History's Biggest Natural Experiment 2020-03-24T02:56:30.070Z · score: 97 (44 votes)
COVID-19's Household Secondary Attack Rate Is Unknown 2020-03-16T23:19:47.117Z · score: 63 (16 votes)
A Significant Portion of COVID-19 Transmission Is Presymptomatic 2020-03-14T05:52:33.734Z · score: 118 (43 votes)
Credibility of the CDC on SARS-CoV-2 2020-03-07T02:00:00.452Z · score: 215 (86 votes)
Effectiveness of Fever-Screening Will Decline 2020-03-06T23:00:16.836Z · score: 68 (26 votes)
For viruses, is presenting with fatigue correlated with causing chronic fatigue? 2020-03-04T21:09:48.149Z · score: 18 (6 votes)
Will COVID-19 survivors suffer lasting disability at a high rate? 2020-02-11T20:23:50.664Z · score: 143 (46 votes)
Jimrandomh's Shortform 2019-07-04T17:06:32.665Z · score: 29 (4 votes)
Recommendation Features on LessWrong 2019-06-15T00:23:18.102Z · score: 62 (19 votes)
Welcome to LessWrong! 2019-06-14T19:42:26.128Z · score: 151 (96 votes)
[April Fools] User GPT2 is Banned 2019-04-02T06:00:21.075Z · score: 64 (18 votes)
User GPT2 Has a Warning for Violating Frontpage Commenting Guidelines 2019-04-01T20:23:11.705Z · score: 50 (18 votes)
LW Update 2019-03-12 -- Bugfixes, small features 2019-03-12T21:56:40.109Z · score: 17 (2 votes)
Karma-Change Notifications 2019-03-02T02:52:58.291Z · score: 96 (26 votes)
Two Small Experiments on GPT-2 2019-02-21T02:59:16.199Z · score: 56 (22 votes)
How does OpenAI's language model affect our AI timeline estimates? 2019-02-15T03:11:51.779Z · score: 51 (16 votes)
Introducing the AI Alignment Forum (FAQ) 2018-10-29T21:07:54.494Z · score: 92 (35 votes)
Boston-area Less Wrong meetup 2018-05-16T22:00:48.446Z · score: 4 (1 votes)
Welcome to Cambridge/Boston Less Wrong 2018-03-14T01:53:37.699Z · score: 4 (2 votes)
Meetup : Cambridge, MA Sunday meetup: Lightning Talks 2017-05-20T21:10:26.587Z · score: 0 (1 votes)
Meetup : Cambridge/Boston Less Wrong: Planning 2017 2016-12-29T22:43:55.164Z · score: 0 (1 votes)
Meetup : Boston Secular Solstice 2016-11-30T04:54:55.035Z · score: 1 (2 votes)
Meetup : Cambridge Less Wrong: Tutoring Wheels 2016-01-17T05:23:05.303Z · score: 1 (2 votes)
Meetup : MIT/Boston Secular Solstice 2015-12-03T01:14:02.376Z · score: 1 (2 votes)
Meetup : Cambridge, MA Sunday meetup: The Contrarian Positions Game 2015-11-13T18:08:19.666Z · score: 1 (2 votes)
Rationality Cardinality 2015-10-03T15:54:03.793Z · score: 21 (22 votes)
An Idea For Corrigible, Recursively Improving Math Oracles 2015-07-20T03:35:11.000Z · score: 5 (5 votes)
Research Priorities for Artificial Intelligence: An Open Letter 2015-01-11T19:52:19.313Z · score: 23 (24 votes)
Petrov Day is September 26 2014-09-18T02:55:19.303Z · score: 24 (18 votes)
Three Parables of Microeconomics 2014-05-09T18:18:23.666Z · score: 25 (35 votes)
Meetup : LW/Methods of Rationality meetup 2013-10-15T04:02:11.785Z · score: 0 (1 votes)
Cambridge Meetup: Talk by Eliezer Yudkowsky: Recursion in rational agents 2013-10-15T04:02:05.988Z · score: 7 (8 votes)
Meetup : Cambridge, MA Meetup 2013-09-28T18:38:54.910Z · score: 4 (5 votes)
Charity Effectiveness and Third-World Economics 2013-06-12T15:50:22.330Z · score: 7 (12 votes)
Meetup : Cambridge First-Sunday Meetup 2013-03-01T17:28:01.249Z · score: 3 (4 votes)
Meetup : Cambridge, MA third-Sunday meetup 2013-02-11T23:48:58.812Z · score: 3 (4 votes)
Meetup : Cambridge First-Sunday Meetup 2013-01-31T20:37:32.207Z · score: 1 (2 votes)
Meetup : Cambridge, MA third-Sunday meetup 2013-01-14T11:36:48.262Z · score: 3 (4 votes)
Meetup : Cambridge, MA first-Sunday meetup 2012-11-30T16:34:04.249Z · score: 1 (2 votes)
Meetup : Cambridge, MA third-Sundays meetup 2012-11-16T18:00:25.436Z · score: 3 (4 votes)
Meetup : Cambridge, MA Sunday meetup 2012-11-02T17:08:17.011Z · score: 1 (2 votes)
Less Wrong Polls in Comments 2012-09-19T16:19:36.221Z · score: 79 (82 votes)
Meetup : Cambridge, MA Meetup 2012-07-22T15:05:10.642Z · score: 2 (3 votes)
Meetup : Cambridge, MA first-Sundays meetup 2012-03-30T17:55:25.558Z · score: 0 (3 votes)
Professional Patients: Fraud that ruins studies 2012-01-05T00:20:55.708Z · score: 16 (25 votes)
[LINK] Question Templates 2011-12-23T19:54:22.907Z · score: 1 (1 votes)
I started a blog: Concept Space Cartography 2011-12-16T21:06:28.888Z · score: 6 (9 votes)
Meetup : Cambridge (MA) Saturday meetup 2011-10-20T03:54:28.892Z · score: 2 (3 votes)
Another Mechanism for the Placebo Effect? 2011-10-05T01:55:11.751Z · score: 8 (22 votes)
Meetup : Cambridge, MA Sunday meetup 2011-10-05T01:37:06.937Z · score: 1 (2 votes)


Comment by jimrandomh on Something about the Pinker Cancellation seems Suspicious · 2020-07-09T02:26:57.396Z · score: 2 (1 votes) · LW · GW

It seems fairly normal to me for an emotionally charged movement to attract people for whom it's difficult to tell whether they're not-too-bright fanatics or agents provocateur.

This is a very good observation, and seems like a pretty big problem for such movements.

Comment by jimrandomh on COVID-19's Household Secondary Attack Rate Is Unknown · 2020-07-06T15:36:59.014Z · score: 2 (1 votes) · LW · GW

This paper (from June 27) collects studies published after this post, does meta-analysis, and corrects for some methodological problems like false negative rates, and gives a central estimate of the household secondary attack rate of 30%.

Comment by jimrandomh on Why are all these domains called from Less Wrong? · 2020-06-27T20:43:25.113Z · score: 29 (10 votes) · LW · GW

LessWrong developer here. Here's an overview of what all those domains are. The code is open source, so you should be able to verify these, with some effort.

Algolia (, is a service we use for site search (what you get when you click the magnifying glass icon on the top-bar). They have a mirror of all searchable data (ie non-draft posts and comments, tag pages, user bios); they receive a copy of searches that are performed through the site search box, which they can associate with IP addresses but not with usernames.

Cloudflare is a CDN that is hosting components of MathJax, the Javascript library that renders LaTeX in posts and comments, and some libraries we use for integrating MathJax with the comment/post editors. The CDN URLs were defaults that came with libraries we're using; we could probably move them to our own domain with a little effort. JsDelivr is hosting some things that similarly came with library defaults, as parts of MathJax3 and Algolia.

Cloudinary is an image-hosting CDN that we use for images in some posts and images that are part of the site UI. and are hosting images that were used in posts, presumably because they were visible in the Recent Discussion section when you loaded the front page. Currently, when users insert images into posts, depending how they do it and which editor they're using, it may point to the original domain of the image. Also, for authors we have set up automatic crossposting for, the crossposts will use the original image URLs. We will hopefully switch this to always upload those images to Cloudinary and host them there instead, partially for privacy reasons but mostly to prevent link rot in archives of old posts.

dl.drop is not a valid domain name; it's either a broken image link in some post that was in Recent Discussion, or a typo in this post.

The Google domains are from Google Analytics, Google Tag Manager, Google Fonts, and ReCaptcha. Google Analytics and Google Tag Manager measure site traffic and aggregate usage patterns. is for the chat icon in the bottom-right corner, used for messaging the admins about the site. is LogRocket. We (the devs) use it to see how the site is being used; we can watch anonymized replays of sessions (anonymized in that the username in the corner is edited out). As policy, we don't read people's direct messages or unpublished drafts, or deanonymize votes, though in principle we have the capability to (both with this tool or with direct database access).

TypeKit, aka Adobe Fonts, is a font library and font hosting service. We could probably consolidate this with one of the other CDNs being used, but font-hosting involves some user-agent-string based compatibility polyfills, which would be somewhat annoying to reproduce ourselves.

Comment by jimrandomh on Self-Predicting Markets · 2020-06-11T01:21:50.435Z · score: 2 (1 votes) · LW · GW

Test comment

Comment by jimrandomh on Jimrandomh's Shortform · 2020-06-10T04:28:01.197Z · score: 12 (6 votes) · LW · GW

Twitter is an unusually angry place. One reason is that the length limit makes people favor punchiness over tact. A less well-known reason is that in addition to notifying you when people like your own tweets, it gives a lot of notifications for people liking replies to you. So if someone replies to disagree, you will get a slow drip of reminders, which will make you feel indignant.

LessWrong is a relatively calm place, because we do the opposite: under default settings, we batch upvote/karma-change notifications together to only one notification per day, to avoid encouraging obsessive-refresh spirals.

Comment by jimrandomh on Jimrandomh's Shortform · 2020-06-06T22:29:39.921Z · score: 33 (9 votes) · LW · GW

Despite the justness of their cause, the protests are bad. They will kill at least thousands, possibly as many as hundreds of thousands, through COVID-19 spread. Many more will be crippled. The deaths will be disproportionately among dark-skinned people, because of the association between disease severity and vitamin D deficiency.

Up to this point, R was about 1; not good enough to win, but good enough that one more upgrade in public health strategy would do it. I wasn't optimistic, but I held out hope that my home city, Berkeley, might become a green zone.

Masks help, and being outdoors helps. They do not help nearly enough.

George Floyd was murdered on May 25. Most protesters protest on weekends; the first weekend after that was May 30-31. Due to ~5-day incubation plus reporting delays, we don't yet know how many were infected during that first weekend of protests; we'll get that number over the next 72 hours or so.

We are now in the second weekend of protests, meaning that anyone who got infected at the first protest is now close to peak infectivity. People who protested last weekend will be superspreaders this weekend; the jump in cases we see over the next 72 hours will be about *the square root* of the number of cases that the protests will generate.

Here's the COVID-19 case count dashboard for Alameda County and for Berkeley. I predict a 72 hours from now, Berkeley's case-count will be 170 (50% CI 125-200; 90% CI 115-500).

(Crossposted on Facebook; abridgeposted on Twitter.)

Comment by jimrandomh on How to convince Y that X has committed a murder with >0.999999 probability? · 2020-05-22T20:48:31.485Z · score: 2 (1 votes) · LW · GW

If I sometimes write down a 6-nines confidence number because I'm sleepy, then this affects your posterior probability after hearing that I wrote down a 6-nines confidence number, but doesn't reduce the validity of 6-nines confidence numbers that I write down when I'm alert. The 6-nines confidence number is inside an argument, while your posterior is outside the argument.

Comment by jimrandomh on How to convince Y that X has committed a murder with >0.999999 probability? · 2020-05-22T04:02:33.613Z · score: 2 (1 votes) · LW · GW

Six nines of reliability sounds like a lot, and it's more than is usually achieved in criminal cases, but it's hardly insurmountable. You just need to be confident enough that, given one million similar cases, you would make only one mistake. A combination of recorded video and DNA evidence, with reasonably good validation of the video chain of custody and of the DNA evidence-processing lab's procedures, would probably clear this bar.

Comment by jimrandomh on Why don't we tape surgical masks to the face to seal them airtight? · 2020-05-08T05:33:33.369Z · score: 2 (1 votes) · LW · GW

As a diabetic, I have a few things (insulin infusion canula, continuous glucose monitor) that attach to skin with adhesive. In principle, you could use medical tape around the edges of a normal mask, and it would improve the seal. I think the reason people don't do this is because it's a lot of effort to put on (effort which could be spent improving the fit in other ways), and it's physically painful to take off. This limits its usefulness to the range where an imperfectly-fitted N95 isn't good enough, but a positive-pressure suite isn't necessary; I'm not sure situations in that range are at all common.

Comment by jimrandomh on "Stuck In The Middle With Bruce" · 2020-04-29T18:47:26.866Z · score: 2 (1 votes) · LW · GW

Thanks; I have fixed the link.

Comment by jimrandomh on Will COVID-19 survivors suffer lasting disability at a high rate? · 2020-04-27T23:14:39.895Z · score: 17 (5 votes) · LW · GW

Responding to this news article which is responding to Bornstein et al on the subject of diabetes as a complication of COVID-19 infection.

The paper is primarily about management of COVID19 in patients with existing diabetes, rather than the risk of new-onset diabetes as a result of COVID infection, so it's on shakier ground than you might expect given the news article. The relevant arguments given are: (1) pancreatic beta cells express ACE2 in a mouse model, (2) SARS1 was known to directly damage pancreatic beta cells, and (3) Italian physicians anecdotally report a high rate of DKA in new-onset COVID19 patients (no percentage or citation).

This is strong enough to convince me that this is a thing that happens, for at least a non-negligible (but not necessarily large) subset of the patients who are admitted to ICU.

Some background for people less familiar with diabetes. Pancreatic beta cells produce insulin, which is a hormone that signals to the rest of the body that they should eat the sugar that's in the blood. Under normal circumstances, this keeps blood sugar within a narrow range (70-110mg/dL). However, if the pancreas is damaged or if the pancreatic-function-to-body-size ratio is too low, it can't produce enough insulin, so blood sugar rises higher than it's supposed to. Very high blood sugar is toxic to pancreatic beta cells themselves, causing a feedback loop which leads to a state called diabetic ketoacidosis (DKA), which is reliably fatal if left untreated.

For SARS1, says:

Twenty of the 39 followed-up patients were diabetic during hospitalization. After 3 years, only two of these patients had diabetes.

Unfortunately the paper didn't say how many of those patients had gotten as far as DKA, as opposed to developing diabetes short of DKA in a hospital setting and having it treated promptly. I haven't verified this yet, but my prior belief was that anyone who enters DKA is probably going to be diabetic forever.

This is all separate from the question of whether there's a lasting risk of diabetes in patients who were not hospitalized, or who survived their hospitalization without obvious blood sugar complications. This is a hard question; it seems plausible, but we don't yet have empirical evidence either way. My guess is probably some increased risk, but not a very large one, and decreasing over time.

Comment by jimrandomh on The Chilling Effect of Confiscation · 2020-04-25T18:00:30.578Z · score: 8 (5 votes) · LW · GW

I have a suspicion that some of these seizures are not actually being done by the federal government, but actually are straightforward robberies where the thieves lie about their identity.

Comment by jimrandomh on [Site Meta] Feature Update: More Tags! (Experimental) · 2020-04-23T23:47:32.665Z · score: 3 (2 votes) · LW · GW

Yes, I think That Alien Message should have the AI Alignment tag. (In general, if older posts don't have tags, it mostly means no one has considered yet whether the tag should be applied.)

Comment by jimrandomh on Helping the kids post · 2020-04-17T22:36:41.285Z · score: 16 (6 votes) · LW · GW

These are adorable.

If she does decide to post to LessWrong, the reception will probably be better if it's an Open Thread comment or in the Shortform section, than if it's a top-level post.

Comment by jimrandomh on COVID-19's Household Secondary Attack Rate Is Unknown · 2020-04-15T00:37:38.268Z · score: 4 (2 votes) · LW · GW

This is definitely an improvement over the US CDC and Shenzhen papers, but I still have reservations about it. The first issue is that it's based on calling people and asking about symptoms, not based on testing. So it doesn't count asymptomatic people, nor people with mild symptoms who don't disclose them. The second issue is that their numbers imply an average household size of 6.4, which implies a definition of "household" which is somehow not as expected.

They track contacts of the first 30 identified cases of COVID-19 in South Korea, and find 119 household contacts, of which 9 are infected. Table 2 describes every transmission they found, and whether it was a household transmission. Of the first 30 cases, 8 of them got it by household transmission from someone else who was also one of the first 30 cases, so that's 22 distinct households.

(30 people + (119 contacts - 8 already counted)) / 22 households = 141/22 = 6.4 people per household.

Comment by jimrandomh on Jimrandomh's Shortform · 2020-04-14T20:44:29.588Z · score: 4 (3 votes) · LW · GW

Genetic engineering is ruled out, but gain-of-function research isn't.

Comment by jimrandomh on Jimrandomh's Shortform · 2020-04-14T20:42:28.238Z · score: 15 (7 votes) · LW · GW

First, a clarification: whether SARS-CoV-2 was laboratory-constructed or manipulated is a separate question from whether it escaped from a lab. The main reason a lab would be working with SARS-like coronavirus is to test drugs against it in preparation for a possible future outbreak from a zoonotic source; those experiments would involve culturing it, but not manipulating it.

But also: If it had been the subject of gain-of-function research, this probably wouldn't be detectable. The example I'm most familiar with, the controversial 2012 US A/H5N1 gain of function study, used a method which would not have left any genetic evidence of manipulation.

Comment by jimrandomh on Jimrandomh's Shortform · 2020-04-14T19:55:58.508Z · score: 7 (4 votes) · LW · GW

That's overstating it. They're the only BSL-4 lab. Whether BSL-3 labs were allowed to deal with this class of virus, is something that someone should research.

Comment by jimrandomh on Jimrandomh's Shortform · 2020-04-14T19:25:16.438Z · score: 11 (7 votes) · LW · GW

I agree that this is technically correct, but the prior for "escaped specifically from a lab in Wuhan" is also probably ~100 times lower than the prior for "escaped from any biolab in China"

I don't think this is true. The Wuhan Institute of Virology is the only biolab in China with a BSL-4 certification, and therefore is probably the only biolab in China which could legally have been studying this class of virus. While the BSL-3 Chinese Institute of Virology in Beijing studied SARS in the past and had laboratory escapes, I expect all of that research to have been shut down or moved, given the history, and I expect a review of Chinese publications will not find any studies involving live virus testing outside of WIV. While the existence of one or two more labs in China studying SARS would not be super surprising, the existence of 100 would be extremely surprising, and would be a major scandal in itself.

Comment by jimrandomh on Jimrandomh's Shortform · 2020-04-14T19:16:48.052Z · score: 70 (29 votes) · LW · GW

I am now reasonably convinced (p>0.8) that SARS-CoV-2 originated in an accidental laboratory escape from the Wuhan Institute of Virology.

1. If SARS-CoV-2 originated in a non-laboratory zoonotic transmission, then the geographic location of the initial outbreak would be drawn from a distribution which is approximately uniformly distributed over China (population-weighted); whereas if it originated in a laboratory, the geographic location is drawn from the commuting region of a lab studying that class of viruses, of which there is currently only one. Wuhan has <1% of the population of China, so this is (order of magnitude) a 100:1 update.

2. No factor other than the presence of the Wuhan Institute of Virology and related biotech organizations distinguishes Wuhan or Hubei from the rest of China. It is not the location of the bat-caves that SARS was found in; those are in Yunnan. It is not the location of any previous outbreaks. It does not have documented higher consumption of bats than the rest of China.

3. There have been publicly reported laboratory escapes of SARS twice before in Beijing, so we know this class of virus is difficult to contain in a laboratory setting.

4. We know that the Wuhan Institute of Virology was studying SARS-like bat coronaviruses. As reported in the Washington Post today, US diplomats had expressed serious concerns about the lab's safety.

5. China has adopted a policy of suppressing research into the origins of SARS-CoV-2, which they would not have done if they expected that research to clear them of scandal. Some Chinese officials are in a position to know.

To be clear, I don't think this was an intentional release. I don't think it was intended for use as a bioweapon. I don't think it underwent genetic engineering or gain-of-function research, although nothing about it conclusively rules this out. I think the researchers had good intentions, and screwed up.

Comment by jimrandomh on Jimrandomh's Shortform · 2020-04-04T05:45:01.934Z · score: 15 (4 votes) · LW · GW

This tweet raised the question of whether masks really are more effective if placed on sick people (blocking outgoing droplets) or if placed on healthy people (blocking incoming droplets). Everyone in public or in a risky setting should have a mask, of course, but we still need to allocate the higher-quality vs lower-quality masks somehow. When sick people are few and are obvious, and masks are scarce, masks should obviously go on the sick people. However, COVID-19 transmission is often presymptomatic, and masks (especially lower-quality improvised masks) are not becoming less scarce over time.

If you have two people in a room and one mask, one infected and one healthy, which person should wear the mask? Thinking about the physics of liquid droplets, I think the answer is that the infected person should wear it.

  1. A mask on a sick person prevents the creation of fomites; masks on healthy people don't.
  2. Outgoing particles have a larger size and shrink due to evaporation, so they'll penetrate a mask less, given equal kinetic energy. (However, kinetic energies are not equal; they start out fast and slow down, which would favor putting the mask on the healthy person. I'm not sure how much this matters.)
  3. Particles that stick to a mask but then un-stick lose their kinetic energy in the process, which helps if the mask is on the sick person, but doesn't help if the mask is on the healthy person.

Overall, it seems like for a given contact-pair, a mask does more good if it's on the sick person. However, mask quality also matters in proportion to the number of healthy-sick contacts it affects; so, upgrading the masks of all of the patients in a hospital would help more than upgrading the masks of all the workers in that hospital, but since the patients outnumber the workers, upgrading the workers' masks probably helps more per-mask.

Comment by jimrandomh on Has the effectiveness of fever screening declined? · 2020-03-27T22:27:15.276Z · score: 4 (2 votes) · LW · GW

For now, mods do it. This will be expanded to all users and to a wider variety of tags when tagging leaves beta.

Comment by jimrandomh on History's Biggest Natural Experiment · 2020-03-25T18:00:35.440Z · score: 5 (3 votes) · LW · GW

It would show up as people with a particular year of birth having a much lower risk than people born one year earlier or later. Since most research includes collecting date of birth, this should be easy to check.

Comment by jimrandomh on Virus As A Power Optimisation Process: The Problem Of Next Wave · 2020-03-22T22:43:43.795Z · score: 13 (6 votes) · LW · GW

Biological global catastrophic risks were neglected for years, while AGI risks were on the top.

This is a true statement about the attention allocation on LessWrong, but definitely not a true statement about the world's overall resource allocation. Total spending on pandemic preparedness is and was orders of magnitude greater than spending on AGI risk. It's just a hard problem, which requires a lot of expensive physical infrastructure to prepare for.

Comment by jimrandomh on Will COVID-19 survivors suffer lasting disability at a high rate? · 2020-03-22T04:21:51.304Z · score: 5 (3 votes) · LW · GW

"Impaired consciousness" doesn't sound unusual for patients with severe fever, but five strokes out of 214 hospitalized patients is pretty noteworthy.

Comment by jimrandomh on SARS-CoV-2 pool-testing algorithm puzzle · 2020-03-21T06:23:20.857Z · score: 2 (1 votes) · LW · GW

It's a newspaper article based on an unpublished paper; that reference class of writing can't be trusted to report the caveats.

(I could be wrong about the mechanics of PCR; I'm not an expert in it; but the article itself doesn't provide much information about that.)

Comment by jimrandomh on SARS-CoV-2 pool-testing algorithm puzzle · 2020-03-20T18:19:16.943Z · score: 2 (1 votes) · LW · GW

This can only be used on groups where everyone is asymptomatic, and there will be low limits on the pool size even then.

The first step of a PCR test is RNA amplification; you use enzymes which take a small amount of RNA in the sample, and produce a large number of copies. The problem is that there are other RNA viruses besides SARS-CoV-2, such as influenza, and depending when in the disease course the samples were taken, the amount of irrelevant RNA might exceed the amount of SARS-CoV-2 RNA by orders of magnitude, which would lead to a false negative.

Comment by jimrandomh on Preprint says R0=~5 (!) / infection fatality ratio=~0.1%. Thoughts? · 2020-03-20T17:59:38.827Z · score: 22 (9 votes) · LW · GW

tl;dr: Someone wrote buggy R code and rushed a preprint out the door without proofreading or sanity checking the numbers.

The main claim of the paper is this:

The total number of estimated laboratory–confirmed cases (i.e. cumulative cases) is 18913 (95% CrI: 16444–19705) while the actual numbers of reported laboratory–confirmed cases during our study period is 19559 as of February 11th, 2020. Moreover, we inferred the total number of COVID-19 infections (Figure S1). Our results indicate that the total number of infections (i.e. cumulative infections) is 1905526 (95%CrI: 1350283– 2655936)

So, they conclude that less than 1% of cases were detected. They claim 95% confidence that no more than 1.5% of cases were detected. They combine this with the (unstated) assumption that 100% of deaths were detected and reported, and that therefore the IFR is two orders of magnitude lower than is commonly believed. This is an extraordinary claim, which the paper doesn't even really acknowledge; they just sort of throw numbers out and fail to mention that their numbers are wildly different from everyone else's. Their input data is

the daily series of laboratory–confirmed COVID-19 cases and deaths  in Wuhan City and epidemiological data of Japanese evacuees from Wuhan City on board government–chartered flights

This is not a dataset which is capable of supporting such a conclusion. On top of that, the paper has other major signals of low quality. The paper is riddled with typos. And there's this bit:

Serial interval estimates of COVID-19 were derived from previous studies of nCov, indicating that it follows a gamma distribution with the mean and SD at 7.5 and 3.4 days, respectively, based on [14]

In this post I collected estimates of COVID-19's serial interval. 7.5 days was the chronologically first published estimate, was the highest estimate, and was an outlier with small sample size. Strangely, reference [14] does not point to the paper which estimated 7.5 days; that's reference 21, whereas reference 14 points to this paper which makes no mention of the serial interval at all.

Comment by jimrandomh on Is the Covid-19 crisis a good time for x-risk outreach? · 2020-03-20T00:13:56.022Z · score: 20 (8 votes) · LW · GW

Right now, most people are hyperfocused on COVID-19; this creates an obvious incentive for people to try to tie their pet issues to it, which I expect a variety of groups to try and which I expect to mostly backfire if tried in the short run. (See for example the receptiontthe WHO got when they tried to talk about stigma and discriminatio; people interpreted it as the output of an "always tie my pet issue to the topic du jour" algorithm and ridiculed then for it. Talking about AI risk in the current environment risks provoking the same reaction, because it probably would in fact be coming from a tie-my-pet-topic algorithm.

A month from now, however, will be a different matter. Once people start feeling like they have attention to spare, and have burned out on COVID-19 news, I expect them to be much more receptive to arguments about tail risk and to model-based extrapolation of the future than they were before.

Comment by jimrandomh on Covid-19 Points of Leverage, Travel Bans and Eradication · 2020-03-19T18:36:47.000Z · score: 10 (5 votes) · LW · GW

To start, the severity estimates that Joshua assumed were worst case and are implausible. The very alarmist Fergeson et al paper has much lower numbers than Joshua's [Joscha Bach's] claim that "20% will develop a severe case and need medical support to survive."

I believe the 20% figure comes from the WHO joint report which says

13.8% have severe disease (dyspnea, respiratory frequency ≥30/minute, blood oxygen saturation ≤93%, PaO2/FiO2 ratio <300, and/or lung infiltrates >50% of the lung field within 24-48 hours) and 6.1% are critical (respiratory failure, septic shock, and/or multiple organ dysfunction/failure).

There are a lot of modeling assumptions that go into this, and the true number is probably lower, but not so low as to invalidate Joscha's point.

Comment by jimrandomh on [UPDATED] COVID-19 cabin secondary attack rates on Diamond Princess · 2020-03-19T01:22:56.007Z · score: 15 (4 votes) · LW · GW

Thank you, this is exactly the sort of clever analysis I was hoping people would come up with when I wrote my post.

This site has floor-plan images of Diamond Princess cabins, from which we can make a few inferences about cabin occupancy. Most of the cabin layouts contain a single bed which fits two people, so two-person cabins will almost exclusively couples sharing a bed. If I assume the rate at which people in single-person cabins get infected (8%) is the rate of infection outside the cabin, and that the higher rate of infection in two-person cabins is caused entirely by within-cabin secondary transmission, then it looks like each person would have to infect their partner an average of 1.5 times each. This also tells us that the transmission rate between elderly couples sharing a bed is likely to be extremely high, and also that people in single-person cabins must be different in some way--perhaps they spent less time in the ship's common areas.

Three- and four-person cabins seem harder to interpret. These would originally have been couples with children, but there aren't many children aboard as of Feb 5th, and they probably moved people around to free up single cabins for extra-vulnerable people and for confirmed cases that they needed to isolate.

Comment by jimrandomh on COVID-19's Household Secondary Attack Rate Is Unknown · 2020-03-19T00:36:41.760Z · score: 2 (1 votes) · LW · GW

This paper analyzes specific incidents in which a group of one infected person plus some uninfected people sat down together, and some uninfected people got it. They find a secondary attack rate (from mostly non-household interactions) of 35%.

There are two big issues that prevent this paper from being used to draw good inferences about the household secondary attack rate. First, the incidents were found by specifically looking for superspreading events, and does not include any events where transmission didn't happen. And second, the events are single gatherings, whereas living with someone may involve many opportunities to get infected.

Comment by jimrandomh on Does the 14-month vaccine safety test make sense for COVID-19? · 2020-03-18T19:48:28.914Z · score: 11 (7 votes) · LW · GW

The main piece of data that would help answer this question is case-studies of past vaccines, whether they had safety problems and what those problems were, and when the problems manifested. Given that there's a new influenza vaccine every year, and I've never heard of any year's influenza vaccine being rejected on safety grounds, my guess is that 18 months is much too conservative.

Comment by jimrandomh on Ubiquitous Far-Ultraviolet Light Could Control the Spread of Covid-19 and Other Pandemics · 2020-03-18T19:09:02.539Z · score: 10 (2 votes) · LW · GW

The indentation of the table of contents is determined by the heading levels. It looks like you may have set some of the headings to "Heading 2" style and others to "Heading 3" style in Docs, then adjusted the font size to make them look the same. If you use Docs' heading format presets and use the same one for all the sections, they should be at the same indent level.

Comment by jimrandomh on Ubiquitous Far-Ultraviolet Light Could Control the Spread of Covid-19 and Other Pandemics · 2020-03-18T19:00:06.709Z · score: 2 (1 votes) · LW · GW

Unfortunately the LW codebase doesn't support multiple authors with edit access to the same post yet; we're working on this (as part of a broader overhaul of the post editor, which will allow Google Docs-style simultaneous editing), but it isn't ready yet. In the mean time, the easiest way to handle this is to make edits in a shared Google Doc, and have the primary author paste them in and save. (Copy-paste between the post editor and Google Docs should just work.)

Comment by jimrandomh on Ubiquitous Far-Ultraviolet Light Could Control the Spread of Covid-19 and Other Pandemics · 2020-03-18T18:29:14.517Z · score: 6 (3 votes) · LW · GW

(I added Roko to the metadata as a coauthor. Tagged coauthors are a beta-feature which currently can only be edited by moderators, since we haven't implemented the mechanics of having authors approve each other.)

Comment by jimrandomh on Good News: the Containment Measures are Working · 2020-03-17T16:24:47.215Z · score: 2 (1 votes) · LW · GW

In this post I collected papers which estimate the incubation period (time from exposure to symptom onset) and the serial interval (time from exposure to infecting the next person in the chain). Studies get varying results because they're done in different populations and have methodological differences, but find reasonably similar medians. A few of them also provide full distributions over incubation periods and serial intervals.

Comment by jimrandomh on Good News: the Containment Measures are Working · 2020-03-17T08:14:00.212Z · score: 8 (4 votes) · LW · GW

What does it mean for the future? That it takes about a week of a severe lockdown to switch from the exponential growth to linear, about two weeks to switch to leveling off, and three to four weeks to start seeing a meaningful decline.

This is because of the incubation period (3-14d) and the delay between people becoming symptomatic and getting tested. The reduction in transmissions is immediate, it just takes awhile to notice.

Comment by jimrandomh on COVID-19's Household Secondary Attack Rate Is Unknown · 2020-03-17T00:09:19.577Z · score: 15 (4 votes) · LW · GW

R0 is the number of people that each person will go on to infect, on average. R0 for COVID-19 is high compared to other common diseases, indicating high transmissibility.

I avoided stating a quantitative estimate of the attack rate because my confidence intervals are too wide to be useful. If I had to bet, I'd say 90% CI 15-85%, 50% CI 30-65%. I'm hoping people can gather weak evidence of various forms (secondary attack rates and R0s of other diseases, anecdotes in which household members do or don't get it, or in the best case a dataset with household memberships labelled).

Comment by jimrandomh on Will COVID-19 survivors suffer lasting disability at a high rate? · 2020-03-16T23:13:01.281Z · score: 13 (3 votes) · LW · GW

Relevant post by Sarah Constantin: COVID-19 Risks For Young People

Comment by jimrandomh on Could you save lives in your community by buying oxygen concentrators from Alibaba? · 2020-03-16T20:52:12.936Z · score: 8 (4 votes) · LW · GW

This isn't a question that can really be answered, because access to resources like RNs and places to isolate people is very local but the top-level post is about the effectiveness of oxygen concentrators in general.

That said, the Bay Area rationality community, at least, does have these things.

Comment by jimrandomh on A Significant Portion of COVID-19 Transmission Is Presymptomatic · 2020-03-15T04:56:25.720Z · score: 8 (4 votes) · LW · GW

Looks like I was about a day ahead of the curve. I wonder whether Elizabeth Cohen reads LessWrong? In any case, I'm glad the media is now making this more well known. That article is mostly based on the same papers as my post, but adds some other evidence, such as Sandra Ciesek's measurements of viral shedding,

Comment by jimrandomh on A Significant Portion of COVID-19 Transmission Is Presymptomatic · 2020-03-14T17:23:31.077Z · score: 6 (3 votes) · LW · GW

I found this article which makes that claim, and cites the WHO joint mission report. However, I can't find the claim (or the number "25%" used in any capacity) in the cited document at all. I suspect that the article mixed up percent of cases which are asymptomatic with the percent of transmissions which are from asymptomatic sources.

Comment by jimrandomh on A Significant Portion of COVID-19 Transmission Is Presymptomatic · 2020-03-14T17:02:32.946Z · score: 6 (4 votes) · LW · GW

Tapiwa Ganyani et al, the one with quantitative estimates of presymptomatic transmission, used data from Singapore up to Feb 26th and Tianjin up to Feb 27th. Both are cities which seem to have achieved containment over the relevant time period. I'm going to focus on Singapore, because information about what Singapore has been doing is easier to come by.

This article has a graph of Singapore's case counts over the relevant time period, and appears to show it as being contained (R<1) during the relevant time period. This paper estimates the latency between people becoming symptomatic and being isolated in Singapore, and finds that it's about 3 days for local cases as of Feb 26th, but longer on earlier dates. Their endpoint is "hospitalization or quarantine", but reading Singapore's FAQ, it sounds like they have a tiered system with two lesser levels of isolation that the paper doesn't mention: leave of absence (advice to isolate with no legal force) and stay-home notice (which has legal force against leaving the house, but only a non-binding advisory against having visitors). The lesser isolation tiers are for traced contacts, and would be effective at preventing asymptomatic transmission as well.

My impression from all this is that Singapore's measures would have driven symptomatic transmission down more than asymptomatic transmission, but substantially driven down both.

Comment by jimrandomh on A Significant Portion of COVID-19 Transmission Is Presymptomatic · 2020-03-14T07:05:53.780Z · score: 3 (2 votes) · LW · GW

When they identify a person with symptoms, they certainly isolate that person, but they also do contact tracing. Standard practice is to tell everyone who's interacted with an infected person to self-quarantine for 14 days, even if they aren't symptomatic. If that happens soon enough and is thorough enough, it can in principle prevent all presymptomatic transmission.

Comment by jimrandomh on A Significant Portion of COVID-19 Transmission Is Presymptomatic · 2020-03-14T06:22:07.307Z · score: 5 (3 votes) · LW · GW

Asked on Facebook crosspost: If it's transmitted by cough, how do asymptomatic people transmit it? Is it dependent on other diseases or allergies to provoke coughing?

None of the studies cited address the question of how presymptomatic transmission occurs, but my guess is that it's the same: respiratory droplets. People also emit droplets when they talk and breathe. There aren't as many and they don't travel as far as when they cough, but they're there.

Comment by jimrandomh on Will COVID-19 survivors suffer lasting disability at a high rate? · 2020-03-11T23:50:57.986Z · score: 25 (6 votes) · LW · GW

Acute respiratory distress syndrome (ARDS) occurs in about 20% of the subset of patients severe enough to be hospitalized, which are themselves the ~20% more-severe subset. This study finds 50% survival among patients with ARDS, and 91% survival among patients without. So, very approximately, the risk of death-or-ARDS is about twice as high as the risk of death.

Functional Disability 5 Years after Acute Respiratory Distress Syndrome tracked the outcomes of patients with ARDS from other sources, mostly pneumonia (not from COVID-19) but also sepsis (other infections), trauma, and other causes.

Main findings: ~10% died within 1 year, ~20% within 5 years. Of those who survived to 5-year followup, score on the SF-36 physical health survey was reduced 1 SD relative to age-matched controls. At 5-year follow-up, 77% of patients had returned to work, of which 94% of these patients returned to their original work.

Comment by jimrandomh on How useful are masks during an epidemic? · 2020-03-11T21:53:25.582Z · score: 18 (7 votes) · LW · GW

The majority of studies either compare two different kinds of masks to each other, in a hospital setting, which is interesting but not the main thing we want, which is a comparison between mask and no mask. There are also a few studies looking at households where the housemates of a person with influenza-like illness are randomly assigned to use masks (or not); these studies find masks are not effective, but also show low compliance. In a household setting, most of the variance in whether a household member gets infected is likely explained by vaccination, other sources of prior immunity, and kitchen hygiene; whereas when considering infections of coronavirus occurring in public, none of these are factors. All studies had problems with noncompliance, and confounding between compliance and other precautions (ie people who comply with masks wash their hands at rates.)

This study is interesting in that it effectively has a "placebo mask" arm. There were three arms: a no-intervention arm in which health care workers continued wearing whatever masks they did before (less often than in either intervention arm, but still a significant amount), an intervention arm where hospital workers are given reusable cloth masks which do not work, and an intervention arm in which they're given disposable medical masks. The disposable medical mask arm did best, the cloth-mask arm did worst (worse than the no-intervention arm, due to a combination of not using other masks and taking fewer non-mask precautions). Presumably, this was able to get past an IRB because the study authors didn't know how bad cloth masks are.

Compliance (defined as "mask wearing more than 70% of working hours") was 57% in both the cloth mask and medical groups, and 24% in the no-intervention group.

The majority of infections in the study were from rhinovirus, which is transmitted via aerosol and contact droplets. These are the same modes of transmission as SARS-CoV2, but in different proportions; rhinovirus causes sneezing, so it generates a lot of aerosol, whereas SARS-CoV2 doesn't and aerosol is believed to be responsible for only a small portion of its transmissions. Other diagnosed infections in the study were from hMPV and influenza B. Rhinovirus has a smaller diameter than SARS-CoV2 (30nm), so it's unlikely that mask aerosol penetration of rhinovirus is higher than that of SARS-CoV2. In particle penetration tests, the cloth masks were almost completely ineffective, and the medical masks had some effectiveness but much less than N95.

Compared to cloth masks, medical masks reduced clinical respiratory illness from 7.6 to 4.8%, laboratory-confirmed viral infection from 5.4 to 3.3%, and influenza-like illness from 2.3 to 0.2%.

Comment by jimrandomh on March Coronavirus Open Thread · 2020-03-10T18:37:00.496Z · score: 17 (7 votes) · LW · GW

Legally speak, yes they would. Practically speaking, however, the FDA has no enforcement power over secret programs in the intelligence community.

I think a lot of people are seriously overestimating the FDA's actual power, and that's causing pretty severe problems. Consider for example this tweet (and a long series like it) by the mayor of NYC, begging the FDA for approvals. While there is no legal precedent to refer to, it's extremely implausible that the FDA could ever get or enforce a judgment of the city of New York for actions taken during a state of emergency, when the FDA itself caused that emergency with culpable negligence.

Comment by jimrandomh on March Coronavirus Open Thread · 2020-03-10T04:00:41.191Z · score: 4 (2 votes) · LW · GW

In principle, with enough resources, multiple vaccines could be tested this way in parallel. Not that there are that many vaccine candidates to try, as far as I know; but if there were some software that bulk-generated candidate molecules, it could be done, in principle. The limiting input is mobilized resources, not time.