Ruby

• Ruby 14 Nov 2019 8:40 UTC

  4 points

  in reply to: Ben Pace's comment on: Ben Pace’s Short­form Feed

  That clar­ifies it and makes a lot of sense. Seems my ob­jec­tion rested upon a mi­s­un­der­stand­ing of your true in­ten­tion. In short, no wor­ries.

  I look for­wards to figur­ing this out to­gether.

• Ruby 14 Nov 2019 8:25 UTC

  2 points

  in reply to: Ben Pace's comment on: Ben Pace’s Short­form Feed

  I’m cur­rently work­ing through my own thoughts and vi­sion for tag­ging.

  If and when we end up build­ing a tag­ging sys­tem on LessWrong, the goal will be to dis­t­in­guish the main types of post peo­ple are in­ter­ested in view­ing, and cre­ate a limited num­ber of tags that de­ter­mine this. I think build­ing this will mainly help users who are view­ing new con­tent on the front­page, and that for much more gran­u­lar sort­ing of his­tor­i­cal con­tent, a wiki is bet­ter placed.

  I’m pretty sure I dis­agree with this and ob­ject to you mak­ing an as­ser­tion that makes it sound like the team is definitely de­cided about what the goal of tag­ging sys­tem will be.

  I’ll write a proper re­sponse to­mor­row.

• Ruby 12 Nov 2019 0:09 UTC

  8 points

  in reply to: Raemon's comment on: Rae­mon’s Short­form Feed

  I re­ally like the green-un­read on post pages. On Re­cent Dis­cus­sion I have so much of it that I think I don’t re­ally pay at­ten­tion to it.

• Ruby 11 Nov 2019 20:01 UTC

  2 points

  in reply to: Pattern's comment on: LW Team Up­dates—Novem­ber 2019 (Sub­scrip­tions & More)

  Fixed, thanks!

• Ruby 9 Nov 2019 2:07 UTC

  4 points

  in reply to: Vaniver's comment on: [Team Up­date] Why we spent Q3 op­ti­miz­ing for karma

  I find this com­pel­ling (along with the “find­ing out which things mat­ter that you didn’t re­al­ize mat­tered) and think this is a rea­son for us to be­gin do­ing A/​B test­ing some­time in not too dis­tant fu­ture.

• Ruby 9 Nov 2019 2:04 UTC

  12 points

  in reply to: johnswentworth's comment on: [Team Up­date] Why we spent Q3 op­ti­miz­ing for karma

  I see the spirit of what you’re say­ing and think there’s some­thing to it though it doesn’t feel com­pletely cor­rect. That said, I don’t think any­one on the team has ex­pe­rience with that kind of A/​B test­ing loop and given that lack of ex­pe­rience, we should try it out for at least a while on some pro­jects.

  To date, I’ve been work­ing just to get us to have more of an an­a­lyt­ics-mind­set plus ba­sic thor­ough an­a­lyt­ics through­out the app, e.g. track­ing on each of the fea­tures/​but­tons we build, etc. (This wasn’t triv­ial to do with e.g. Google Tag Man­ager so we’ve ended up build­ing stuff in-house.) I think try­ing out A/​B test­ing would likely make sense soon, but as above, I think there’s a lot of value even be­fore it with more dumb/​naive an­a­lyt­ics.

  We tri­alled Ful­lS­tory for a few weeks and I agree it’s good, but also we just weren’t us­ing it enough to jus­tify it. LogRocket offers monthly sub­scrip­tion though and likely we’ll sign up for that soon. (Once we’re ac­tu­ally us­ing it fully, not just tri­al­ling, we’ll need to post about it prop­erly, build opt-out, etc. and be good around pri­vacy—already in trial we hid e.g. vot­ing, user­names.)

  To come back to the open­ing points in the OP, we prob­a­bly shouldn’t get too bogged down try­ing to op­ti­mize spe­cific sim­ple met­rics by get­ting all the but­tons perfect, etc., given the un­cer­tainty over which met­rics are even cor­rect to fo­cus on. For ex­am­ple, there isn’t any clear met­ric (that I can think of) that definitely an­swers how much to fo­cus on bring­ing in new users and get­ting them up to speed vs build­ing tools for ex­ist­ing users already pro­duc­ing good in­tel­lec­tual progress. I think it’s cor­rect that have to use high-level mod­els and fuzzier tech­niques to think about big pro­ject pri­ori­ti­za­tion. A/​B tests won’t re­solve the most cru­cial un­cer­tain­ties we have though I do think they’re likely to hugely helpful in re­fin­ing our de­sign sense.

• Ruby 9 Nov 2019 1:51 UTC

  6 points

  in reply to: cousin_it's comment on: [Team Up­date] Why we spent Q3 op­ti­miz­ing for karma

  This is roughly the pro­ce­dure we usu­ally fol­low.

LW Team Up­dates—Novem­ber 2019 (Sub­scrip­tions & More)

• Ruby 8 Nov 2019 2:33 UTC

  8 points

  in reply to: johnswentworth's comment on: [Team Up­date] Why we spent Q3 op­ti­miz­ing for karma

  Try­ing to op­ti­mize a met­ric with­out even hav­ing a test frame­work in place adds a lot of ev­i­dence to that story—cer­tainly in my own start-up ex­pe­rience, we never had any idea what we were do­ing un­til well af­ter the test frame­work was in place (at any of the com­pa­nies I’ve worked at). An­a­lyt­ics more gen­er­ally were also always cru­cial for figur­ing out where the low-hang­ing fruit was and which pro­jects to pri­ori­tize, and it sounds like you guys are cur­rently still fly­ing blind in that de­part­ment.

  I think I agree with the gen­eral spirit here. Through­out my year with the LessWrong team, I’ve been pro­gres­sively build­ing out an­a­lyt­ics in­fras­truc­ture to re­duce my sense of the “fly­ing blind” you speak of. We’re not done yet, but I’ve now got a lot of data at my finger­tips. I think the dis­agree­ment here would be over whether any­thing short of A/​B test­ing is valuable. I’m pretty sure that it is.

• Ruby 8 Nov 2019 2:28 UTC

  9 points

  in reply to: habryka's comment on: [Team Up­date] Why we spent Q3 op­ti­miz­ing for karma

  A num­ber of these pro­jects were already on our docket, but less visi­ble is the pro­jects which were de­layed and the fact that those se­lected might not have been done now oth­er­wise. For ex­am­ple, if we hadn’t been do­ing met­ric quar­ter, I’d like have spent more of my time con­tin­u­ing work on the Open Ques­tions plat­form and much less of my time do­ing in­ter­views and talk­ing to au­thors. Ad­mit­tedly, sub­scrip­tions and the new ed­i­tor are pro­jects we were already com­mit­ted to and had been work­ing on, but if we hadn’t thought they’d help with the met­ric, we’d have de­layed it to the next quar­ter the way we did with many of other pro­ject ideas.

  We did brain­storm how­ever, but as Oli said, it wasn’t easy to come with any ideas which were ob­vi­ously much bet­ter.

• Ruby 8 Nov 2019 2:21 UTC

  13 points

  in reply to: habryka's comment on: [Team Up­date] Why we spent Q3 op­ti­miz­ing for karma

  Heck­ling ap­pre­ci­ated. I’ll add a bit more to Habryka’s re­sponse.

  Separate from the ques­tion of whether A/​B would have been ap­pli­ca­ble to our pro­jects, I’m not sure why think it’s pointless to try to make in­fer­ences with­out them. True, A/​B tests are cleaner and more defini­tive, and what we ob­served is plau­si­bly what would have hap­pened even with differ­ent ac­tivi­ties, but that isn’t to say we don’t learn a lot when the out­come is one of a) met­ric/​growth stays flat, b) small de­crease, c) small in­crease, d) large de­crease, e) large in­crease. In par­tic­u­lar, the growth we saw (in­crease in ab­solute and rate) is sug­ges­tive of do­ing some­thing real and also strong ev­i­dence against the hy­poth­e­sis that it’d be very easy to drive a lot of growth.

  Gen­er­ally, it’s at least sug­ges­tive that the first quar­ter where we ex­plic­itly we fo­cus on growth is one where we see 40% growth from last quar­ter (com­pared to 20% in the pre­vi­ous quar­ter to the one be­fore). It could be a co­in­ci­dence, but I feel like there are still like­li­hood ra­tios here.

  When it comes to at­tri­bu­tion too, with some of these pro­jects it’s easy to get much more of an idea even with­out A/​B test­ing. I can look at the posts from au­thors who we con­tacted and rea­son­ably be­lieve coun­ter­fac­tu­ally would not have oth­er­wise posted and see how much karma that gen­er­ated. Same from Petrov Days and MSFP.

[Team Up­date] Why we spent Q3 op­ti­miz­ing for karma

• Ruby 31 Oct 2019 22:42 UTC

  4 points

  in reply to: Ben Pace's comment on: Aella on Ra­tion­al­ity and the Void

  Se­conded.

• Ruby 31 Oct 2019 16:47 UTC

  2 points

  in reply to: toonalfrink's comment on: steve2152′s Shortform

  Be­cause they’re in­ter­ested in weapons and mak­ing peo­ple dis­tinctly not safe.

• Ruby 31 Oct 2019 14:51 UTC

  9 points

  in reply to: steve2152's comment on: steve2152′s Shortform

  A friend in the AI space who vis­ited Wash­ing­ton told me that mil­i­tary lead­ers dis­tinctly do not like the term “safety”.

• Ruby 30 Oct 2019 22:35 UTC

  2 points

  in reply to: Ben Pace's comment on: On In­ter­nal Fam­ily Sys­tems and multi-agent minds: a re­ply to PJ Eby

  Re­cently clar­ified guidelines: https://​​www.less­wrong.com/​​posts/​​5conQhfa4rgb4SaWx/​​site-guide-per­sonal-blog­posts-vs-front­page-posts

• Ruby 30 Oct 2019 19:57 UTC

  3 points

  in reply to: Ericf's comment on: Theater Tick­ets, Sleep­ing Pills, and the Idiosyn­crasies of Del­e­gated Risk Management

  But sleep­ing pills aren’t just about stress, they’re very con­cretely about putting you to sleep which they do.

• Ruby 30 Oct 2019 19:48 UTC

  18 points

  in reply to: MakoYass's comment on: Turn­ing air into bread

  No mat­ter how many false phys­i­cal con­straints we over­turn the sec­ond law of ther­mo­dy­nam­ics seems to guaran­tee (this is de­bat­able) that we will even­tu­ally hit a wall . . .

  Granted that we will even­tu­ally hit a wall, there’s a good chance the wall is so un­be­liev­ably far off that it might as not ex­ist for an­other mil­lion or billion years and al­low for as­tro­nom­i­cal (liter­ally) amounts of growth. Heck, even what we get out of the Earth alone could be in­creased mul­ti­ple or­ders of mag­ni­tude. Sup­pose there’s a point at think­ing about slow­ing down, I think that point is very far away.

  I’ll quote a bit from my sum­mary of Eter­nity in Six Hours, which I find cred­ible:

  • Trav­el­ling at 50c% there are 116 mil­lion galax­ies reach­able; at 80% there are 762 mil­lion galax­ies reach­able; at 99%c, you get 4.13 billion galax­ies.

    • For refer­ence, there are 100 to 400 billion stars in the Milky Way, and from a quick check it might be rea­son­able to as­sume 100 billion is the av­er­age galaxy.

      • The abil­ity to colonize the uni­verse as op­posed to just the Milky Way is the differ­ence be­tween ~10^8 stars and ~10^16 or ~10^17 starts. A fac­tor of 100 mil­lion.

  Similarly, the sun’s es­ti­mated en­ergy out­put is 3.8x10^26W (Joules per Se­cond) whereas civ­i­liza­tion’s cur­rent en­ergy us­age is es­ti­mate at ~10^24J/​year in a re­cent year (2012 or 2014?). That’s some­thing like 9 or­ders of mag­ni­tude of more en­ergy that’s be­ing ex­pended than we cur­rently use (sim­plify­ing a whole bunch).

  There’s finite and there’s finite, some of those finite’s are freak­ing huge. I say let’s get ’em.

  But be­ing more se­ri­ous, if we think about the EV of differ­ent strate­gies, I think the EV con­tin­u­ing to pur­sue growth (as ja­son­crawford defines it) for the fore­see­able fu­ture is bet­ter than very pre­ma­turely try­ing to limit growth and be “sus­tain­able” notwith­stand­ing the risks that even­tu­ally there will be some kind of crunch.

  Ad­mit­tedly, I could be wrong about the limits of po­ten­tial tech­nolog­i­cal ca­pa­bil­ities. If for some rea­son we hit a a limit of what we can do far ear­lier, then there might be a wall far sooner than when we run out of en­ergy. But even such a wall seems at least quite a ways off.

• Ruby 30 Oct 2019 19:09 UTC

  4 points

  on: Theater Tick­ets, Sleep­ing Pills, and the Idiosyn­crasies of Del­e­gated Risk Management

  Good post. I like these ex­am­ples, though I find the names you’ve given to each type to not that be that evoca­tive and I ex­pect to strug­gle to re­mem­ber them or ex­plain them to oth­ers. For ex­am­ple, sleep­ing pills do some­thing real. Per­haps you mean some­thing like sugar (placebo) pills? Or an­tibiotics for a cold?

• Ruby 30 Oct 2019 18:25 UTC

  4 points

  on: Rugby & Re­gres­sion Towards the Mean

  I re­ally ap­pre­ci­ate see­ing this kind of ap­plied statis­ti­cal anal­y­sis to a stray in­ter­est­ing-sound­ing fact you heard.