GuySrinivasan

• GuySrinivasan 20 May 2020 18:01 UTC

  4 points

  on: Race for the Galaxy: 9 Choose 6

  I like the idea of 9 choose 6, and might see my way to be­ing on board with “No IV turn 1” with the var­i­ant. Without the var­i­ant, of course, trade must re­main, or else wind­fall starts be­come quite bad. The ex­tra free­dom to play Set­tle on turn 1 with­out wor­ries seems pretty bal­anced with not hav­ing a start­ing world in the first place.

  As to this par­tic­u­lar draw, I’ll as­sume 4p. Smug­gling Lair is very very good here. It’s free which lets you keep the more-syn­er­gis­tic-than-usual cards you’re go­ing to start with, comes with a good, lets you leech off of Con­sume/​Trades very well, and gives you an ex­tra even­tual point. Re­search Labs is not very good if you and your op­po­nents are good; it’s clunky, usu­ally steals your green goods for points when you would rather cards, has a niche Pro­duce benefit that is less likely to come up when you’ve got green syn­er­gies go­ing, and has a strong Ex­plore benefit but good play­ers will be avoid­ing the Ex­plore ac­tion as much as pos­si­ble. I would definitely keep Smug­gling Lair, dis­card Re­search Labs, and play Develop, with an eye to­ward your next 3 turns in­volv­ing leech­ing off oth­ers’ IV while get­ting down 2-3 of Ge­net­ics Labs, Rebel War­rior Race, and An­cient Race, then de­cid­ing what to do from there. Turn two de­pends on oth­ers’ turn 1 plays, but the most likely line is Develop for Ge­net­ics Lab to put pres­sure on oth­ers’ abil­ity to use­fully play de­vel­op­ments or maybe Set­tle if 1+ folks played Bank/​Cred­its/​Works and there wasn’t a Set­tle turn 1, to put pres­sure on oth­ers’ econ­omy in gen­eral and en­sure your card­flow be­gins, which again would be safe with Smug­gling Lair as your Set­tle be­cause if some­one fi­nesses you with IV you still get the 2 cards.

• GuySrinivasan 18 May 2020 12:56 UTC

  5 points

  in reply to: blueanchovy's comment on: What are your great­est one-shot life im­prove­ments?

  Ah, right. Twix. I for­got about this one. While de­pressed, I grad­u­ally ramped up my con­sump­tion of Twix from the vend­ing ma­chine near my office to over 1 per day (2 since they came in pairs?). I wanted to stop, tried sev­eral times and failed, re­mem­bered about a tech­nique I’d seen on LW. Tried it. Have not had a sin­gle Twix for many years.

  Tech­nique: bought one Twix. Slowly, mind­fully, bro­ken off pieces and threw them in the trash, care­fully imag­in­ing that I was tast­ing the Twix and that it was full of mag­gots. Took… a few min­utes? Crav­ing gone.

• GuySrinivasan 17 May 2020 23:53 UTC

  2 points

  in reply to: ChristianKl's comment on: What are your great­est one-shot life im­prove­ments?

  Not sure. It was a hand­made afghan which, when both it and my body were fully flat and at a nor­mal an­gle (not di­ag­o­nal), did not quite reach from above my head to be­low my feet. I am 5′10″ and had a habit of cov­er­ing my head with the blan­ket at night. (I re­tained that habit un­til half a year ago when I got a blind­fold eye cov­er­ing thinger that was ac­tu­ally com­fortable.)

  Now I use the enor­mous Costco blan­ket that’s su­per soft and I highly recom­mend for your­self and for giv­ing as gifts, ex­cept when it’s too hot and I use a blan­ket which cov­ers the en­tirety of a queen-sized bed.

• GuySrinivasan 17 May 2020 2:02 UTC

  3 points

  in reply to: Pongo's comment on: What are your great­est one-shot life im­prove­ments?

  Stum­bled, then con­firmed. Cir­cum­stances were (a) my part­ner no­tified me that hav­ing all that back pain was not in fact nor­mal, and (b) of­ten sleep­ing un­der a sane blan­ket be­cause my fa­vored blan­ket was ob­vi­ously in­ad­e­quate for the both of us. Even­tu­ally con­nected the dots enough to form the hy­poth­e­sis, then did a trial to con­firm be­cause I didn’t want to give up my be­loved blan­ket.

• GuySrinivasan 16 May 2020 22:23 UTC

  8 points

  on: What are your great­est one-shot life im­prove­ments?

  Prob­lem: I had a lot of back pain. Some days I’d wake up in pain, and it wouldn’t go away for 1-4 days af­ter­wards. Al­most once per week? No rhyme or rea­son.

  Solu­tion: Re­place the blan­ket I was us­ing to sleep. My blan­ket was too small, I would curl up too tightly to stay un­der it. Now I still very oc­ca­sion­ally get the same prob­lem, I’m guess­ing be­cause of habits, but it’s more like once ev­ery three months.

• GuySrinivasan 7 May 2020 13:38 UTC

  4 points

  in reply to: Max Hodges's comment on: A non-mys­ti­cal ex­pla­na­tion of in­sight med­i­ta­tion and the three char­ac­ter­is­tics of ex­is­tence: in­tro­duc­tion and preamble

  Some per­sonal ex­pe­rience. On the or­der of… ~100 hrs, I think?

  I am not say­ing it’s dishon­est to say it has worth­while benefits. My guess is that if some­one hears about this en­light­en­ment thing and thinks hmmm I won­der if I can...? and then tries in­sight med­i­ta­tion for 30x 10 minute ses­sions, the modal re­sult is prob­a­bly (a) some worth­while benefits, and (b) be­liev­ing that the scope of po­ten­tial worth­while benefits is far far larger than what 30x 10 minute ses­sions re­sulted in. Or, rephrased, that 30x 10 minute ses­sions just scratched the sur­face. Or, rephrased, that 30x 10 minute ses­sions was not “deep”.

  If you want to throw reg­u­lar 12-per­son themed din­ner par­ties ev­ery week­end, cook­ing ev­ery day for 10 min­utes is a fan­tas­tic place to start learn­ing the skills you need, and some­thing like that is ab­solutely nec­es­sary for long-term suc­cess if you’ve never cooked, and will definitely have worth­while benefits, and ab­solutely is go­ing to miss cov­er­ing tons and tons of nec­es­sary skills. If you told some­one your goal and they said “it’s ac­tu­ally easy, just cook for 10 min­utes a day for 30 days”, they would be tel­ling you a false thing. Their ad­vice, if fol­lowed, might be the best way to start, but that’s not the same thing as true.

• GuySrinivasan 7 May 2020 1:10 UTC

  3 points

  in reply to: Max Hodges's comment on: A non-mys­ti­cal ex­pla­na­tion of in­sight med­i­ta­tion and the three char­ac­ter­is­tics of ex­is­tence: in­tro­duc­tion and preamble

  I mean, sure, it’s a time hon­ored prac­tice to lie to peo­ple and tell them they can make a sub­stan­tial frac­tion of the to­tal progress (“deep”) with very lit­tle effort, in­tend­ing to get them to start when oth­er­wise they sim­ply wouldn’t, so that they’ll find out that it takes very lit­tle effort to make a lot of progress even though they’ll also find out that the to­tal is more vast than they ever imag­ined and their lit­tle amounts of effort didn’t yield a sub­stan­tial frac­tion of the to­tal progress.

  lie

  I still hate it.

• GuySrinivasan 6 May 2020 17:02 UTC

  2 points

  in reply to: Max Hodges's comment on: A non-mys­ti­cal ex­pla­na­tion of in­sight med­i­ta­tion and the three char­ac­ter­is­tics of ex­is­tence: in­tro­duc­tion and preamble

  Every recipe I look up on the in­ter­net says it takes just 5 min­utes.

  Ideal prac­tice might go quite far with 30x 10 minute prac­tices, enough to call it “deep”. Likely prac­tice if you don’t spend any time with a men­tor, read­ing/​watch­ing on­line, fix­ing your sit setup, etc etc etc? Sincerely doubt it would end up enough to call what­ever hap­pened “get­ting deep in med­i­ta­tion”. Time ac­tu­ally spent if you have 30x 10 minute prac­tices, given that you’ll be do­ing the above and also tran­si­tion­ing be­tween other things to prac­tice and back, ex­plain­ing to your part­ner that you need to not be dis­turbed dur­ing those 10 min­utes and an­swer­ing their ques­tions about med­i­ta­tion, etc etc etc? Way more than 30x10 min­utes.

• GuySrinivasan 3 May 2020 3:12 UTC

  5 points

  in reply to: ErickBall's comment on: Against strong bayesianism

  A piece of a cer­tain large cor­po­ra­tion’s spel­ling/​gram­mar checker was at its heart Re­sult <== Prior x Ev­i­dence. Due to legacy code, de­cay­ing in­sti­tu­tional knowl­edge, etc., no one knew this. The code/​math was strewn about many files. Folks had tweaked the code over the years, al­lowed pa­ram­e­ters to vary, fit those pa­ram­e­ters to data.

  I read the code, re­al­ized that fun­da­men­tally it had to be “about” de­ter­min­ing a prior, de­ter­min­ing ev­i­dence, and com­put­ing a pos­te­rior, re­con­structed the ac­tual math be­ing performed, and dis­cov­ered that two ex­po­nents were differ­ent from each other, “best fit to the data”, and I couldn’t think of any good rea­son they should be differ­ent. Brain threw up all sorts of warn­ing bells.

  I ex­am­ined how we trained on the data to de­ter­mine the val­ues we’d use for these ex­po­nents. Turns out, the pro­cess was com­pletely un­jus­tifi­able, and only seemed to give bet­ter re­sults be­cause our test set was sub­tly part of the train­ing set. Now that’s some­thing ev­ery­one un­der­stands im­me­di­ately; you don’t train on your test set. So we fixed our eval­u­a­tion pro­cess, stopped al­low­ing those par­tic­u­lar pa­ram­e­ters to float, and im­proved over­all perfor­mance quite a bit.

  Note, math. Be­cause in­for­ma­tion and Bayes and cor­re­la­tion and such is un­for­tu­nately not sim­ple, it’s en­tirely pos­si­ble that some type of data is bet­ter served by e^(a*ln(P(v|w))-b*ln(P(v|~w))) where a!=b!=1. I dunno. But if you see some­one ob­vi­ously only in­tro­duc­ing a and b and then fit­ting them to data be­cause :shrug: why not, that’s when your red flags go up and you re­al­ize some­one’s put this thing to­gether with­out pay­ing at­ten­tion. And in this case af­ter fix­ing the eval­u­a­tion we did end up leav­ing a==b != 1, which is just Naive Bayes, ba­si­cally. a!=b was the re­ally dis­con­cert­ing bit.

• GuySrinivasan 30 Apr 2020 19:53 UTC

  17 points

  on: Against strong bayesianism

  Towards what­ever-you-call-the-thing-I-got-from-read­ing-LW ism:

  Think­ing about all in­for­ma­tion sys­tems as fun­da­men­tally perform­ing the com­pu­ta­tion “Re­sult <== Prior x Ev­i­dence” has been re­spon­si­ble for 5 of the 7 biggest suc­cesses in my ca­reer thus far. The other 2 had noth­ing to do with math/​in­for­ma­tion/​prob­a­bil­ity. All of the 5 were me notic­ing, where many bet­ter ed­u­cated “more qual­ified” in­di­vi­d­u­als did not, that some part of the ac­tual in­for­ma­tion sys­tem’s im­ple­men­ta­tion was bro­ken w.r.t. “Re­sult <== Prior x Ev­i­dence” and figur­ing out how to phrase the bro­ken­ness in some other way (’cause in­fer­en­tial dis­tance), re­sult­ing in in­sti­tu­tional pres­sure to fix it.

• GuySrinivasan 17 Apr 2020 4:36 UTC

  1 point

  on: Two-player Catan

  Open hands make trad­ing al­most ir­rele­vant. Closed hands means trades could still hap­pen but prob­a­bly shouldn’t of­ten, so maybe it’s not worth the com­plex­ity to al­low trades. For me though al­most all the fun of Set­tlers is in trad­ing—try­ing to be­come one of the part­ners in as many of the in­evitable trades as pos­si­ble, even if the other player gets a lit­tle more than you ev­ery time, be­cause you’re still gain­ing on the non-trad­ing play­ers. The back­drop is the bits of tac­tics and strat­egy on the board, but other games are far richer in that de­part­ment.

  So how could we re­tain in­ter­est­ing trad­ing dy­nam­ics? I’m think­ing some­how add fake play­ers, sup­ply, de­mand to the mix. How about this (com­pletely untested):

  After the board is laid out, be­fore de­ter­min­ing first player, jointly agree on a rank­ing of re­sources from 5 to 1, most im­por­tant to least, prob­a­bly similar to “least pips available to most pips available”. For ex­am­ple, maybe it’s wheat > ore > brick > sheep > wood. I’ll use that rank­ing for illus­tra­tion.

  Create a de­mand deck of 2x wheat, 2x ore, 1x brick, 1x sheep, face­down, shuffled, with a cor­re­spond­ing empty dis­card deck. Similarly cre­ate sup­ply: 2x wood, 2x sheep, 1x brick, 1x ore. Deal 1 de­mand and 1 sup­ply faceup. This is an available trade. Between turns any­one may dis­card cards match­ing the shown de­mand from their hand to gain cards match­ing the shown sup­ply, once.

  De­tails: Player who just took a turn has pri­or­ity. If one of you is about to trade, the other may out­bid them, re­peat un­til a high­est bid is reached, then that trade goes through; dis­card all the de­mand and sup­ply cards to their re­spec­tive dis­card piles. If no one makes a trade, deal an ad­di­tional de­mand and two ad­di­tional sup­ply; the available trades are “the first N de­mand for the first 2N-1 sup­ply”, higher N wins. If a turn ends and there is no available trade, deal one. When­ever an available trade has wlog a sheep in de­mand and in sup­ply, dis­card that match. If ei­ther the available sup­ply or de­mand has no show­ing cards, dis­card all show­ing cards. When­ever re­sources (non-7) are rol­led, add a re­source of each type among the most dis­tributed to the sup­ply dis­card pile and a re­source of each type among the least dis­tributed to the de­mand dis­card pile. When­ever a sup­ply or de­mand pile is empty and you need to draw, shuffle the dis­card, deal 6 face­down to form the new deck, and re­turn the oth­ers to the bank. If there still are not enough cards, dis­card the available trade. When you play a Monopoly you also gain any match­ing cards in show­ing sup­ply. You may not steal from the Shadow Trader.

  Out­bid­ding: only a su­per­set of the show­ing de­mand is a valid bid. Higher N wins. Values are 4, 3, 3, 3, 2 in or­der of re­source rank; high­est sum of val­ues wins. If an ex­tra sup­ply card ex­ists, you may de­cline it and add its value to your to­tal.

  Ex­am­ple of play: there’s hardly any wheat or ore, lots of sheep and wood, so above rank­ing is used. Show­ing trade is de­mand: wheat, sup­ply: wood. Sur­pris­ingly, lots of wheat has been placed on and rol­led. Alice offers a wheat. Bob offers a wheat and a sheep. Alice folds, Bob trades. Bob takes his turn; 2 wood and 1 brick are dis­tributed, so 1 wood is added to sup­ply dis­card and 1 wheat, ore, and sheep to de­mand dis­card. A trade is dealt, de­mand: sheep, sup­ply: ore. Both play­ers are cha­grined to not have sheep, and do not trade. Alice takes her turn, rol­ling a 7. Ad­di­tional trade cards are dealt, de­mand: sheep /​ wheat, sup­ply: ore /​ sheep, wood. Dis­card the sheep. De­mand: wheat, sup­ply: ore /​ wood. Alice offers a wheat. Bob offers a wheat, de­clin­ing the wood. Alice offers two wheat. Bob folds, Alice trades.

• GuySrinivasan 10 Apr 2020 21:34 UTC

  21 points

  on: How to eval­u­ate (50%) predictions

  I think you’re con­flat­ing im­pres­sive­ness of pre­dic­tions with cal­ibra­tion of pre­dic­tions. It’s pos­si­ble to be perfectly cal­ibrated and unim­pres­sive (for ev­ery state­ment, guess 50% and then ran­domly choose whether to use the state­ment or its nega­tion). It’s also pos­si­ble to be un­cal­ibrated and very im­pres­sive (be re­ally good at find­ing all the ev­i­dence sway­ing things from baseline, but count all ev­i­dence 4x, for in­stance).

  50% pre­dic­tions don’t re­ally tell us about cal­ibra­tion due to the “swap state­ment with nega­tion at ran­dom” strat­egy, but they can tell us plenty about im­pres­sive­ness.

• GuySrinivasan 8 Apr 2020 13:57 UTC

  6 points

  on: How do you de­ter­mine tem­per­a­ture cut­offs for women’s body-tem­per­a­ture?

  Does she know her nor­mal range? My wife’s tem­per­a­ture fluc­tu­ates a full just-over-two-de­grees, for a differ­ent rea­son, so we know when it’s past “nor­mal”. We’ve sim­ply re­signed our­selves to los­ing “slight fever” as a use­ful in­di­ca­tor of any­thing at all. Tech­ni­cally there’s slightly more in­for­ma­tion than that, but in prac­tice fal­libly try­ing to use it would pro­duce more false pos­i­tives than would be worth it.

• GuySrinivasan 17 Mar 2020 17:48 UTC

  2 points

  in reply to: Pattern's comment on: How effec­tive are tul­pas?

  Sure. Seems ex­tremely un­likely IMO. But if you’re de­liber­ately try­ing to change how your brain thinks at a fun­da­men­tal level rather than train­ing an over­lay like we do when learn­ing math or some­thing and let­ting that trickle down or how­ever it usu­ally works, you might suc­ceed at chang­ing but fail at di­rec­tion. This is an ob­vi­ous failure mode to at least con­sider be­fore be­gin­ning. e.g. http://​​med­i­tat­in­gin­safety.org.uk/​​wp-con­tent/​​up­loads/​​2017/​​05/​​Kuijpers_2007.pdf

• GuySrinivasan 10 Mar 2020 17:53 UTC

  19 points

  in reply to: Raven's comment on: How effec­tive are tul­pas?

  If you had not even con­sid­ered the pos­si­bil­ity of break­ing your brain in the pro­cess of try­ing to de­velop a sec­ond per­son, you need to step back and think more be­fore pro­ceed­ing. This failure mode should be one of the first that pops into your head, with­out even try­ing to think of novel failure modes. Right alongside in­tense med­i­ta­tion, psychedelics, etc.

• GuySrinivasan 4 Mar 2020 16:42 UTC

  2 points

  on: Coron­avirus: Jus­tified Prac­ti­cal Ad­vice Thread

  If it’s more im­por­tant for you to avoid con­tract­ing the virus right now than it is for oth­ers, sched­ule any doc­tor’s ap­point­ments you are go­ing to for first thing in the morn­ing. Rea­son­ing: seems like hos­pi­tal con­tam­i­na­tion should rise through­out the day as more and more peo­ple pass through and then drop dur­ing nightly clean­ings. Similarly for any needed out­ings, do it in the morn­ing, be­cause though the virus can sur­vive for days on sur­faces, pre­sum­ably it’s safer to touch just af­ter an 8 hour break from peo­ple than just af­ter a 2 minute break from peo­ple. No idea of effect size but I don’t see why these mechanisms would be wrong.

• GuySrinivasan 31 Jan 2020 15:56 UTC

  5 points

  on: [ELDR Tac­tics] Con­sider switch­ing to (mostly) de­caf.

  I used to overuse caf­feine. Every few months I would go cold turkey for a week or two, to (a) prove to my­self I eas­ily could, but also (b) mostly re­set the amount I needed or thought I needed for the same benefits. I figured this worked all right.

  Then a change in my home life made it nec­es­sary to some­times be able to stay or quickly be­come alert and use­ful at odd hours of the night/​morn­ing, and a mod­er­ate amount of caf­feine just wasn’t enough to break through. So my part­ner and I in­sti­tuted a new rule: (ba­si­cally) no caf­feine af­ter noon, and at most one cup of coffee’s worth be­fore that; but all the de­caf I wanted.

  Benefits: rad­i­cally bet­ter sleep qual­ity, one dose dur­ing night­time emer­gen­cies is suffi­cient, and I no longer par­ti­ci­pate in the tol­er­ance/​turkey cy­cle.

  Cou­ple of other ELDRs:

  • we only buy a very spe­cific brand of eggs, vet­ted to be the most eth­i­cally pro­duced available locally

  • I play as many seden­tary games as I like, but in­cur a cer­tain amount of ex­er­cise debt per [ses­sion]

• GuySrinivasan 5 Nov 2019 17:24 UTC

  6 points

  in reply to: ryan_b's comment on: To­tal horse takeover

  https://​​en.wikipe­dia.org/​​wiki/​​Rat

  https://​​wiki.less­wrong.com/​​wiki/​​Co­her­ent_Ex­trap­o­lated_Volition

• GuySrinivasan 2 Jun 2019 17:51 UTC

  12 points

  on: “But It Doesn’t Mat­ter”

  On the con­trary, one of my go-to tech­niques for de­ci­sion-mak­ing is to ex­am­ine ques­tions that seem rele­vant to see if I can tell by the mag­ni­tude of their pos­si­ble an­swers whether I care about what their pos­si­ble an­swers are. If my choice boils down to “if X > 100, yes, oth­er­wise no” and I am pretty con­fi­dent that X is some­where around 90-110 and I find a ques­tion that seems rele­vant but it turns out any an­swer sways X by at most a tenth of a point, I will dis­miss that ques­tion and look for more im­por­tant ques­tions.

  It is a flag to check for ra­tio­nal­iza­tion, sure.

• GuySrinivasan 1 May 2019 17:21 UTC

  16 points

  in reply to: quanticle's comment on: Nash equil­ibriums can be ar­bi­trar­ily bad

  If your strat­egy re­mains un­changed, I can change my strat­egy to “999,999.99 please” and come away with 1,000,000.01 in to­tal, so that’s not a Nash equil­ibrium.