Lanrian

• Lanrian 5 Jan 2020 13:13 UTC

  10 points

  in reply to: Mark_Friedenbach's comment on: Do­minic Cum­mings: “we’re hiring data sci­en­tists, pro­ject man­agers, policy ex­perts, as­sorted weirdos”

  And yet here’s a ra­tio­nal­ist who up­turned global poli­tics sin­gle­hand­edly, and cred­its LessWrong with his suc­cess.

  Source? I’ve googled his name and LessWrong, but can’t find him say­ing any­thing about it.

• Lanrian 22 Dec 2019 23:20 UTC

  3 points

  on: Can fear of the dark bias us more gen­er­ally?

  This is just anec­do­tal, but me, a friend, and plau­si­bly Ran­dall Mun­roe are sig­nifi­cantly more so­cially risk-tak­ing at night than dur­ing other times of day. This might be di­rectly con­nected to the time of day, or just be a con­se­quence of sleep de­pri­va­tion.

  I also have ir­ra­tional fears dur­ing the night, some­times, but I would guess that this is largely due to be­ing sleepy, stupid, and alone, which causes me to be more sug­gestible to stray thoughts in gen­eral. I wouldn’t be sur­prised if dark­ness also con­tributes, though.

• Lanrian 18 Dec 2019 9:45 UTC

  2 points

  in reply to: lsusr's comment on: Per­sonal Ex­per­i­ment: One Year with­out Junk Media

  You can sculpt a ser­vice like this out of pretty much any ser­vice that keeps its old shows around. For ex­am­ple, you can use ublock ori­gin to re­move all side-bar video-sug­ges­tions from youtube (and also re­move com­ments, if you want). Then you can just for­bid your­self from go­ing to the home page (or au­to­mat­i­cally block it with some­thing like leech­block), and only ever ac­cess videos by do­ing search di­rectly. If you have a way of adding any search en­g­ine to your browser (which I recom­mend get­ting; I think there are easy ways to do this in most browsers, though I use vimium), you can add youtube.com/​re­sults?search_query= or net­flix.com/​search?q= or what­ever you want to it.

• Lanrian 16 Dec 2019 22:34 UTC

  4 points

  on: How time track­ing can help you prioritize

  If you want au­to­matic time track­ing for a mac (with good sup­port for man­ual as­sign­ment), and you’re will­ing to pay for it, I tested sev­eral op­tions two years back and de­cided that timing was best. I’m still happy with it. A nice thing with au­to­matic time-track­ing is that I have de­cent data for a year where I didn’t use it ac­tively at all.

  That said, I don’t want to ruin the call-to-ac­tion by forc­ing any­one to con­sider sev­eral op­tions. I haven’t tried tog­gle, but it seems like a good choice. Go forth and try time track­ing!

• Lanrian 16 Dec 2019 20:01 UTC

  1 point

  in reply to: TAG's comment on: When would an agent do some­thing differ­ent as a re­sult of be­liev­ing the many wor­lds the­ory?

  Wait, what? If com­pat­i­bil­ism doesn’t sug­gest that I’m choos­ing be­tween ac­tions, what am I choos­ing be­tween?

• Lanrian 16 Dec 2019 18:01 UTC

  1 point

  in reply to: TAG's comment on: When would an agent do some­thing differ­ent as a re­sult of be­liev­ing the many wor­lds the­ory?

  MWI is de­ter­minis­tic, so you can’t al­ter the per­centages by any kind of free will, de­spite what peo­ple keep as­sert­ing.

  Nei­ther most col­lapse-the­o­ries nor MWI al­low for su­per-phys­i­cal free will, so that doesn’t seem rele­vant to this ques­tion. Since the ques­tion con­cerns what one should do, it seems rea­son­able to as­sume that some no­tion of choice is pos­si­ble.

  (FWIW, I’d guess com­pat­i­bil­ism is the most pop­u­lar take on free will on LW.)

• Lanrian 12 Dec 2019 21:01 UTC

  2 points

  in reply to: toonalfrink's comment on: toon­alfrink’s Shortform

  We have ab­solutely no rea­son to be­lieve that be­hav­ior cor­re­lates with con­scious­ness.

  The strong ver­sion of this can’t be true. You claiming that you’re con­scious is part of your be­havi­our. Hope­fully, it’s ap­prox­i­mately true that you would claim that you’re con­scious iff you be­lieve that you’re con­scious. If be­havi­our doesn’t at all cor­re­late with con­scious­ness, it fol­lows that your be­lief in con­scious­ness doesn’t at all cor­re­late with you be­ing con­scious. Which is a re­duc­tio, be­cause the whole point with hav­ing be­liefs is to cor­re­late them with the truth.

• Lanrian 9 Dec 2019 22:14 UTC

  4 points

  in reply to: Benquo's comment on: The Les­son To Unlearn

  I’m not sure what you mean with “the spirit of pass­ing tests is hack­ing them”. Do you mean that the tests were in­ten­tion­ally de­signed to be hack­able? Be­cause it seems like Gra­ham is very much not say­ing that:

  Merely talk­ing ex­plic­itly about this phe­nomenon is likely to make things bet­ter, be­cause much of its power comes from the fact that we take it for granted. After you’ve no­ticed it, it seems the elephant in the room, but it’s a pretty well cam­ou­flaged elephant. The phe­nomenon is so old, and so per­va­sive. And it’s sim­ply the re­sult of ne­glect. No one meant things to be this way. This is just what hap­pens when you com­bine learn­ing with grades, com­pe­ti­tion, and the naive as­sump­tion of un­hack­a­bil­ity.

  I’m less con­fi­dent of what Ho­tel Concierge’s point is (partly be­cause it’s a damn long es­say with many dis­tinct points), but at least the end of it I’d sum­ma­rize as: “Suc­cess cor­re­lates with mis­ery, be­cause too much perfec­tion­ism con­tributes to both, and it’s a prob­lem when peo­ple are pushed and se­lected to­wards be­ing too perfec­tion­ist”. Some rele­vant pas­sages:

  I think the psy­chopathol­ogy term for TDTPT [this means The De­sire To Pass Tests] is “perfec­tion­ism.” [...]

  Perfec­tion­ism—like liter­ally ev­ery­thing else—is part of a spec­trum, good in mod­er­a­tion and dan­ger­ous in over­dose. [...]

  And so if you se­lect for high TDTPT, if you take only the high­est scores and most fev­er­ishly ded­i­cated hoop-jump­ing ap­pli­cants, then there is no way around it: you are se­lect­ing for a high frac­tion of un­happy peo­ple. [...]

  I’ve used Scantron-cen­tric ex­am­ples be­cause Scantrons are easy to quan­tify, but tests are ev­ery­where, and I promise you that the same trait that made me check my an­swers ten times over is pre­sent in the girl that spends two and a half hours do­ing her makeup, paus­ing ev­ery five min­utes to ask a room­mate if she looks ugly. TDTPT is the source of anorexia, body dys­mor­phia, worka­holism, anx­iety (“I just can’t find any­thing to say that doesn’t sound stupid”), ob­ses­sion, and a hun­dred mil­lion cases of an­he­do­nia, fa­tigue, and in­ad­e­quacy [...]

  So they’re say­ing that a lit­tle TDTPT is good for you, but many peo­ple have too much, and that’s bad for their men­tal health. Look­ing at that last para­graph, the ex­am­ples aren’t par­tic­u­larly con­nected to the hack­a­bil­ity of tests, as far as I can tell. That there are im­por­tant tests which are test­ing ar­bi­trary things cer­tainly con­tribute to the prob­lem, since be­ing perfec­tion­ist about mean­ingless work is less pro­duc­tive and more likely to lead to burnout, but it’s not es­sen­tial to the point, and it’s not some­thing that Ho­tel Concierge em­pha­sizes.

• Lanrian 9 Dec 2019 18:36 UTC

  8 points

  in reply to: Benquo's comment on: The Les­son To Unlearn

  I don’t think the es­says say the same thing. Paul Gra­ham claims that school in par­tic­u­lar en­courages hack­ing tests, and that the spirit of hack­ing tests is bad for many real-world prob­lems. Ho­tel Concierge equates the will to do well on tests with perfec­tion­ism, and thinks that this helps a lot with suc­cess in all parts of life, but causes mis­ery for the perfec­tion­ists. The dis­tinc­tion be­tween hack­able tests and non-hack­able tests is nei­ther em­pha­sized nor nec­es­sary for the lat­ter, while it’s cen­tral to the former.

• Lanrian 6 Dec 2019 15:43 UTC

  7 points

  in reply to: romeostevensit's comment on: Karate Kid and Real­is­tic Ex­pec­ta­tions for Disagree­ment Resolution

  Peo­ple have a re­ally hard time with in­ter­ven­tions of­ten be­cause they liter­ally do not have a func­tion­ing causal model of the thing in ques­tion. Peo­ple who ap­ply de­liber­ate prac­tice to a work­ing causal model of­ten level up as­ton­ish­ingly quickly.

  Could you give an ex­am­ple of a func­tion­ing causal model that quickly helps you learn a topic?

  I’m not sure whether you’re think­ing about some­thing more meta-level, “what can I prac­tice that will cause me to get bet­ter”, or some­thing more ob­ject-level, “how does me­chan­ics work”, and I think an ex­am­ple would help clar­ify. If it’s the lat­ter, I’m cu­ri­ous about what the differ­ence is be­tween hav­ing a func­tion­ing causal model of the sub­ject (the pre­con­di­tion for learn­ing) and be­ing good at the sub­ject (the re­sult of learn­ing).

• Lanrian 27 Nov 2019 21:25 UTC

  2 points

  in reply to: Viliam's comment on: Effect of Advertising

  Maybe I am too nega­tive about ad­ver­tis­ing, but it seems like its ma­jor strat­egy is to an­noy me. Like that ad­ver­tise­ment I won’t men­tion that I have re­cently seen (the first five sec­onds of) per­haps sev­eral hun­dred times, be­cause YouTube plays it at the be­gin­ning of al­most ev­ery video I see.

  FYI, ad­block­ers (like ublock ori­gin) work fine to pre­vent all of youtube’s ads, in­clud­ing the video ones.

• Lanrian 25 Nov 2019 7:59 UTC

  1 point

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

  That does seem to change things… Although I’m con­fused about what sim­plic­ity is sup­posed to re­fer to, now.

  In a pure bayesian ver­sion of this setup, I think you’d want some sim­plic­ity prior over the wor­lds, and then dis­card in­con­sis­tent wor­lds and renor­mal­ize ev­ery time you en­counter new data. But you’re not speak­ing about sim­plic­ity of wor­lds, you’re speak­ing about sim­plic­ity of propo­si­tions, right?

  Since a propo­si­tions is just a set of wor­lds, I guess you’re speak­ing about the com­bined sim­plic­ity of all the wor­lds. And it makes sense that that would in­crease if the propo­si­tion is con­sis­tent with more wor­lds, since any of the wor­lds would in­deed lead to the propo­si­tion be­ing true.

  So now I’m at “The sim­plic­ity of a propo­si­tion is pro­por­tional to the prior-weighted num­ber of wor­lds that it’s con­sis­tent with”. That’s start­ing to sound closer, but you seem to be say­ing that “The sim­plic­ity of a propo­si­tion is pro­por­tional to the num­ber of other propo­si­tions that it’s con­sis­tent with”? I don’t un­der­stand that yet.

  (Also, in my for­mu­la­tion we need some other kind of sim­plic­ity for the sim­plic­ity prior.)

• Lanrian 24 Nov 2019 19:39 UTC

  1 point

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

  Roughly, A is sim­pler than B if all data that is con­sis­tent with A is a sub­set of all data that is con­sis­tent with B.

  Maybe the less rough ver­sion is bet­ter, but this seems like a re­ally bad for­mu­la­tion. Con­sider (a) an ex­act enu­mer­a­tion of ev­ery event that ever hap­pened, mak­ing no pre­dic­tion of the fu­ture, vs (b) the true laws of physics and the true ini­tial con­di­tions, cor­rectly pre­dict­ing ev­ery event that ever hap­pened and ev­ery event that will hap­pen.

  In­tu­itively, (b) is sim­pler to spec­ify, and we definitely want to as­sign (b) a higher prior prob­a­bil­ity. But ac­cord­ing to this for­mu­la­tion, (a) is sim­pler, since all fu­ture events are con­sis­tent with (a), while al­most none are con­sis­tent with (b). Since both the­o­ries have equally much ev­i­dence, we’d be forced to as­sign higher prob­a­bil­ity to (a).

• Lanrian 24 Nov 2019 19:28 UTC

  14 points

  on: RAISE post-mortem

  Good post!

  Maybe this is too nit­picky, but “the most im­pact­ful years of your life will be 100x more im­pact­ful than the av­er­age” is nec­es­sar­ily false, be­cause your ca­reer is so short that those years will in­crease the av­er­age. For ex­am­ple, if you have a 50-year ca­reer and all of your im­pact hap­pens dur­ing a pe­riod of two years, your av­er­age yearly im­pact is 2/​50=1/​25 times as high as your im­pact dur­ing those two years. How­ever, “the most im­pact­ful years of your life will be 100x more im­pact­ful than the me­dian” could be true.

• Lanrian 7 Nov 2019 20:33 UTC

  8 points

  in reply to: johnswentworth's comment on: Build­ing In­tu­itions On Non-Em­piri­cal Ar­gu­ments In Science

  I feel like this is mostly a ques­tion of what you mean with “at­lantis”.

  • If you want to calcu­late P(ev­i­dence | the_spe­cific_at­lantis_that_newa­gers_speci­fied_af­ter_hear­ing_the_ev­i­dence) * P(the_spe­cific_at­lantis_that_newa­gers_speci­fied_af­ter_hear­ing_the_ev­i­dence), then the first term is go­ing to be pretty high, and the sec­ond term would be very low (be­cause it speci­fies a lot of things about what the atlean­tans did).

  • But if you want to calcu­late P(ev­i­dence | the_type_of_at­lantis_that_peo­ple_mostly_as­so­ci­ate_to_be­fore_think­ing_about_the_sphinx) * P(the_type_of_at­lantis_that_peo­ple_mostly_as­so­ci­ate_to_be­fore_think­ing_about_the_sphinx), the first term would be very low, while the sec­ond term would be some­what higher.

  The differ­ence be­tween the two cases is whether you think about the new agers as hold­ing ex­actly one hy­poth­e­sis and ly­ing about what it pre­dicts (as it can­not as­sign high prob­a­bil­ity to all of the things, since you’re cor­rect that the differ­ent prob­a­bil­ities must sum to 1), or whether you think about the new agers as switch­ing to a new hy­poth­e­sis ev­ery time they dis­cover a new fact about the sphinx /​ ev­ery time they’re asked a new ques­tion.

  In this par­tic­u­lar ar­ti­cle, Scott mostly wants to make a point about cases where the­o­ries have similar P(E|T) but differ in the prior prob­a­bil­ities, so he fo­cused on the first case.

• Lanrian 7 Nov 2019 19:54 UTC

  4 points

  in reply to: Dagon's comment on: Nor­ma­tive reductionism

  It’s harm­less (but silly (note: I LIKE silly, in part be­cause it’s silly to do so)) to have such prefer­ences, and usu­ally harm­less to act on them.

  I don’t re­ally un­der­stand why prefer­ences about things that you can’t ob­serve are more silly than other prefer­ences, but that’s ok. I mostly wanted to clear up the ter­minol­ogy, and note that it seems more like com­mon us­age of ‘prefer­ence’ and ‘util­ity’ to say “That’s a silly prefer­ence to have, be­cause X, Y, Z” and “I think we should only care about things that can af­fect us” in­stead of say­ing “Your satis­fac­tion of that prefer­ence has noth­ing to do with their con­fi­dence, it’s all about whether you ac­tu­ally find out” and “Without some per­cep­ti­ble differ­ence, your util­ity can­not be differ­ent”.

• Lanrian 7 Nov 2019 17:52 UTC

  −2 points

  in reply to: Dagon's comment on: Nor­ma­tive reductionism

  Does it help if you don’t think about a ‘prefer­ence’ as some­thing on­tolog­i­cally fun­da­men­tal, but just as a con­ve­nient short­hand for some­thing that an agent is op­ti­mis­ing for? It’s cer­tainly pos­si­ble for an agent to op­ti­mise for some­thing even if they’ll never re­ceive any ev­i­dence of if they suc­ceeded. gjm gives a few ex­am­ples in the sibling-com­ment to mine.

• Lanrian 6 Nov 2019 20:00 UTC

  1 point

  in reply to: Lanrian's comment on: Nor­ma­tive reductionism

  I’ve per­son­ally used to­tal con­se­quen­tial­ism for this in the past (when ar­gu­ing that non-causal de­ci­sion the­o­ries + to­tal con­se­quen­tial­ism im­plies that we should as­sume that alien civil­i­sa­tions are com­mon) and would sup­port it be­ing stan­dard ter­minol­ogy. Many peo­ple know what to­tal util­i­tar­i­anism is, and mak­ing the switch for con­se­quen­tial­ism is quite in­tu­itive.

• Lanrian 6 Nov 2019 19:30 UTC

  1 point

  in reply to: Dagon's comment on: Nor­ma­tive reductionism

  Without some per­cep­ti­ble differ­ence, your util­ity can­not be differ­ent.

  This defi­ni­tion of “util­ity” (and your defi­ni­tion of “prefer­ence”) is differ­ent from the one that most LWers use, differ­ent from the one that economists use, and differ­ent from the one that (at least some) pro­fes­sional philoso­phers use.

  Eco­momists use it to define any prefer­ence or­der­ing over wor­lds, and don’t re­quire it to be defined only over your own ex­pe­riences. Some eth­i­cal the­o­ries in philos­o­phy (e.g. he­do­nis­tic util­i­tar­i­anism) define it as a di­rect func­tion of your ex­pe­riences, but oth­ers, (e.g. prefer­ence util­i­tar­i­anism) define it as some­thing that can be af­fected by things you don’t know about. As ev­i­dence for the lat­ter, this SEP page states:

  If a per­son de­sires or prefers to have true friends and true ac­com­plish­ments and not to be de­luded, then hook­ing this per­son up to the ex­pe­rience ma­chine need not max­i­mize de­sire satis­fac­tion. Utili­tar­i­ans who adopt this the­ory of value can then claim that an agent morally ought to do an act if and only if that act max­i­mizes de­sire satis­fac­tion or prefer­ence fulfill­ment (that is, the de­gree to which the act achieves what­ever is de­sired or preferred). What max­i­mizes de­sire satis­fac­tion or prefer­ence fulfill­ment need not max­i­mize sen­sa­tions of plea­sure when what is de­sired or preferred is not a sen­sa­tion of plea­sure. This po­si­tion is usu­ally de­scribed as prefer­ence util­i­tar­i­anism.

  If you’re a he­do­nis­tic util­i­tar­ian, feel free to ar­gue for he­do­nis­tic util­i­tar­i­anism, but do that di­rectly in­stead of mak­ing claims about what other peo­ple are or aren’t al­lowed to have prefer­ences about.

• Lanrian 6 Nov 2019 0:10 UTC

  8 points

  on: Nor­ma­tive reductionism

  This SEP page defines:

  Ag­grega­tive Con­se­quen­tial­ism = which con­se­quences are best is some func­tion of the val­ues of parts of those con­se­quences (as op­posed to rank­ings of whole wor­lds or sets of con­se­quences).

  To­tal Con­se­quen­tial­ism = moral right­ness de­pends only on the to­tal net good in the con­se­quences (as op­posed to the av­er­age net good per per­son).

  as two re­lated and po­ten­tially iden­ti­cal con­cepts. When defin­ing nor­ma­tive re­duc­tion­ism, do you mean that the value of the world is equal to the sum of the value of its parts? If so, to­tal con­se­quen­tial­ism is prob­a­bly the clos­est term (though it’s a bit un­for­tu­nate that they only con­trast it with av­er­age util­i­tar­i­anism).