The modern internet is replete with feeds such as Twitter, Facebook, Insta, TikTok, Substack, etc. They're bad in ways but also good in ways. I've been exploring the idea that LessWrong could have a very good feed.

I'm posting this announcement with disjunctive hopes: (a) to find enthusiastic early adopters who will refine this into a great product, or (b) find people who'll lead us to an understanding that we shouldn't launch this or should launch it only if designed a very specific way.

You can check it out right now: www.lesswrong.com/feed

From there, you can also enable it on the frontpage in place of Recent Discussion. Below I have some practical notes on using the New Feed.

Note! This feature is very much in beta. It's rough around the edges.

Why have a "feed"?

Far more than with other LessWrong feature announcements^[1], I find myself starting to write this one from a defensive place: no, this feed is good actually.

The modern internet seems to have converged on feeds as a content presentation format that engages people effectively. There's the Facebook newsfeed, Twitter, Insta, TikTok, Substack, etc. I think these feeds are rightly regarded with suspicion – I think it's half that inherently the form factor, by being easier to consume^[2], tempts you away from other content forms you more endorse spending attention on, e.g. reading a proper book. The other half is that the feed operators, via their chosen algorithms, are out to get you. They want you addicted, spending as much time as you can rather than spending the optimal time to get the optimal amount of content at high efficiency on your attention.

Why then should LessWrong get in on this game?^[3] Well, because feeds actually have some useful properties as a way of finding and consuming content:

1. It's quick and easy to sample content at varying depths (e.g just read the title, read a sentence, read a paragraph) when choosing what to read. In contrast with a posts list, you only have the title, karma, author to go by unless you click into it or read the hover-preview (finicky and not present on mobile).

2. From a UI-perspective, it's easy to present many different kinds of content. Current LessWrong is crammed with sections: there's the section with a list of posts, quick takes, popular comments, featured content,  and then you have pages for wikitags, events/community, sequences, your bookmarks, stuff you started reading but didn't finish, etc. With luck, a user will find and look at the sections of interest to them. Yet I don't think we have that much luck and users miss a lot of stuff they'd like to have seen.

With a feed, you get a single place to focus your attention and we can serve up our best guess of what you'd like to see: here are some of the latest posts on topics you like, and by the way, here's an event in your area you might like, also gentle reminder you bookmarked some stuff last week, and PSA here are some standout imported wiki pages from Arbital. Due to the nature of a feed, it's easy to skip and it's easy for us to learn what you do and don't like^[4].

To me, this is a lot of the promise of feeds. It makes it a lot simpler to present users with content of interest across diverse content types.

3. There are various conversations that seem good for people to be having that aren't and shouldn't be tied to a single post and being in its comments section. Someone writes a paragraph or two and then people discuss it. Twitter is where I see this happening, sometimes well. LessWrong's Quick Takes move us in that direction too and we get some of it.

My feeling is that feeds are a good way for people to continue to participate in and follow along ongoing conversations. You open the feed, and it shows you the latest in the threads you were engaging in.

Contrast: a few months ago I subscribed to comments on a few LessWrong users^[5]. Previously I'd subscribed to posts by some authors and this generated a small manageable number of notifications. Not so with comments. Suddenly my notifications drawer was filled with comments. I wanted to read them, but it wasn't a great experience^[6]. I realized that what I really wanted was a feed of comments by users who typically wrote comments I wanted to see.

Making Our Feed Good Actually

Incentive-wise, I think LessWrong escapes the worst of the pressures to make an addictive feed but perhaps not all of them. There's no direct monetization of user attention whereby marginal scrolls equate strongly to marginal dollars. That said, donors are fond of activity graphs going up and I personally get clout if I can boast that the feature I built was highly successful.

Beyond claiming that there's less incentive to be terrible, there are two principles I've followed so far in the design of the LessWrong feed that I think help.

1. Transparency. I'll tell you how the algorithm is deciding what to show you^[7], and you can decide/complain if you think it's not working in a way you endorse.
2. Control. Beyond transparency, the user has a lot of ability to change what the feed shows and customize.
3. Encouraging longform content creation and consumption. I elaborate in the next section.

There are pressures against 1 and 2, supposing they're even helpful. First, very few users adjust settings even when it would give them something they like more, so the LessWrong team still has to choose defaults. Second, in the long-term I think algorithmic output quality might be improved by using relatively opaque AI/ML methods as opposed to the simple dumb rules the current system uses, and that makes transparency harder.

A model of feeds taking over and why that's bad

Suppose that as you consume content (written, audio, visual, etc), you experience enjoyment on a zero to ten scale^[8]. I think what we have observed over decades is a trend towards shorter and shorter content with more concentrated peaks of reward. Movies replaced by TV episodes replaced by 5-minute YouTube videos replaced by 15-second TikTok reels. Books replaced by comics replaced by tweets and memes. Food has gotten sweeter (higher caloric density) and content has too.

Rather than watch a 5-minute YouTube video with a single reward high of "8", a person can watch 10 TikTok reels with rewards like [2, 5, 9, 3, 7...] and get some nice intermittent reinforcement gambling action in there too. The format allows you to stop engaging with one piece of content and seek another very cheaply as soon as you are bored.

I can see this being in fact very bad. I think that for some content, the payoff requires you to wade through some effortful or less rewarding set up. If people are trained to bounce off of that and jump to more readily rewarding content, people lose the ability to stick through the harder stuff and there's less reward for creating it.

Even if you suppose that all content could be made short and punchy and high-reward-density, it is still not good to put such a strong selection pressure on that over other things like is your content valuable, well-explained, etc. 

The other content platforms are entirely happy to engage in Molochian destroying-everyone's-general-intelligence games in exchange for money since, after all, it works.  At least in the moment, users "want" higher density reward and opt for that. And if your platform doesn't do it, you lose out to those who do.

It's this kind of model that makes me wonder if LessWrong shouldn't join the party. You can't just use the One Ring for good. But I don't know.

If people would like to mindfully try out the New Feed and other feeds they use, and report back on the experience, that'd be swell.

The third principle I've had in mind with design of the feed so far is to make it so the feed facilitates doing long-form content consumption even when habitually using the feed. This means things like:

• The feed makes it easy to read long comments seamlessly, we're not punishing content longer than N characters, etc.
• It's easy to launch into reading a long post or comment mid-feed, either in-place, in a nice popup view, or with navigation in a way that if you leave, you're not worried about losing your place
• While the feed contains quick takes and comment threads, a lot of what we're recommending is long form content such as posts, sequences, long good wikitags, like those imported from Arbital.

Another direction I'm keen on is using the Feed as a place to remind people of content they've bookmarked but haven't read yet, or stuff they started reading but didn't finish (either a long post or not finishing all posts in a sequence). LessWrong has had a (not that successful) Continue Reading section that I think just needed more iterations rather than it was a bad idea (proof: Kindle and Netflix have "continue reading" sections that people do use).

Metrics-wise, I'd more want to target "time spent consuming LessWrong content that was initiated from the feed"^[9] and "how quickly engaging with the feed led to someone reading something substantive" rather than just "time spent on the feed itself". This is saying the feed should function a lot as content discovery mechanism rather than solely as a final destination.

Unregretted User Minutes

Unregretted user minutes seems like a good metric if you can measure it well. I have some confidence that LessWrong users can judge after the fact whether they endorse time they spent doing something or not.

I think it's not crazy that after N minutes or M items in your feed, you get a card which polls you on how you're feeling about your feed usage. This over time can be monitored as a signal as we make changes to the feed.

It's really my least favorite argument, but perhaps it's still valid to say that given LessWrong is competing for people's attention with Twitter, etc., we should have a feed too. I don't like it, but also personally when it's late in the evening and I'm unwinding, I find I like having the LessWrong feed as an alternative to e.g. Twitter. It makes me think of how we're often told that people say "LessWrong is the form of procrastination they feel least bad about".

Proceeding with Caution

I think there's reason to proceed but do so with caution. I feel like LessWrong's current content-discovery mechanisms and user attention guidance are inadequate and are not getting people to see everything they'd want to see. Feeds have promise. I have enough hope to try doing a feed and hope to execute it with enough good judgment and integrity that it's pretty great.

Practical Tips about the New Feed

You can try out the new (very much still under construction) LessWrong feed at: https://www.lesswrong.com/feed

I'm seeking beta users who are interested in regular interviews/chats about what is and isn't working in making this a valuable feature. Please comment or message if you are.

Avoiding Confusion

Great products are self-explanatory and don't require manuals, walkthroughs, etc. Unfortunately, I don't think The Feed has hit that bar yet, so here are some things you might want to know:

Branching Comments Linear Slices

The LessWrong feed [currently] makes an opinionated choice about how to display comments. While comment threads by nature branch, the feed only ever displays a single "slice" (parent-child-grandchild-greatgrandchild) in one block. I believe this makes the reading experience a lot more straightforward. Other branches can be accessed by clicking on the comment icon (that has count of direct children), or waiting for other branches to appear in the feed^[10].

Choose How Much Text To Display

A big question for the feed is how much content to display by default (e.g if a comment is 657 words, how many do you show?) and what the expand/read more behavior is.

At this stage, I've made this highly configurable so early adopters can experiment and help me figure out what works best as a default. You can adjust these in the New Feed settings.

Following/Subscribing

This is not currently emphasized, but LessWrong and the feed support subscribing/following and following users. Anyone you follow is by default more likely to show up in your feed and the settings allow you to increase how heavily they show up. 

Although I haven't made it, I think there's merit to having a "Following"/"Subscribed" feed view where it's only content from people you follow^[11]. If nothing else, it means people have an easy alternative if they don't like the recommendations you're giving them. I might want to add this.

I anticipate that the new feed will supercede the Subscribed tab. Following or subscribing to a user in any form will treat you as following them for purposes of display in the new feed.

Settings, Settings Everywhere

The settings menu currently offers a lot of choice, though I eventually expect to figure out good defaults and hide or hardcode the settings.

As per the title, this feature is in beta and I'm seeking feedback. Any and all appreciated! And remember this is LessWrong: we put the doom in doomscrolling.

1. ^{^}

   With the exception of announcing that we are experimenting with the frontpage posts list including suggestions by a recommender system – something closely related in spirit to this feature.

2. ^{^}

   Feeds are eminently "low-brain" compatible.

3. ^{^}

   We are trying to not be evil.

4. ^{^}

   There are dangers here that I hope are navigable.

5. ^{^}

   I think they included @habryka and @gwern.

6. ^{^}

   In the notifications drawer it's either with an easy to lose hover or you're expensively navigating away, plus it's harder to get context on any comment.

7. ^{^}

   At this time, the feed is very transparent with the exception of suggested posts drawn from the ~opaque recommendation engine that also fuels the Enriched and Recommended post lists. That engine is pretty good and there are worlds in which we use across other content too, or something like it.

8. ^{^}

   Or better yet a minus ten to ten scale to allow for anti-enjoyment but that's not important for this argument.

9. ^{^}

   i.e. minutes spent reading on the feed + minutes spent reading a post you opened from the feed.

10. ^{^}

    If a comment thread with multiple branches deemed of interest, the feed is likely to eventually show them all. Parent comments that have been seen will be collapsed/deemphasized.

11. ^{^}

    Personally, on Twitter I go back and forth between "For You" and "Following" with it being a toss-up on what has better content.