Recommendation Features on LessWrong

To­day, we’re rol­ling out sev­eral new beta fea­tures on the home page, which dis­play recom­mended posts to read. The first is a Con­tinue Read­ing sec­tion: if you start read­ing a multi-post se­quence, it will sug­gest that you read an­other post from that se­quence. The sec­ond is a From the Archives sec­tion, which recom­mends highly-rated posts that you haven’t read, from all of LessWrong’s his­tory.

To use these fea­tures, please en­sure you are logged-in.

Con­tinue Reading

Se­quences are a mechanism on LessWrong for or­ga­niz­ing col­lec­tions of re­lated posts. Any­one can cre­ate a Se­quence from the library page. If you write a se­ries of posts and add them to a Se­quence, they will have Pre­vi­ous and Next links at the top and bot­tom; if you cre­ate a Se­quence out of posts by other au­thors, they will have Pre­vi­ous and Next links for read­ers who came to them via the Se­quence.

When you are logged in and read a post from a Se­quence, the first un­read post from that se­quence will be added as a recom­men­da­tion in the Con­tinue Read­ing sec­tion, like this:

If you de­cide not to finish, hover over the recom­men­da­tion and click the X but­ton to dis­miss the recom­men­da­tion. For logged-out users, the Con­tinue Read­ing sec­tion is re­placed with a Core Read­ing sec­tion, which sug­gests the first posts of Ra­tion­al­ity: A-Z, The Codex, and Harry Pot­ter and the Meth­ods of Ra­tion­al­ity.

From the Archives

The home page now has a From the Archives sec­tion, which dis­plays three posts ran­domly se­lected from the en­tire his­tory of LessWrong. Cur­rently, a post can ap­pear in this sec­tion if:

(1) you’ve never read it while logged in (in­clud­ing on old-LessWrong),

(2) it has a score of at least 50, and

(3) it is not in the Meta sec­tion, or man­u­ally ex­cluded from recom­men­da­tion by mod­er­a­tors. (We man­u­ally ex­clude posts if they aged poorly in a way that wouldn’t be cap­tured by votes at the time—for ex­am­ple, an­nounce­ments of con­fer­ences that have already hap­pened, and re­port­ing of stud­ies that later failed to repli­cate.)

Cur­rently, if a post is el­i­gible to ap­pear in the From the Archives sec­tion, it will ap­pear with prob­a­bil­ity pro­por­tional to the cube of its score.

Why Th­ese Features

LessWrong as a Re­pos­i­tory for “long con­tent”

Gw­ern’s about page has in­fluenced me a lot in think­ing about the fu­ture of LessWrong. Gw­ern uses the fol­low­ing quote:

The In­ter­net is self de­struc­t­ing pa­per. A place where any­thing writ­ten is soon de­stroyed by ra­pa­cious com­pe­ti­tion and the only preser­va­tion is to for­ever copy writ­ing from sheet to sheet faster than they can burn. If it’s worth writ­ing, it’s worth keep­ing. If it can be kept, it might be worth writ­ing…If you store your writ­ing on a third party site like Blog­ger, Live­jour­nal or even on your own site, but in the com­plex for­mat used by blog/​wiki soft­ware du jour you will lose it for­ever as soon as hy­per­sonic wings of In­ter­net la­bor flows di­rect peo­ple’s en­er­gies el­se­where. For most in­for­ma­tion pub­lished on the In­ter­net, per­haps that is not a mo­ment too soon, but how can the muse of origi­nal­ity soar when im­mo­lat­ing tran­sience brushes ev­ery feather?
-- Ju­lian As­sange (“Self de­struc­t­ing pa­per”, 5 De­cem­ber 2006)

Most of the con­tent on the in­ter­net is de­signed to be read and for­got­ten in a very short pe­riod of time. Ex­ist­ing dis­cus­sion plat­forms like Red­dit and many fo­rums even close threads au­to­mat­i­cally af­ter a cer­tain pe­riod of time to en­sure that all dis­cus­sion cen­ters around the most re­cent ac­tivity on the site.

One of the goals I have with LessWrong is to be a place where we can build on each other’s ideas over the course of mul­ti­ple decades, and where if you show up, you en­gage with the con­tent on the site in a fo­cused way, more similar to a text­book than a nor­mal in­ter­net fo­rum. And like our best text­books, good in­tro­duc­tions into core top­ics tend to stand the test of time quite well (e.g. the Feyn­man Lec­tures, which is still one of, if not the best in­tro­duc­tion to physics even 60 years later).

The Con­tinue Read­ing sys­tem is a key part of that goal, be­cause it makes it much eas­ier to use the site as a tool for fo­cused study, since con­tin­u­ing to read the se­quences you started is now one of the core ac­tions on the site.

The recom­men­da­tion sys­tem is also a key part of that goal, be­cause it cre­ates a way to dis­cover con­tent from the com­plete his­tory of LessWrong, in­stead of just the last week, which strikes me as a nec­es­sary com­po­nent to make col­lec­tive in­tel­lec­tual progress that can span mul­ti­ple decades. The best things for al­most any­one to read have very likely not been writ­ten in the past week.

LessWrong as a Nudge

Con­tinue Read­ing is a nudge to en­courage read­ing a few long things, rather than a lot of short things. Longer writ­ing al­lows top­ics to be ex­plored in greater depth, and also en­ables more ex­plicit de­ci­sions about what to read, since mak­ing one de­ci­sion per se­quence is a lot less work than mak­ing one de­ci­sion per post.

From the Archives is a nudge to read bet­ter posts. When we choose what to read, there is of­ten a re­cency bias; the best writ­ing of the past ten years will be bet­ter, on av­er­age, than the best writ­ing of the past week, but ac­tive con­ver­sa­tions will fo­cus on the most re­cent things. A good in­for­ma­tion diet con­tains a mix of re­cent writ­ing and of time­less clas­sics; by putting From the Archives on the home page, we are say­ing, on the mar­gin, read more of the past.

I also think that From the Archives will have a pos­i­tive effect on what peo­ple write on LessWrong. There are many good ideas in LessWrong’s archives, wait­ing to be built upon, which haven’t re­ceived at­ten­tion re­cently; my hope is that recom­men­da­tions of older posts will in­spire more good writ­ing.

Caveat: Addictiveness

Read­ing the lat­est posts on LessWrong is finite; you will run out of in­ter­est­ing-seem­ing re­cent posts, which cre­ates a nat­u­ral limit on time spent. Read­ing posts from the archives is effec­tively in­finite; LessWrong’s archives are deep enough that you prob­a­bly won’t ever run out of things to read. Th­ese new recom­men­da­tion fea­tures there­fore offer an op­por­tu­nity to spend a lot of time by ac­ci­dent. We’d rather you make a de­liber­ate de­ci­sion about what and how much to read on LessWrong.

If you find you’re spend­ing more time read­ing LessWrong’s recom­mended posts than you want, or ex­pect that you would spend more time than you want to, you can turn off the Con­tinue Read­ing and/​or From the Archives sec­tions by click­ing the gear icon. (This re­quires that you be logged in to save the set­ting.)

Th­ese Are Beta

Th­ese fea­tures are beta, and prob­a­bly have bugs. The From the Archives post se­lec­tion al­gorithm we’re cur­rently us­ing (based on post scores) seems to work okay for now, but scores are heav­ily af­fected by post visi­bil­ity as well as qual­ity, so some posts (es­pe­cially im­ported posts) aren’t be­ing recom­mended that should be, and post scores will suffer a pos­i­tive-feed­back effect where be­ing recom­mended causes posts to be recom­mended more. So, we ex­pect to rely less on the raw post score in the fu­ture, and more on other eval­u­a­tion mechanisms such as ask­ing users for ret­ro­spec­tive eval­u­a­tions of posts they’ve pre­vi­ously read, read-com­ple­tion and click­through rates, vote-to-views ra­tios, etc. The recom­men­da­tion al­gorithm is likely to be­come too com­plex to straight­for­wardly ex­plain, though its work­ings will always be know­able to those will­ing to dive into the source code.