Unnamed

• Unnamed 21 May 2020 22:08 UTC

  23 points

  on: What are ob­jects that have made your life bet­ter?

  Some lists that peo­ple have made of prod­ucts that they use & recom­mend:

  Sam Bow­man, 2017

  Sam Bow­man, 2019

  Robert Wiblin, 2019

  Ar­den Koehler, 2019

  Rosie Camp­bell, 2019

• Unnamed 21 May 2020 21:46 UTC

  3 points

  on: What are ob­jects that have made your life bet­ter?

  The Time Timer Audible Count­down Timer.

  This is the timer that I like to use when work­ing, e.g. if I de­cide “alright, I’m go­ing to spend the next half hour work­ing on this thing.” It is a vi­sual timer, where the frac­tion of the cir­cle that is red tells you what frac­tion of an hour is left. Ig­nore its bizarre name—its best fea­ture is that it is com­pletely inaudible.
  
  Fea­tures that I like:
  - it counts down silently, with­out any tick­ing
  - I can (and do) set it to end silently, with­out any alarm sound
  - it is easy to tell at a glance about how much time is left
  - it is quick & straight­for­ward to set the timer, with­out any but­ton press­ing
  - it is a phys­i­cal ob­ject rather than a pro­gram on a com­put­ing de­vice
  
  Fea­tures that it lacks which some peo­ple might miss:
  - you can’t choose a nice sound for the alarm, ei­ther it’s silent or there’s the one kinda an­noy­ing alarm sound
  - it is not a pro­gram on your com­put­ing de­vice, but rather a sep­a­rate ob­ject you need to have with you
  - it can’t be set to more than an hour
  - it can’t be set precisely

• Unnamed 3 May 2020 5:13 UTC

  12 points

  on: Book Re­view: Narconomics

  The eco­nomic ar­gu­ment seems wrong in the “Burn­ing coca leaves won’t win the war” sec­tion.

  The to­tal amount of a good that con­sumers buy must be less than or equal to the amount that is pro­duced (and not de­stroyed). So if enough of the crop gets de­stroyed, then less of it will get con­sumed. And that’ll hap­pen re­gard­less of whether the sup­pli­ers are in a com­pet­i­tive mar­ket or monop­sony or threaten peo­ple with guns.

  I framed this in terms of quan­tities rather than prices be­cause the ar­gu­ment seems more straight­for­ward this way. Also, it seems like re­duc­ing the quan­tity sold is more di­rectly re­lated to what anti-drug folks care about than rais­ing the price. Also, the street price for US con­sumers would pre­sum­ably go up if the availa­bil­ity went down, since the peo­ple who sell drugs to con­sumers would be able to make more profit by rais­ing their prices.

  If there are prob­lems with the eco­nomic ar­gu­ment in the post, that doesn’t nec­es­sar­ily mean the con­clu­sion is wrong. “Burn­ing lots of coca crops will have lit­tle to no effect on the price or quan­tity of co­caine in the US” does seem plau­si­ble, mainly be­cause pro­duc­ers can just grow a lot more coca leaves than they need. Pro­duc­ers can pre­dict in ad­vance that lots of their crop might get de­stroyed (or their product lost in tran­sit or similar), and grow­ing coca leaves is not that ex­pen­sive rel­a­tive to their op­er­a­tion, so they can add a lot of slack by grow­ing more than they need. (This doesn’t de­pend on monop­sony or vi­o­lence.)

• Unnamed 14 Apr 2020 9:08 UTC

  4 points

  on: The One Mis­take Rule

  One ob­vi­ously mis­taken model that I got a lot of use out of dur­ing a stretch of Feb-Mar is the one where the cu­mu­la­tive num­ber of coro­n­avirus in­fec­tions in a re­gion dou­bles ev­ery n days (for some globally fixed, un­known value of n).

  This model has ridicu­lous im­pli­ca­tions if you ex­tend it for­ward for a few months, as well as var­i­ous other flaws. I was aware of those ridicu­lous im­pli­ca­tions and some of those other flaws, and used it any­ways for sev­eral days be­fore try­ing to find less flawed mod­els.

  I’m glad that I did, since it helped me have a bet­ter grasp of the situ­a­tion and be more pre­pared for what was com­ing. And I don’t think it would’ve made much differ­ence at the time if I’d learned more about SEIR mod­els and so on.

  It’s un­clear how ex­am­ples like this are sup­posed to fit with the One Mis­take Rule or the ex­cep­tions in the last para­graph.

• Unnamed 10 Apr 2020 21:20 UTC

  12 points

  in reply to: orthonormal’s comment on: The One Mis­take Rule

  This seems im­por­tant.

  Another fea­ture of com­pet­i­tive mar­kets is that “not bet­ting” is always available as a safe de­fault op­tion. Maybe that means wait­ing to bet un­til some un­known fu­ture date when your mod­els are good enough, maybe it means never bet­ting in that mar­ket. In many other con­texts (like re­spond­ing to covid-19) there is no safe de­fault op­tion.

• Unnamed 4 Apr 2020 3:55 UTC

  5 points

  in reply to: habryka’s comment on: Has LessWrong been a good early alarm bell for the pan­demic?

  In the broader ra­tio­nal­ity/​EA com­mu­nity there was also a Siderea post on Jan 30 and an 80K pod­cast on Feb 3 (along with a fol­lowup pod­cast on Feb 14).

  Th­ese two, plus Matthew Bar­nett’s late Jan EA Fo­rum post (which you linked), are the three ex­am­ples I re­call which look most like early visi­ble pub­lic alarms from the ra­tio­nal­ity/​EA com­mu­nity.

  Other writ­ing was less visi­ble (e.g., on Twit­ter, Face­book, or Me­tac­u­lus), less alarm-like (dis­cus­sions of some as­pect of what was hap­pen­ing rather than a call to at­ten­tion), or later (like the putanu­monit See­ing the Smoke post on Feb 27).

• Unnamed 3 Apr 2020 22:34 UTC

  7 points

  on: Has LessWrong been a good early alarm bell for the pan­demic?

  I think this post is giv­ing the stock mar­ket too much credit.

  I’d date the start of the stock mar­ket fall as Fe­bru­ary 24 rather than Fe­bru­ary 20. The S&P close on Feb 20 & Feb 21 was roughly the same as it had been over the pre­vi­ous cou­ple weeks, and higher than the close on Feb 7, 5, 4, or 3. The first no­table dip hap­pened on Fe­bru­ary 24th; that was the first day that set a low for the month of Feb 2020 (and Feb 25 was the first day that set a low for cal­en­dar year 2020).

  Also, that was just the start of the crash. The stock mar­ket con­tinued fal­ling sharply and er­rat­i­cally for a cou­ple more weeks, and didn’t get within 10% of its cur­rent level un­til March 12th (2.5 weeks af­ter it started its fall on Feb 24).

• Unnamed 1 Apr 2020 8:50 UTC

  38 points

  on: April Fools: An­nounc­ing LessWrong 3.0 – Now in VR!

  This is now my fa­vorite way to read HPMOR. I love the Star Wars feel.

  

• Unnamed 29 Mar 2020 5:15 UTC

  2 points

  on: mind viruses about body viruses

  I think Scott linked to Pueyo’s es­say as an illus­tra­tion of the ideas, not as the source from which the smart peo­ple got the ideas.

  Which means that this post’s at­tempt to track & eval­u­ate the in­for­ma­tion flows is work­ing off of an in­ac­cu­rate map of how in­for­ma­tion has flowed.

• Unnamed 25 Mar 2020 22:34 UTC

  4 points

  in reply to: Chris Hibbert’s comment on: Coron­avirus Open Thread

  Keep in mind that the trend in the num­ber of con­firmed cases only pro­vides hints about the trend in new in­fec­tions. The num­ber of con­firmed cases is highly de­pen­dent on the amount of test­ing, and in­creases in test­ing ca­pac­ity will tend to lead to more con­firmed cases. Also, there is a sub­stan­tial de­lay be­tween when a per­son is in­fected and when they test pos­i­tive, typ­i­cally some­where in the range of 1-2 weeks (with the length of the de­lay also de­pend­ing on the test­ing regime).

• Unnamed 23 Mar 2020 19:59 UTC

  2 points

  in reply to: Elizabeth’s comment on: Us­ing smart ther­mome­ter data to es­ti­mate the num­ber of coro­n­avirus cases

  I think that’s right. Although the data still can tell us some­thing af­ter we get into that am­bigu­ous range where it’s hard to dis­t­in­guish in­creas­ing covid and de­creas­ing flu.

  One nice thing about this pat­tern is that it pro­vides some ev­i­dence that the anti-covid in­ter­ven­tions are re­duc­ing the spread of fever-in­duc­ing dis­eases. And the size of the drop in to­tal fev­ers tells us some­thing about how well they’re work­ing on the whole, even if it doesn’t tell us the pre­cise trend in covid cases.

  Another thing that might be pos­si­ble is to find other sources of data on the ac­tual prevalence of flu, and use that to come up with a bet­ter “baseline” which re­flects ac­tual cur­rent con­di­tions rather than an es­ti­mate of the trendline in the coun­ter­fac­tual world where there was no coro­n­avirus pan­demic.

  A third thing is that 0 is a lower bound on the num­ber of non-covid fev­ers, so the trend in to­tal fev­ers is an up­per bound on the num­ber of covid cases.

  This third thing already tells us some­thing about Seat­tle (King County). Their peak in ex­cess fev­ers hap­pened March 9 at 1.76 scale points (ob­served minus ex­pected), and the March 22 data show the to­tal fev­ers at 2.77 scale points. As an up­per bound, if those are all covid fev­ers, that is 1.6x as many new daily cases on March 22 com­pared to March 9. That’s 13 days, and not even a full dou­bling in the num­ber of daily new fev­ers. Which sug­gests that sup­pres­sion there is ei­ther work­ing or com­ing very close to work­ing (even though the num­ber of con­firmed cases has kept curv­ing up­ward, at least through March 21).

• Unnamed 23 Mar 2020 19:38 UTC

  2 points

  in reply to: Douglas_Knight’s comment on: Us­ing smart ther­mome­ter data to es­ti­mate the num­ber of coro­n­avirus cases

  If you look at the time se­ries for King County (Seat­tle area), it shows a spike peak­ing on March 9 with the up­ward trend be­gin­ning some­time around Feb 28 - Mar 2.

  I think the pat­tern of a spike and then flat­ten­ing & maybe de­cline (which has hap­pened at differ­ent times in differ­ent re­gions) re­flects a drop in the num­ber of in­fluenza cases, as peo­ple’s anti-covid pre­cau­tions also pre­vent flu trans­mis­sion. So the baseline es­ti­mate of how many new fev­ers there would be if there wasn’t a coro­n­avirus pan­demic doesn’t ac­tu­ally rep­re­sent the num­ber of non-covid fev­ers, be­cause there are fewer non-covid fev­ers than there would’ve been with­out this pan­demic.

  Eliz­a­beth’s com­ment also de­scribes this.

• Unnamed 23 Mar 2020 4:34 UTC

  12 points

  on: How can we es­ti­mate how many peo­ple are C19 in­fected in an area?

  Kinsa, a com­pany that sells smart ther­mome­ters, has a dash­board that shows which re­gions of the US have an un­usu­ally high num­ber of fev­ers. They have pre­vi­ously used these meth­ods to track re­gional flu trends in the US. (FitBit has done some­thing similar.)

  I wrote a post here de­scribing my at­tempt to turn their data into a rough es­ti­mate of the to­tal num­ber of coro­n­avirus in­fec­tions in the United States. Some­thing similar could be done for smaller re­gions.

  What links here?

  • LW Coron­avirus Agenda Up­date 3/​23 by Elizabeth (24 Mar 2020 1:05 UTC; 46 points)

Us­ing smart ther­mome­ter data to es­ti­mate the num­ber of coro­n­avirus cases

• Unnamed 23 Mar 2020 2:42 UTC

  11 points

  on: Us­ing the Quan­tified Self paradigma for COVID-19

  I agree that a lot could be done with those sorts of data.

  One com­pany that already is mak­ing some use of a similar dataset is Kinsa, who sells smart ther­mome­ters. They started a few years ago, track­ing trends in the flu in the US based on the tem­per­a­ture read­ings of the peo­ple us­ing their ther­mome­ters (along with lo­ca­tion, age, and gen­der). Now they have a coro­n­avirus track­ing web­site up. It looks like the biggest use­ful thing that they’ve been able to do so far with their data is to quickly iden­tify hotspots—parts of the coun­try where there has been a spike in the num­ber of peo­ple with a fever. That used to be a sign of a lo­cal flu out­break, now it’s a sign of a lo­cal coro­n­avirus out­break. From the NYTimes:

  Just last Satur­day, Kinsa’s data in­di­cated an un­usual rise in fev­ers in South Florida, even though it was not known to be a Covid-19 epi­cen­ter. Within days, test­ing showed that South Florida had in­deed be­come an epi­cen­ter.

  Com­pa­nies like Fit­bit could make a similar pivot, look­ing to see if they can find atyp­i­cal trends in their data in the Seat­tle area Feb 28 - Mar 9, the Mi­ami area Mar 2-19, etc. And they might be able to take the ex­tra step of iden­ti­fy­ing new in­di­ca­tors that help iden­tify in­di­vi­d­u­als who may have coro­n­avirus (un­like Kinsa, as high body tem­per­a­ture was already a known in­di­ca­tor).

  There are po­ten­tially a bunch more use­ful things that could be done with all of these datasets, if more re­searchers had ac­cess to them. For ex­am­ple, it might be pos­si­ble to get much more ac­cu­rate es­ti­mates of the num­ber of peo­ple who have been in­fected with coro­n­avirus. I may make an­other post about this soon.

• Unnamed 17 Mar 2020 0:51 UTC

  9 points

  on: COVID-19′s House­hold Se­condary At­tack Rate Is Unknown

  Has there been re­search from other similar­ish dis­eases break­ing down the house­hold sec­ondary at­tack rate by rele­vant vari­ables? It seems like there could be large differ­ences be­tween:

  ro­man­tic part­ners who sleep in the same bed vs. house­mates who sleep in differ­ent rooms

  cir­cum­stances where the house­hold has height­ened con­cerns and is tak­ing pre­cau­tions vs. un­sus­pect­ing households

  situ­a­tions where peo­ple are re­moved from the house­hold shortly af­ter they’re in­fected vs. house­holds where peo­ple con­tinue to live af­ter infection

  
  

  Group houses are mostly in the safer of the two pos­si­bil­ities for the first 2 of these 3.

• Unnamed 16 Mar 2020 5:13 UTC

  5 points

  in reply to: Unnamed’s comment on: A Sig­nifi­cant Por­tion of COVID-19 Trans­mis­sion Is Presymptomatic

  I was look­ing at this pa­per (for other rea­sons) and saw that it es­ti­mated a mean se­rial in­ter­val of 6.3 days in Shen­zhen while there was ag­gres­sive test­ing, con­tact trac­ing, and iso­lat­ing. They re­port that the mean se­rial in­ter­val was 3.6 days among pa­tients who were in­fected by some­one who was iso­lated within 2 days of symp­tom on­set, and 8.1 days among pa­tients who were in­fected by some­one who wasn’t iso­lated un­til 3+ days af­ter symp­tom on­set, for an over­all av­er­age se­rial in­ter­val of 6.3 in their pop­u­la­tion. They found R=0.4 - an av­er­age of 0.4 known trans­mis­sions from each in­fected per­son.

• Unnamed 16 Mar 2020 2:29 UTC

  2 points

  in reply to: Unnamed’s comment on: Coron­avirus Open Thread

  This pa­per looks at cases which were con­firmed in Shen­zhen (Guang­dong, China) Jan 14 - Feb 12, which is while coro­n­avirus was be­ing brought un­der con­trol there (by the end of the study the cases had fallen to less than ¹⁄₃ of their peak). I sus­pect that they qual­ify for point 1, a place with an un­usu­ally good test­ing regime.

  The pa­per re­ports that “Cases de­tected through symp­tom-based surveillance were con­firmed on av­er­age 5.5 days (95% CI 5.0, 5.9) af­ter symp­tom on­set (Figure 3, Table S2); com­pared to 3.2 days (95% CI 2.6,3.7) in those de­tected by con­tact-based surveillance”, and also that the me­dian in­cu­ba­tion pe­riod was 4.8 days from in­fec­tion to symp­tom on­set (in the smaller sam­ple where both of those dates were known).

  Ad­ding 5.5+4.8, that im­plies that an av­er­age of 10.3 days passed be­tween when a per­son be­came in­fected and when they tested pos­i­tive for cases de­tected based on symp­toms, and 8.0 days for those de­tected by con­tact trac­ing. Since the pa­per re­ports that 77% of cases were de­tected through symp­tom-based surveillance, that gives an over­all av­er­age of 9.8 days. (And this is only for the cases that were de­tected; it’s not ad­just­ing at all for peo­ple who were in­fected by never got a pos­i­tive test.)

  That means that in places where test­ing is as good as it was in Shen­zhen, then the num­ber of pos­i­tive tests is tel­ling us about the num­ber of in­fec­tions 9.8 days ago. If the num­ber of cases in that re­gion is dou­bling ev­ery 4 days, then that’s 2.4 dou­blings, so the num­ber of con­firmed cases would only be 18% of the ac­tual num­ber of cases due to the de­lay in test­ing (again, with­out fac­tor­ing in peo­ple who never got tested). (With a 3 day dou­bling pe­riod it would be 10%, with a 5 day dou­bling pe­riod 26%.)

  So in places that don’t have a good test­ing regime it would be sig­nifi­cantly less than that.

• Unnamed 14 Mar 2020 8:18 UTC

  3 points

  in reply to: jimrandomh’s comment on: A Sig­nifi­cant Por­tion of COVID-19 Trans­mis­sion Is Presymptomatic

  Yeah, I agree that con­tact trac­ing & test­ing/​quaran­tin­ing con­tacts is good, and that presymp­tomatic trans­mis­sion is pos­si­ble.

  It looked to me like you were claiming that the hy­poth­e­sis “stop­ping all symp­tomatic trans­mis­sion is suffi­cient to pre­vent the num­ber of COVID-19 cases from curv­ing up­wards” has been tested by some coun­tries’ mea­sures and found to be false, and I am ques­tion­ing that ap­par­ent as­ser­tion.

• Unnamed 14 Mar 2020 7:08 UTC

  3 points

  on: A Sig­nifi­cant Por­tion of COVID-19 Trans­mis­sion Is Presymptomatic

  I no­tice that the es­ti­mates of se­rial in­ter­val (al­most?) all come from places that had pretty ag­gres­sive & suc­cess­ful con­tain­ment mea­sures in place, such as iden­ti­fy­ing & iso­lat­ing po­ten­tial car­ri­ers (in­clud­ing peo­ple who show symp­toms, traced con­tacts, and high-risk trav­el­ers). That would tend to shorten the se­rial in­ter­val, since peo­ple who are iden­ti­fied early in their in­fec­tion lose the op­por­tu­nity to trans­mit dur­ing the later por­tion of their ill­ness.

  Are there es­ti­mates of what R was for these pop­u­la­tions? If it’s a lot less than the 2-3 that other stud­ies have found that would be some ev­i­dence that a lot of later-stage trans­mis­sions were pre­vented.