Posts

Outperforming the human Atari benchmark 2020-03-31T19:33:46.355Z · score: 58 (22 votes)
Mod Notice about Election Discussion 2020-01-29T01:35:53.947Z · score: 58 (19 votes)
Circling as Cousin to Rationality 2020-01-01T01:16:42.727Z · score: 72 (35 votes)
Self and No-Self 2019-12-29T06:15:50.192Z · score: 47 (17 votes)
T-Shaped Organizations 2019-12-16T23:48:13.101Z · score: 51 (14 votes)
ialdabaoth is banned 2019-12-13T06:34:41.756Z · score: 30 (17 votes)
The Bus Ticket Theory of Genius 2019-11-23T22:12:17.966Z · score: 64 (18 votes)
Vaniver's Shortform 2019-10-06T19:34:49.931Z · score: 10 (1 votes)
Vaniver's View on Factored Cognition 2019-08-23T02:54:00.915Z · score: 41 (9 votes)
Conversation on forecasting with Vaniver and Ozzie Gooen 2019-07-30T11:16:58.633Z · score: 41 (10 votes)
Commentary On "The Abolition of Man" 2019-07-15T18:56:27.295Z · score: 65 (15 votes)
Is there a guide to 'Problems that are too fast to Google'? 2019-06-17T05:04:39.613Z · score: 49 (15 votes)
Welcome to LessWrong! 2019-06-14T19:42:26.128Z · score: 122 (72 votes)
Steelmanning Divination 2019-06-05T22:53:54.615Z · score: 152 (61 votes)
Public Positions and Private Guts 2018-10-11T19:38:25.567Z · score: 95 (30 votes)
Maps of Meaning: Abridged and Translated 2018-10-11T00:27:20.974Z · score: 54 (22 votes)
Compact vs. Wide Models 2018-07-16T04:09:10.075Z · score: 32 (13 votes)
Thoughts on AI Safety via Debate 2018-05-09T19:46:00.417Z · score: 88 (21 votes)
Turning 30 2018-05-08T05:37:45.001Z · score: 75 (24 votes)
My confusions with Paul's Agenda 2018-04-20T17:24:13.466Z · score: 90 (22 votes)
LW Migration Announcement 2018-03-22T02:18:19.892Z · score: 139 (37 votes)
LW Migration Announcement 2018-03-22T02:17:13.927Z · score: 2 (2 votes)
Leaving beta: Voting on moving to LessWrong.com 2018-03-11T23:40:26.663Z · score: 6 (6 votes)
Leaving beta: Voting on moving to LessWrong.com 2018-03-11T22:53:17.721Z · score: 139 (42 votes)
LW 2.0 Open Beta Live 2017-09-21T01:15:53.341Z · score: 23 (23 votes)
LW 2.0 Open Beta starts 9/20 2017-09-15T02:57:10.729Z · score: 24 (24 votes)
Pair Debug to Understand, not Fix 2017-06-21T23:25:40.480Z · score: 8 (8 votes)
Don't Shoot the Messenger 2017-04-19T22:14:45.585Z · score: 11 (11 votes)
The Quaker and the Parselmouth 2017-01-20T21:24:12.010Z · score: 6 (7 votes)
Announcement: Intelligence in Literature Prize 2017-01-04T20:07:50.745Z · score: 9 (9 votes)
Community needs, individual needs, and a model of adult development 2016-12-17T00:18:17.718Z · score: 12 (13 votes)
Contra Robinson on Schooling 2016-12-02T19:05:13.922Z · score: 4 (5 votes)
Downvotes temporarily disabled 2016-12-01T17:31:41.763Z · score: 17 (18 votes)
Articles in Main 2016-11-29T21:35:17.618Z · score: 3 (4 votes)
Linkposts now live! 2016-09-28T15:13:19.542Z · score: 27 (30 votes)
Yudkowsky's Guide to Writing Intelligent Characters 2016-09-28T14:36:48.583Z · score: 4 (5 votes)
Meetup : Welcome Scott Aaronson to Texas 2016-07-25T01:27:43.908Z · score: 1 (2 votes)
Happy Notice Your Surprise Day! 2016-04-01T13:02:33.530Z · score: 14 (15 votes)
Posting to Main currently disabled 2016-02-19T03:55:08.370Z · score: 22 (25 votes)
Upcoming LW Changes 2016-02-03T05:34:34.472Z · score: 46 (47 votes)
LessWrong 2.0 2015-12-09T18:59:37.232Z · score: 92 (96 votes)
Meetup : Austin, TX - Petrov Day Celebration 2015-09-15T00:36:13.593Z · score: 1 (2 votes)
Conceptual Specialization of Labor Enables Precision 2015-06-08T02:11:20.991Z · score: 10 (11 votes)
Rationality Quotes Thread May 2015 2015-05-01T14:31:04.391Z · score: 9 (10 votes)
Meetup : Austin, TX - Schelling Day 2015-04-13T14:19:21.680Z · score: 1 (2 votes)
Sapiens 2015-04-08T02:56:25.114Z · score: 42 (36 votes)
Thinking well 2015-04-01T22:03:41.634Z · score: 28 (29 votes)
Rationality Quotes Thread April 2015 2015-04-01T13:35:48.660Z · score: 7 (9 votes)
Meetup : Austin, TX - Quack's 2015-03-20T15:12:31.376Z · score: 1 (2 votes)
Rationality Quotes Thread March 2015 2015-03-02T23:38:48.068Z · score: 8 (8 votes)

Comments

Comment by vaniver on Has LessWrong been a good early alarm bell for the pandemic? · 2020-04-03T22:35:05.145Z · score: 7 (4 votes) · LW · GW

"Will ncov survivors suffer lasting disability at a high rate?" is a medical question that makes no implication about broader covid risk.

This seems wrong to me, in part because the hypothesis that there could be widespread negative effects even for survivors was a compelling reason for 1) me to take it seriously (at the time, I estimated my disability risk was something like 5x the importance of my mortality risk) and 2) people to expect spread to be bad in a way that shows up in many indicators (like GDP).

Comment by vaniver on Has LessWrong been a good early alarm bell for the pandemic? · 2020-04-03T20:42:08.697Z · score: 11 (4 votes) · LW · GW

Even with them included, though, there's an obvious sharp discontinuity in volume of posting starting on Feb. 27th. 

Is this the right comparison? It seems to me that the interesting question is "what was the balance of information available on LW before the 20th?" and "how much information was there on LW before the 20th?", not "did the amount of discussion on LW increase over time?". In worlds where we had posted the perfect pandemic survival guide on Feb 11th, and then as more and more people realized the crisis was real, posted questions here, the posts graph would look a lot like the one you posted.

To the best of my knowledge, LW had very little minimization or pushback against preparation (and what pushback I recall was generally of precautions that probably were too extreme or could have been accomplished more cheaply).

Based on that, I'd say that we have not yet achieved any kind of substantial collective coordination or debiasing. By and large, we're still passively waiting for the consensus to come to us and shake us out of complacency.

I mean, this seems true in the sense that most online communities are 90% lurkers.

I think my main regret was something like "assuming more people were on top of this," explicitly or implicitly; I was spooked and preparing early enough that when I went on one of my last outings to get something with Duncan, our conversation spooked him and he started preparing, and then he was on top of it in time to help other people prepare. But you won't see any warnings from me on LessWrong or Facebook, because my reaction was closer to "ah now I have a bunch of chores to get ready myself, and I'm not making inferences on any private data, so others can come to the same conclusion if they want to" than "oh jeez I need to make sure everyone is aware of this / I need to publicly vouch for my inference to best shift the group epistemology."

But also most of the conversations that I was having about this were private, in one way or another; what should this group house do? What do I think of this draft doc my boyfriend's group house is working on? Some of this eventually made its way to LW, mostly through the actions of a few heroes, but also often after days of private discussion and fact-checking.

Comment by vaniver on March 2020 gwern.net newsletter · 2020-04-03T18:47:48.119Z · score: 6 (3 votes) · LW · GW

From the article about Cameo:

(There are other uses, too: New York Times media columnist Ben Smith just paid former Giants linebacker Leonard Marshall to convince his dad to take more precautions against the coronavirus.)

Comment by vaniver on How special are human brains among animal brains? · 2020-04-02T18:53:39.730Z · score: 2 (1 votes) · LW · GW

The other question is whether we'd allow another species to master language. I've never considered this question before, but my guess is that we would.

At this point, we'd encourage it. (See people trying to communicate with dolphins, or dogs, or gorillas, or parrots, or...)

But the relevant time period was probably when there were multiple species in the homo genus; as most similar to humans, they were probably also the fewest steps away from language and also the most likely to be a competitor for the same ecological niche. There's much more reward to anatomically modern humans for driving neanderthals to extinction than driving parrots to extinction, and so we don't see our near competitors in the race to language anymore. 

Comment by vaniver on How special are human brains among animal brains? · 2020-04-02T18:47:33.406Z · score: 2 (1 votes) · LW · GW

Perhaps the idea is something like "Some species had to get there first. That species will be the 'first observer', in some meaningful sense. Whenever that happened, and whatever species became that first observer, there'd likely be a while in which no other species had language, and that species wondered why that was so."

I think this is the idea. You're right that it doesn't change our estimate of how difficult language is from the non-existence of a second species with language; the thing that it does is point out that "even if you observe 1 element of a rare set, you shouldn't think the set is common instead of rare, because you were conditioning on observing at least one element of that set." [That is, we're not seeing any of the planets that have life but no language, or directly observing any of the 50 kiloyear time periods when Earth was one of those.]

Comment by vaniver on Taking Initial Viral Load Seriously · 2020-04-02T01:21:50.203Z · score: 7 (4 votes) · LW · GW

The biggest change that this suggests, I believe, is that people should not isolate with their families if they become sick. (Basically for the same reason that the first kid with measles is ok, but their siblings are in trouble.)

The 'temporary hospitals', where someone with COVID goes there to recover and hopefully doesn't infect anyone else (or get too much additional virus from the other patients), seem like they get around this.

Comment by vaniver on Why do we have offices? · 2020-04-01T23:47:40.446Z · score: 4 (2 votes) · LW · GW

I think high bandwidth communication (and meta communication) is the core factor, with social presence and accountability as secondary factors.

Comment by vaniver on Why do we have offices? · 2020-04-01T23:46:05.841Z · score: 4 (2 votes) · LW · GW

Note that this sort of inspectability is, in principle, accessible for remote work; you could have software that captures what's going on for each employee's screen (or webcam), and either randomly sample it or programmatically check for long periods of inactivity, or attempts to fool the system, or so on.

Comment by vaniver on Is the coronavirus the most important thing to be focusing on right now? · 2020-04-01T21:13:55.260Z · score: 4 (2 votes) · LW · GW

I do indeed think that during World War 2 it would have been reasonable for many people on LessWrong to participate in the war effort, and think the same is true in this case.

It feels to me like there are three reasons this could be the case:

  1. Counterfactual impact on the war; if the LWers of the time chose to act instead of not act, they shift the probabilities of who ends up winning / what collateral damage happens over the course of the resolution.
  2. Social obligation; if LW conscientiously objected from doing its part, or thought other things were more important, this would be terrible PR / weaken LW's position after the fact. (Or maybe the reason to be an EA and the reason to sign up to fight in the war have a common cause that's hard to turn off.)
  3. Ability to impact other things that happen as a result of war participation; sign up, be excellent, get promoted, and then set up good systems that last after the crisis. (This looks like the standard argument for being in public service, except argues it's an unusually good time to enter it.)

Is this basically what you had in mind, or is there something else I'm missing?

Comment by vaniver on When to assume neural networks can solve a problem · 2020-03-27T22:33:05.132Z · score: 5 (2 votes) · LW · GW

That’s partly because I’ve never seen a consistent top-to-bottom reasoning for it.

I think it's difficult to find a consistent top-to-bottom story because the overall argument is disjunctive.

That is, a conjunction is the intersection of different events ("the sidewalk is wet and it's raining" requires it to both be true that "the sidewalk is wet" and "it's raining"), whereas a disjunction is the union of different (potentially overlapping) events ("the sidewalk is wet" can be reached by either "the sidewalk is wet and it's raining" and "the sidewalk is wet and the fire hydrant is leaking").

So if you have a conclusion, like "autonomous vehicles will be commercially available in 2030", the more different ways there are for it to be true, the more likely it is. But also, the more different ways there are for it to be true, the less it makes sense to commit to any particular way. "Autonomous cars are commercially available in 2030 because Uber developed them" has more details, but those details are burdensome.

Also, it seems important to point out that the Bostromian position is about the future. That is, the state of autonomous vehicles today can tell you about whether or not they'll be commercially available in 2030, but there's no hard evidence and it requires careful reasoning to just get provisional conclusions. 

And thus, just like the state of neural networks in 2010 was only weakly informative about what would be possible in 2020, it seems reasonable to expect the state of things in 2020 will be only weakly informative and about will be possible in 2030. Which is a very different question from how you should try to solve practical problems now.

Comment by vaniver on Alignment as Translation · 2020-03-26T20:20:02.597Z · score: 2 (1 votes) · LW · GW

I think you run into a problem that most animal communication is closer to a library of different sounds, each of which maps to a whole message, than it is something whose content is determined by internal structure, so you don't have the sort of corpus you need for unsupervised learning (while you do have the ability to do supervised learning).

Comment by vaniver on Against Dog Ownership · 2020-03-24T22:11:43.836Z · score: 15 (6 votes) · LW · GW

I came in agreeing with several of the author's conclusions (many 'aesthetic' breeds are animal cruelty, owning a dog as a single person with a full-time job is probably cruel, dogs are a poor substitute for children, etc.), and yet found something about the article highly offputting.

First is that I think "dominance" is the wrong frame, and having the wrong frame often generates lots of "well, I don't disagree with what you're saying here, but somehow I disagree with the whole thing."

I think from the dog-owner's perspective, the right frame is closer to 'being needed.' Think about the greentext about shrimp and this bit of The Bell Curve:

The broadest goal is a society in which people throughout the functional range of intelligence can find, and feel they have found, a valued place for themselves. For “valued place,” we offer a pragmatic definition: You occupy a valued place if other people would miss you if you were gone. The fact that you would be missed means that you were valued. Both the quality and quantity of valued places are important. Most people hope to find a soulmate for life, and that means someone who would “miss you” in the widest and most intense way. The definition captures the reason why children are so important in defining a valued place. But besides the quality of the valuing, quantity too is important. If a single person would miss you and no one else, you have a fragile hold on your place in society, no matter how much that one person cares for you. To have many different people who would miss you, in many different parts of your life and at many levels of intensity, is a hallmark of a person whose place is well and thoroughly valued. One way of thinking about policy options is to ask whether they aid or obstruct this goal of creating valued places.

That said, many of the same complaints apply--why not be needed for something productive, instead of manufacturing something where the need is the feature, instead of the bug? If no one in your life needs you, and you buy/rescue a dog and now one dog needs you, is that an improvement / is that healthy

I think the answer is "yes," and thinking about the word "healthy" clarifies why. Suppose someone is writing about food, and points out the ways in which food grown without pesticides is healthier than food grown with pesticides. If you're worried about second-order effects, of what additional chemicals you're ingesting, this is right; if you're worried about first-order effects, of whether or not you'll be malnourished, this is wrong. (As an important background fact, pesticides increase yields, such that organic farms produce fewer calories per unit of land and effort.)

In general, I try to be allergic to the "everyone should have <luxury version> of <good>" argument, because in fact people are better in tenements than living on the street, and if we had more tenements we might have fewer people living on the street, and so banning tenements is probably harmful. Similarly with minimum wage laws, and so on and so on.

Consider this section: 

What does it say about a human who enjoys this emotional transaction? It says that on some level they like the idea of having dominance over another being. And, they want that dominance to be a feature of their daily life.

I don’t think there is anything wrong with enjoying dominance, per se. Sexual dominance is clearly a popular tendency, and likewise, the desire to dominate others in competitions is a useful inborn characteristic which incentivizes ambitiousness and effort. I think identifying and pursuing both of those forms of dominance can bring pleasure and satisfaction in a healthy way.

That is, the author isn't opposed to dominance, or A being better than B. They just think there are good ways to do it and sad ways to do it, and dog ownership is one of the sad ways. If we analogize to video games, they're claiming that playing competitively is good, and only scrubs play against AI instead of other humans.

There's a part of this that seems right--people who win at competitive video games are better at gaming than people who can't win, and people who win competitions / status games are better at competing than people who can't win those competitions--but also a part that seems mistaken, in that it won't be the case that everyone can be above average, unless you include competitors that are 'outside everyone' while still engaging in the correct way.

And in especially in the context of "minimum wage laws" or "looking down on the worse version of things", it seems especially cruel to cut off opportunities for people who aren't very needed / aren't very respected to get an easy source of need and respect, not because it's harmful but because it reflects poorly on them for being on the bottom of the pyramid. That is, in an ordered system, someone is going to be on the bottom, and we get to decide whether it's people or dogs.

[There's a different argument you can make, where you say the relationship is bad for the other side; I currently think it's the case that humanity has made a pretty good deal with cows from the cow's point of view, for example, but don't think that humanity has made a pretty good deal with chickens from the chicken's point of view. The author considers this argument but only accepts it in a limited way, in approximately the same way I do, but I think the 'family dogs' and 'lapdogs' have way more meaning than the author thinks.]

Comment by vaniver on Against Dog Ownership · 2020-03-24T20:27:57.653Z · score: 3 (2 votes) · LW · GW

I also think that dogs probably have real feelings and don't just act the part like this creepy robot child, although I wonder how can one actually test this.

I mean, this depends on what you mean by "real feelings," but as far as I can tell the physiological cause of emotions is basically shared by all mammals. (If anything, emotions likely play a larger part in the mental processing of non-human animals, because there's less of other deliberative faculties to play against them.)

Comment by vaniver on March Coronavirus Open Thread · 2020-03-19T17:51:25.255Z · score: 5 (3 votes) · LW · GW

But ultimately, nurses/EMTs/medical students can be trained to do all this in a few days. If someone's competent and confident and has adequate back-up in-case of issues.


I think we might want to be in the world where we train a substantial fraction of the recently unemployed, or the National Guard, or whoever to do this, which requires starting from a lower point than nurses/EMTs/medical students.

Comment by vaniver on LessWrong Coronavirus Agenda · 2020-03-18T18:21:27.290Z · score: 4 (2 votes) · LW · GW

Done.

Comment by vaniver on March Coronavirus Open Thread · 2020-03-18T18:12:42.529Z · score: 3 (2 votes) · LW · GW

They assume a fixed ICU capacity (where I think the main limit is ventilators, not just beds). Does anyone have any models/estimates of how much UK/US can expand intensive care capacity (e.g., with "wartime" style all hands on deck manufacturing and innovation)?

My understanding is that the treatment requires significant monitoring and skill; the ventilation is often invasive (they have to get the tube into your lung, rather than just into your mouth).

But for a while people have been suggesting compartmentalizing the medical system further. If you just want someone to be a 'ventilator nurse', able to intubate a patient and then manage a ventilator for that patient, could you do that with a 30-day training program? Seems likely and worthwhile, but will require some sort of emergency legislation to authorize in most places, and some rapid development of curricula and testing.

Similarly, expanding production runs into legal issues. You may have heard about the volunteers who 3D printed ICU valves; they asked the company for blueprints, and the company threatened to sue for the IP violation. You might also have heard about the patent troll who sued the makers of a COVID19 test for infringement; they dropped the case once it was public that the use was a COVID19 test. It seems like a potentially sensible government action here is to nationalize (or otherwise force licensing) of technology that's useful in a disaster, with the government paying for the IP after-the-fact based on actual usage out of the overall disaster response fund.

But in general, our 'peacetime' standards for medical devices are very high. If you want to take your toaster factory (or w/e) and start spitting out ventilators instead, there's a lengthy approval process because this is complicated stuff with many ways things can go wrong. When the alternative is nothing, it's probably good to have rush jobs available, but there's nothing in place (that I'm aware of) to allow this sort of rapid ramping.

Comment by vaniver on March Coronavirus Open Thread · 2020-03-18T17:48:31.182Z · score: 4 (2 votes) · LW · GW

This abstract says "Sauna takers should avoid bathing during acute respiratory infections." I haven't read the paper to figure out why they think that.

Comment by vaniver on Refactoring EMH – Thoughts following the latest market crash · 2020-03-17T03:54:23.878Z · score: 4 (3 votes) · LW · GW

Dalio even thought about COVID significantly before the crash.

I think he thought about the reference class of pandemics, more than he thought about COVID. I think the key details in this becoming as bad as it is are mostly missing from that post.

Comment by vaniver on Coronavirus Justified Practical Advice Summary · 2020-03-17T03:10:15.434Z · score: 10 (5 votes) · LW · GW

Also, "viable virus could be detected" seems potentially different from "you could get infected from handling it" (in either direction). Your immune system is more robust than the cell culture they use to detect, so many 'detectable levels of virus' are still safe, and the detection method requires diluting the contaminated swab, which means that you coming into contact with the undiluted thing might have nonzero risk even when the test can't detect anything.

But based on my limited understanding of what's going on, the threshold is within a doubling or two of what's sensible; not diluting at all should correspond to the 10^0 line on the graph, which their model suggests should be hit within 20-45 (point estimate: 30) hours of when virus is deposited on the box. So unless you're somehow managing to concentrate the virus, that's the point at which it couldn't infect a cell culture without an immune system.

Comment by vaniver on Coronavirus Justified Practical Advice Summary · 2020-03-17T02:54:11.613Z · score: 12 (7 votes) · LW · GW

I'm glad that they ran this experiment, since this is a core uncertainty, and think we should accept that the price of posting drafts is that you have to read them carefully to spot potential errors; having the graph of the raw data a few days earlier was worth having to eyeball the analysis myself to check their numbers.

Comment by vaniver on Comprehensive COVID-19 Disinfection Protocol for Packages and Envelopes · 2020-03-16T17:05:05.739Z · score: 8 (2 votes) · LW · GW

I'm working off this paper, which did test cardboard.

Comment by vaniver on March Coronavirus Open Thread · 2020-03-16T02:12:50.862Z · score: 4 (2 votes) · LW · GW

According to some as summarized by wikipedia, there's not all that much evidence that people who didn't prepare were bitten by it, or that fixing ahead of time was cheaper / better than fix-on-failure.

Comment by vaniver on March Coronavirus Open Thread · 2020-03-16T02:10:31.110Z · score: 25 (7 votes) · LW · GW

From A Technical Explanation of Technical Explanation:

How would I explain the event of my left arm being replaced by a blue tentacle? The answer is that I wouldn’t. It isn’t going to happen.

If a miracle happens, then a miracle happens. I'm not holding my breath.

The ways in which I do expect Vaniver_2021 to look back at Vaniver_2020 and think "yeah, he was worried about that but it didn't turn out to be relevant" are various unknowns about the virus that might be fine or might be bad. For example, we don't know how bad surface transmission will be, but that's a big factor in what sort of isolation protocols you need to have. We don't know whether existing anti-virals will be effective. We don't know how long immunity will last, but that's a big factor in whether or not 'herd immunity' strategies will work, and how valuable it is to not catch it. We don't know how big a deal antibody-dependent enhancement will be, or how that will interact with the duration of immunity. We don't know what long-term effects of infection (think fatigue, disability, infertility, etc.) look like. We don't know how long people are infectious before they show noticeable symptoms.

For all of those things, I put significant probability on the "it's fine" side of the uncertainty. But it being not fine is quite bad compared to it being fine, such that the expected utility shakes out that I should take it seriously until we know more. For example, I now think that if you're taking your temperature every day, the "infectious before noticeable symptoms" window is probably about a day, which seems pretty tolerable, but don't think I made a mistake in my assessment before. If the long-term disability risk turns out to be closer to 1% than 10%, then I'll adjust my prior on long-term disability for next time (in the obvious way that I'll have two datapoints instead of one), but I won't think "oh, I cried wolf."

Comment by vaniver on Comprehensive COVID-19 Disinfection Protocol for Packages and Envelopes · 2020-03-15T17:07:06.214Z · score: 9 (5 votes) · LW · GW

wait a day, and the virus will be dead to an extent that you can't get infected.

The paper that I've seen that tried to estimate this only reported TCID50/mL; how do you convert from that to infection risk?

[I think their methodology also might have been the equivalent of 'licking the box' instead of, say, touching the box with your finger and then touching your lips with your finger and then licking your lips, but for simplicity's sake let's assume I'm licking the box.]

Comment by vaniver on March Coronavirus Open Thread · 2020-03-15T16:58:51.556Z · score: 9 (4 votes) · LW · GW

Is somebody keeping track of the "what if we're wrong and it turns out this is another Y2K" scenario? 

That is, a combination of "prevention work successfully means no big disasters" and "absence of prevention work doesn't cause any major disasters"? I think that cat is already out of the bag on the latter one; people might end up disagreeing on whether it was better to be in Iran or Wuhan, but they won't be able to disagree that the lockdown in Wuhan had an effect.

I think there will be variation in what sorts of social distancing happen, which we should be able to back out data on, and similarly demonstrate that social distancing had an effect. (I expect it'll be smaller than many people hope it'll be, but it'll still be noticeable.) Like, we could see the effect in 1918 influenza data, and we have a much better ability now to track how people come into contact with each other.

[I expect the main thing to happen is that people take insufficient protective measures, which makes them look like a waste, or we get stuff like "ah look, extensive social distancing meant the peak happened two weeks later!", which is of unclear value compared to the costs.]

Comment by vaniver on Should we heat our houses to ~78F to reduce coronavirus risk? · 2020-03-15T16:30:28.593Z · score: 12 (3 votes) · LW · GW

See also discussion of this (and related) papers here.

Comment by vaniver on What is a School? · 2020-03-15T02:36:20.342Z · score: 4 (2 votes) · LW · GW

Online schools have been around for a while, but I think are generally less popular, and the main users are people who live too far from a regular school to think it's worth the trip, or who got bullied too much by other students, or so on.

More broadly, tho, I think the thing schools are selling is a package, and the 'book learning' part of it is something like a third (or less) of the value of that package for the typical student (and parent), but is the main bit that can also be duplicated by online schooling.

Comment by vaniver on A practical out-of-the-box solution to slow down COVID-19: Turn up the heat · 2020-03-15T01:15:16.662Z · score: 4 (2 votes) · LW · GW

Ah, good to know!

Comment by vaniver on A practical out-of-the-box solution to slow down COVID-19: Turn up the heat · 2020-03-14T16:36:08.258Z · score: 2 (1 votes) · LW · GW

Looks like he tested positive and then negative? Hopefully that means he doesn't have it, but I seem to recall these tests can be inaccurate in both ways.

Comment by vaniver on A practical out-of-the-box solution to slow down COVID-19: Turn up the heat · 2020-03-14T16:34:16.587Z · score: 4 (2 votes) · LW · GW

Here's a linear regression using Chinese cities, which also looks some at national data for other countries. They find a significant negative effect on transmission from both temperature and humidity, which look to be of roughly the same strength; however, basically everywhere is still in the "R>1 exponential growth" phase for months. 

Comment by vaniver on March Coronavirus Open Thread · 2020-03-14T00:25:05.554Z · score: 14 (5 votes) · LW · GW

The idea that smart investors don't understand exponential curves is absurd on its face

I don't think this is necessarily absurd or false. Like, this is what Black Swan Farming was about.

I think people in finance are used to exponential curves with doubling times of 20 years, and this doesn't give them much of an edge when it comes to doubling times of 2 days. Like, even in semiconductor manufacturing, the progress of Moore's Law over someone's 40-year career corresponds to about a month of viral growth at that rate. 

Startup finance people do work with stuff at roughly the same scale, and correspondingly freaked out much more.

The sheer insanity of such a prediction should give you an idea of how uncertain this whole thing still is.

I don't think this is crazy, once you consider healthcare system failure. What does the world look like if no one receives medical care for any condition besides a COVID infection for the next three months?

Comment by vaniver on What is a School? · 2020-03-13T22:18:47.995Z · score: 33 (15 votes) · LW · GW

Of course, this may be because they think online schooling or homework will be an adequate substitute for in-person schooling, and assume that by default those will be assigned to make up for lost in-class time.

[I think school is highly suboptimal and quite possibly net negative compared to not having mandatory or subsidized schooling, but I think the inference Zvi is trying to draw here is shaky.]

Comment by vaniver on A practical out-of-the-box solution to slow down COVID-19: Turn up the heat · 2020-03-13T22:05:14.506Z · score: 4 (2 votes) · LW · GW

Here's a province-level infection map of Iran as of the 12th:

src: https://www.nytimes.com/interactive/2020/world/coronavirus-maps-italy-iran-korea.html
Comment by vaniver on What will be the big-picture implications of the coronavirus, assuming it eventually infects >10% of the world? · 2020-03-13T19:51:50.169Z · score: 4 (2 votes) · LW · GW

For many places, like the US, needing approvals is more of a problem than cost, when it comes to building new hospitals or ventilators or so on. It seems quite possible that the regulations will shift as a result of the experience here.

Comment by vaniver on A practical out-of-the-box solution to slow down COVID-19: Turn up the heat · 2020-03-12T20:53:35.613Z · score: 5 (3 votes) · LW · GW

Brazil looks like it's also rising pretty quickly, but is especially interesting because both the president and his communications director seem to have it. [Edit: Probably not?]

Comment by vaniver on Coronavirus: Justified Practical Advice Thread · 2020-03-12T18:35:13.109Z · score: 5 (2 votes) · LW · GW

I've emailed the author; we'll see if she has time to respond (or if the code goes up on Github; I can't find it yet).

Comment by vaniver on A practical out-of-the-box solution to slow down COVID-19: Turn up the heat · 2020-03-12T17:40:43.908Z · score: 13 (6 votes) · LW · GW

Why Iran and Italy and not Thailand and the Philipines?

I think it's too soon to say "not Thailand and the Philippines"; three days ago there were 10 reported cases in the Philippines, then 33, then 49. Italy's trend at a comparable level was steeper--3 then 9 then 76--and then here we are, two and a half weeks later.

As for "why Italy and Iran first", there's both the possibility of "lower spread rate" and the possibility of "weird founder effects." If a careless Italian healthcare worker made a mistake three weeks ago, and a careless Philippine healthcare worker made a mistake one week ago, we'd see two additional weeks of growth in the Italian numbers and not the Philippine numbers.

Comment by vaniver on A practical out-of-the-box solution to slow down COVID-19: Turn up the heat · 2020-03-12T17:20:49.991Z · score: 4 (2 votes) · LW · GW

Do we have an infection map of Iran? I've only been able to find national numbers, and they're consistent with the disease primarily spreading in Tehran. [That said, I also buy that the numbers in Iran might be a 100-1000x underestimate at this point, which might mean this is just telling us about testing instead of infections.]

Comment by vaniver on A practical out-of-the-box solution to slow down COVID-19: Turn up the heat · 2020-03-12T02:27:34.183Z · score: 23 (8 votes) · LW · GW

Now granted there could be a higher spread in these countries that is underreported, but if they had it bad as in Italy or Iran it wouldn't go unnoticed.

I don't buy it, with the current level of undertesting. The rhetorical work is being done by the red and dark red areas not reporting many cases, but it seems premature to be sure of that, since many of those are places that don't have the medical infrastructure to do widespread testing. It looks like Africa has done 400 tests; some people suspect that's enough for the counts to be accurate, but I wouldn't bank on it yet.

It's not obvious we would know about Seattle if some heroes hadn't tested people for coronavirus despite being told not to, and so it's not clear the higher number of cases reported in Washington state than New York or California is a true difference or testing artifact. Also, Spain (yellow to orange on the map) has more confirmed cases than the entirety of the US.

I can't find data for Shiraz (a hot big city in Iran), and amusingly looking for that data took me to this paper (since one of the authors is affiliated with a university there), but that seems like one of the big tests of this theory.

I also think we have more reason to believe humidity matters than temperature, and wonder if adding that to this map makes this look more or less accurate.

Comment by vaniver on Zoom In: An Introduction to Circuits · 2020-03-11T17:43:48.933Z · score: 8 (4 votes) · LW · GW

I think also Conway's game of life has a large bestiary of 'stable patterns' that you could figure out and then dramatically increase your ability to predict things.

Comment by vaniver on Why don't singularitarians bet on the creation of AGI by buying stocks? · 2020-03-11T17:40:17.658Z · score: 11 (7 votes) · LW · GW

buying stocks at random should perform roughly as well as the market as a whole

Except you have higher variance, which is normally something to avoid. Also note that average mutual fund performance is worse than index funds, even without taking fees into account; a similar story is likely true for individual investors.

But I do buy that being a reader of lesswrong dot com gives you inside information about particular sorts of things (of which covid19 was one), and I think betting big on such opportunities makes sense. I think it's not obvious that this gives you much ability to time the market or make good short-term bets, because it relies on understanding how the market reacts to things. 

That is, suppose you have perfect accuracy on how bad covid19 will be; that doesn't actually tell you all that much about how much the market will drop when. So you might want to sell put options at a wide range of maturities instead of doing so for a single maturity date.

So similarly, if there's a company that has many irons in the fire and you think one particular bet will or won't pay off, you're still exposing yourself to all the other bets the company is making, which will increase the variance of your bet a lot. (I, for example, was long Netflix in 2011 on their DVD business and wanted to be short on their streaming business, since I thought the content owners would renegotiate for a larger share of the pie than the market expected they would, but couldn't easily figure out how the market was pricing the DVD business and the streaming business, and so wasn't sure what to do; as it happened, the market thought something like 90% of the value of Netflix was the streaming business, and so I should have been short the stock as a whole.)

Comment by vaniver on Coronavirus: Justified Practical Advice Thread · 2020-03-11T16:49:51.302Z · score: 19 (7 votes) · LW · GW

This study describes "detecting viable virus" as having a threshold of 10^0.5 TCID50/mL, and they assume exponential decay of viable virus particles. 

I'm really confused by their numbers, tho; it looks like cardboard has a hundred-fold reduction in 23 hours, from 10^2.5 to their detection threshold of 10^0.5, which I can't square with the 8.5 hour half-life. [Edit: it looks like I'm potentially confused about what TCID50/mL means?]

I also don't know how to compare their detection threshold with the point at which I should be willing to handle a cardboard box (with varying levels of cleaning and PPE). Is their test basically as sensitive as my immune system (in that I shouldn't handle something where they could see a viable virus, and can handle something where they can't)? Or should I be letting boxes sit for 3 days?

Comment by vaniver on Zoom In: An Introduction to Circuits · 2020-03-10T22:23:34.493Z · score: 8 (4 votes) · LW · GW

I think for the remaining 5% to be hiding really big important stuff like the presence of optimization (which is to say, mesa-optimization) or deceptive cognition, it has to be the case that there was adversarial obfuscation (e.g. gradient hacking). Of course, I'm only hypothesizing here, but it seems quite unlikely for that sort of stuff to just be randomly obfuscated.

I read Adversarial Examples are Features Not Bugs as suggesting that this sort of thing happens by default, and the main question is "sure, some of it happens by default, but can really big stuff happen by default?". But if you imagine a LSTM implementing a finite state machine, or something, it seems quite possible to me that it will mostly be hard to unravel instead of easy to unravel, while still being a relevant part of the computation.

Comment by vaniver on Thoughts on LessWrong's Infohazard Policies · 2020-03-09T22:55:55.819Z · score: 2 (1 votes) · LW · GW

See also Bioinfohazards, which steps through infohazards from the biorisk lens. It seems to have missed overall the category of "public advice," and is mostly concerned with the question of what gets unleashed, as opposed to how to handle scenarios where something has been unleashed.

That is, one can both popularize advice that is misleading or harmful ("having sex with a virgin will cure you of HIV") or one can fail to popularize advice that is informative or helpful. The mixed case, where the public health authorities have decided it's worth lying to the public ("masks won't help" as opposed to "please reserve masks for medical professionals who need them more"), is also perhaps worth consideration.

Comment by vaniver on Thoughts on LessWrong's Infohazard Policies · 2020-03-09T22:45:19.203Z · score: 2 (1 votes) · LW · GW

on LessWrong people should be obligated to ensure the criticism is true before saying it

I understood Davidmanheim to be arguing that it was true, but that they shouldn't have said it; that is, the standard is it being positive EV, not factually accurate. [Or, at least, that the truth wasn't the crux, and the EV was the crux.]

Comment by vaniver on Coronavirus: Justified Practical Advice Thread · 2020-02-28T16:12:22.704Z · score: 47 (21 votes) · LW · GW

Given the below death rates

Note that while your personal survival is quite important, getting infected and surviving can have quite awful effects. We don't know what the long-term effects are like yet (because we haven't hit the long-term yet), but I won't be surprised if post-viral fatigue is common.

Being ill is also unpleasant, and you become a risk to your community, and especially any elderly people in your community.

Comment by vaniver on What will be the big-picture implications of the coronavirus, assuming it eventually infects >10% of the world? · 2020-02-28T16:08:46.101Z · score: 4 (2 votes) · LW · GW

Depends on if immunity is long-lasting

I've seen speculation that COVID-19 does not confer meaningful immunity, and we've already seen a reinfection. If true, this looks less like "we have a new winter disease" and more like "this is the new normal."

Comment by vaniver on Tessellating Hills: a toy model for demons in imperfect search · 2020-02-21T00:44:34.660Z · score: 3 (1 votes) · LW · GW

Ah, that'd do it too.

Comment by vaniver on On unfixably unsafe AGI architectures · 2020-02-20T23:00:21.167Z · score: 3 (1 votes) · LW · GW

MIRI's hiring; maybe you should email Buck.

Comment by vaniver on Tessellating Hills: a toy model for demons in imperfect search · 2020-02-20T22:20:33.516Z · score: 3 (1 votes) · LW · GW

Finally, this demon becomes so strong that the search gets stuck in a local valley and further progress stops.

I don't see why the gradient with respect to x0 ever changes, and so am confused about why it would ever stop increasing in the x0 direction. Does this have to do with using a fixed step size instead of learning rate? 

[Edit: my current thought is that it looks like there's periodic oscillation in the 3rd phase, which is probably an important part of the story; the gradient is mostly about how to point at the center of that well, which means it orbits that center, and x0 progress grinds to a crawl because it's a small fraction of the overall gradient, whereas it would continue at a regular pace if it were a constant learning rate instead, I think.]

Also, did you use any regularization? [Edit: if so, the decrease in response to x0 might actually be present in a one-dimensional version of this, suggesting it's a very different story.]