If a user has <2500 karma, often when they publish a post, it will get literally zero time on the homepage. And so ~nobody will see it. Probably some of these posts are good but since they never appear on the homepage they never get to experience the virtuous cycle of getting upvoted ⇔ more people reading them.
Meanwhile when I go to lesswrong.com, I mostly see posts that either I've already read or I've already decided not to read, since I go to lesswrong.com multiple times per day.
It would be slightly better for me if the algorithm put more weight on recency when showing me posts, and it would be better for others because I'd sometimes read good recent-low-karma posts and upvote them so others could see them!
And I sometimes feel discouraged about posting because I think maybe I'll write a post people-would-like-if-they-saw but nobody will see it. The solutions are bad:
Sometimes I write things-that-should-be-posts as shortforms because it's easier for shortforms to be seen. (But I think the shortform algorithm should also be tweaked, similarly to the post algorithm.)
Sometimes when I publish something and the homepage-bar that day is too high for a brand-new 9-karma post to be shown (or it drops off the homepage after just a few minutes, before experiencing the virtuous cycle) I get a friend to upvote it. It's bad that I have to do that! And often those posts then do well; it's not like the algorithm is correctly predicting that the post won't do well.
And I'm privileged to have my posts start with 9 karma — it's way worse for people whose posts start at ≤6 karma!
(I've DMed the LW team about this several times, to no avail, but maybe if I post here it'll work. I feel like this might be a super easy way to make LW like 5% better — partially from showing-people-better-posts and partially from making authors feel better about writing posts.)
I think probably your original claim is right (i.e. tweak algorithm), but, one thing that occurs to me is that AI might be good enough to fine-tune on "estimate karma", and use that as more like a prior that helps it out over the first day but is then fades out by the 2nd day.
The percentage of requests like this that make it to implementation should be better now than it used to be. We've recently set up a pair of agents, one of which takes bug reports and feature requests from Intercom messages and turn them into Slack threads for discussion, and one of which takes Slack threads and turns them into branches to merge.
14Q Home
I complained about a similar thing to the LW team. I disliked that "recent discussion" got axed and that there was no option to just sort by recency (maybe with a slight bias by karma). IMO sorting by recency is fundamental to any forum, because it's the most egalitarian sort. That's why "shortform" and "questions" sections (the latter IIRC doesn't exist anymore) are so good.