimulatorGPT. On the other hand, it might be that I have just fully mined HN as an information source. It's likely this is true, in which case I should find some mechanism to reduce the content here.
One idea I have had is the idea of overlay networks - essentially the underlying data of the forum is the same, but we place intentionally different filtering mechanisms on the view layer. One could be "high karma users". Another could be "actively followed users". Anyway, just thoughts.