Some brief thoughts, in order that they arose (incomplete as well, but posted anyway):
Agreed that incentives can produce unwanted behaviors and stepping back can be wise, though I think it’s worth an experiment (like the at-AI feature).
Would be interesting to have AI extract an argument into one of the structured argument formatting tools, like Kialo. Such tools are great, but historically the primary barrier is that it’s too cumbersome to set up and maintain them. Would be neat to have something like that embeddable in LW.
I notice for myself that I can be in two frames of mind:
I actually prefer being in mindset #2. Feels freeing and generous. With mindset #1, I often gravitate strongly toward criticism, since it’s easier to pounce on something demonstrably wrong than to contribute something new and valuable.
Re: (1) I operate in this manner at times too; my internal QC threshold changes based on mood and other factors though. Re: (2) still with the latter frame in mind, I will "search" mentally and remark "what remains to be said" or more specifically something akin to "what angle has not been pursuing here that makes the picture richer or more complete". On criticism, imagining that there are errors to be found w/ the framing, premises, evidence presented, or evidence synthesis, etc..., and then "finding" and sharing them via comments or remarks (on LW or in conversation) is one elicitation method I use, especially if the levels of "epistemic humility" in the presenter/writer seem low.
Should there be a norm, "It's okay to @-mention another user (even a famous/busy person) in a comment, just to ask them to comment on a post"?
(Do famous/busy LW users want to have other humans direct their attention to posts that they might comment on?)
We've noticed that LessWrong has been declining and is saturated with content about AI. There are many hypotheses explaining this decline: community maturity; competition with Substack, AI chat, and old posts; that we're a victim of our own success as our ideas have entered the water supply; an eternal september effect as the AI boom drives new participants who come for the AI, not the rationality. I suspect they're all part of the problem.
One solution has been to pay writers for posts. LW has prizes and a culture of commissioning posts or reviews out of personal funds, and now Inkhaven. My experience of having received commissions is that they didn't produce my best work. In fact, I think the consistent pattern is that people would approach me because of the posts I wrote for free that they liked, offer me money to write a post catered to their interests, and then feel disappointed that the content or quality wasn't what they were hoping for, or that the post didn't get more karma. I genuinely appreciate their generosity and willingness to take a risk on my writing. However, that disappointment was one of the nails in the coffin for my own period of writing on LessWrong.
Paying for quality content to revitalize an intellectual community is tricky in a way that's analogous to using political donations effectively. I think it's worth considering in detail what sustains the motivation to write over the long term, so that we can better target our interventions.
LessWrong's top authors comment less on each others' posts than they did in the past
My hypothesis is that what's been lost isn't writing, it's an engaged readership. Looking back at Overcoming Bias, early LessWrong posts, and SlateStarCodex, it seemed to me that the top authors in that era commented much more often on each others' posts than they do now. This fits with my impression that what sustains an amateur writing community isn't the prospect of earning money, but a sense of participating in an intense, running conversation with a community they are influenced by and care about influencing. People aren't just writing less, they are reading each others' work less, and everybody starts to feel like their writing is primarily interesting to trolls.
With Claude Code, I used the LW API to scrape all posts and comments going back to 2010, which confirmed my hypothesis. Posts are growing, but comments are declining, and top authors comment less on each others' posts.
Year
Posts
Comments
Comments per post
Top-writer peers per top post
Top posts with ≥3 peers
Top-writer cross-comment pairs
Top-writer comments per post written
Comments that are replies
Replies per top-level comment
2009
861
46,563
54.1
12.9
90%
1,300
30.9
75%
3.6
2010
1,163
86,517
74.4
15.7
98%
2,240
47.9
84%
5.2
2011
3,023
106,669
35.3
15.1
98%
2,304
15.5
79%
3.6
2012
2,657
119,920
45.1
10.9
98%
1,632
18.4
82%
4.3
2013
2,019
96,557
47.8
8.4
86%
1,476
22.0
82%
4.4
2014
1,820
71,056
39.0
6.1
78%
1,132
15.0
81%
4.1
2015
2,022
56,069
27.7
4.0
54%
914
9.7
79%
3.7
2016
1,315
25,347
19.3
4.4
68%
662
11.5
73%
2.5
2017
1,654
20,766
12.6
6.3
76%
869
6.9
74%
2.7
2018
1,647
19,828
12.0
8.5
94%
1,109
9.9
67%
2.1
2019
2,084
24,363
11.7
8.5
98%
1,392
8.8
65%
1.9
2020
3,031
35,597
11.7
7.0
88%
1,320
7.5
62%
1.6
2021
3,177
42,979
13.5
6.7
90%
1,126
8.7
65%
1.8
2022
4,412
51,159
11.6
7.4
88%
977
7.6
63%
1.7
2023
5,231
60,277
11.5
5.8
82%
865
6.8
65%
1.8
2024
4,203
42,029
10.0
4.3
66%
687
5.9
62%
1.6
2025
5,680
45,603
8.0
5.4
76%
837
6.7
62%
1.5
2026*
6,020
41,667
6.9
4.6
60%
744
4.4
58%
1.3
*2026 is through November 20.
Definitions. Data: all LessWrong posts and comments from the public GraphQL API (events, drafts and deleted comments excluded). Top writers for a year are the 50 authors with the highest total post karma that year; "peers" means top writers of that year or the previous one, excluding the post's author. Top post means one of the year's 50 highest-karma posts. Peers per top post is the mean number of distinct peers who commented on it. Cross-comment pairs counts distinct (commenter, author) pairs in which a top writer commented on another top writer's post that year. Comments per post written is the year's top writers' total comments divided by their total posts. Writers are ranked within each year rather than by raw karma, because LW 2.0's vote weighting inflated karma. Comment-only users with high standing are not counted as top writers.
Paying for posts: too little money, too much pressure
Prizes and commissions on LW have, in my experience, been on the order of hundreds to low thousands of dollars. That's a meaningful sum of money for a single post. However, prizes are seasonal and the amount you can win is limited. Commissions are rare and unpredictable. The money isn't enough to live on. And it's also generally a distraction from whatever else you could have been doing to make money: your job, networking, training. Putting your effort into pursuing prizes and fulfilling commissions has an opportunity cost. That opportunity cost is higher for the most capable writers, who could be doing such writing on substack or perhaps taking a job that would offer them steadier pay, more prestige, and more access.
My experience of prizes is that earning them was fun and a source of pride. I'd already written the post, so receiving a prize for it was just an added bonus. The "best of LessWrong" prize comes with a review process. The benefit isn't just the money, it's a second round of recognition by a valued community, and the sense of ceremony.
Commissions, by contrast, felt to me like they came with frontloaded pressure to deliver a high-quality post on a deadline, which when published, lands with an unceremonious thud and often drops out of sight within a day. There is no recognition for the commissioner, beyond perhaps an acknowledgement in the post. Instead of recognizing that community members are commissioning each other's writing, which is a rare and special thing, nearly the whole interaction between commissioner and author lives and dies in DMs, emails, and phone calls.
Takeaway: rebuild the LessWrong comments
I think that anyone who's interested in revitalizing LessWrong should start with the comment section, particularly top authors. Reply to posts, and reply to comments, especially by authors with less standing than you. Treat authors here like they have something worth thinking about. Act like there might be a conversation worth having in the comment section, and write comments that stand a chance of eliciting it.
LessWrong could design prizes for good comments. $1000 pays for 200 $5 prizes. We could have a "best comment section" award. Events to celebrate good conversations on LessWrong could be held weekly or monthly. I could imagine a "comment of the week" or "commenter of the month" award (and probably better to celebrate more people, not just one). We could have an "online hangout" time, where everybody shows up at the same time to comment on each others' posts, so that you can feel like you're participating in a live chat with the site. The site could find a way to elevate the visibility of the most thriving comment threads. Since the site's experimenting with integrating AI into the site, perhaps it's worth figuring out how to build AI summaries of those threads so that new participants can jump into them, since it's harder to follow a conversation from the post through a nested comment chain.
Good comments, especially by top authors, can make the author of a post feel like they've been taken seriously by a community they cared enough about to write for. It makes participants feel like they're participating in a thriving community, and that can be a self-fulfilling prophecy. Comments can also be a lower-stakes way to start writing for the community. Some of my own best-received writing on this site has been comments, including those that sparked my own most influential posts.
Do you think that improving the LessWrong comments is an underrated way to revitalize the site? How would you approach it? Do you think the primary issue is the presence of poor comments (rude, off-topic, etc), the lack of peer engagement amongst high-quality authors, or something else? I'd love to hear what you have to say in the comments!