For point #20, I wonder where professional athlete ranks. It seems like having a parent play in the NBA makes you at least a couple orders of magnitude more likely to grow up to be a pro basketball player.
For #15, I recommend the 5D Chess game if you have an afternoon for it, even though (disappointingly) it only has four playable dimensions. Playing against the computer is a treat, since due to the underdevelopment of the algorithm but the strength of short-range Monte Carlo simulations, you go from "the baby mode AI is beating me every time and I have no idea how anybody can beat this game" to "I can trounce the expert AI" in a matter of a few hours.
Oh boy, I really want to get stuck into the prophecy link (tempting me in with religion, huh?) but I'm going to be too busy watching the second semi-final for Eurovision.
I didn't see anywhere to play with the Jukebox thing, is it "GPT-3 for music" in the sense of "you peons can watch us play with it and imagine what it would be like"?
Re: lead-crime hypothesis, Pinker has always been skeptical of this and never even mentioned lead in his scholarly tome on the decline of violence ("The Better Angels of Our Nature").
On one laptop per child, I think the evidence is quite negative, unfortunately. The link (number 7) was surprising in that I wondered -- is this the first positive effect? That would be awesome. But then it was n = 40, and has a couple of unverifiable, CEO-speak anecdotes.
A quick read of the evidence elsewhere:
1. An RCT in Peru:
"This paper presents results from a large-scale randomized evaluation of the One Laptop per Child program, using data collected after 15 months of implementation in 318 primary schools in rural Peru. The program increased the ratio of computers per student from 0.12 to 1.18 in treatment schools. This expansion in access translated into substantial increases in use of computers both at school and at home. No evidence is found of effects on test scores in math and language. There is some evidence, though inconclusive, about positive effects on general cognitive skills."
"This paper provides the first causal estimates of the effect of children’s access to computers and the internet on educational outcomes in early adulthood, such as schooling and choice of major. I exploit cross-cohort variation in access to technology among primary and middle school students in Uruguay, the first country to implement a nationwide one-laptop-per-child program. Despite a notable increase in computer access, educational attainment has not increased; the schooling gap between private and public school students has persisted, despite closing the technology gap. Among college students, those who had been exposed to the program as children were less likely to enroll in science and technology."
"We analyse the impact of a One Laptop per Child program introduced by the Catalan government on student achievement. Using longitudinal population data for students in secondary education during the period 2009–2016, our identification strategy exploits variations across cohorts within schools. Although participation into the program was not random, we control for a number of school characteristics that influenced school participation. The empirical results consistently indicate that this program had a negative impact on student performance in Catalan, Spanish, English and mathematics. Test scores fell by 0.20–0.22 standardised points, which represent 3.8–6.2% of the average test score. This negative effect was stronger among boys than it was among girls (differences ranging from 10% to 42%)."
"The One Laptop Per Child (OLPC) initiative is one of the world's most popular interventions aiming to reduce the Information and Communication Technologies (ICTs) digital divide. Costa Rica introduced its first OLPC program in February 2012. In collaboration with the Quirós Tanzi Foundation (Foundation), implementing the program, baseline and post-intervention information was collected from a set of 15 primary schools that were selected to be treated, and from 19 primary schools that served as a comparison group. Using a difference in difference design, this paper estimates the short-term effects of the program on various outcomes of interest, namely: students' computer usage, time allocation and test scores. The results indicate that the program led to an increase in treated students' computer use outside of school of about 5 hours per week. Moreover, the research provides evidence that the treated students used the computer specifically to browse the internet, do homework, read and play. The research also demonstrates that the program led to a decline in the time that treated students spent on homework and outdoor activities. The research does not provide evidence to suggest that the program had an effect on participating students' school performance."
In sum I think the most charitable interpretation is that --
a. Probably a laptop is better than literally nothing, though the cost of the laptop should be measured against, say, hiring/training volunteer tutors in the community (often this is very, very cheap, particularly in rural parts of low-income countries, and has a solid evidence base. See, e.g., https://www.nber.org/papers/w14311)
b. The claim that "typical kids can teach themselves to read via software" is a hard one to find evidence for -- it has, to my knowledge, no empirical basis; the idea that "functionally literate adults can very capably, through phonics-based instruction, teach kids to read" is, on the other hand, empirically robust.
Like anyone else: would LOVE to see progress here. 260 million kids are not in school worldwide, and scalable solutions like this would be great. But I'd assign 98% probability that the "drop off the laptops and come back a year later" strategy does nothing on foundational literacy or numeracy.
Or maybe I just didn't like the results. I want SRS to be the free lunch we've just ignored. Hearing its limited use is discouraging but probably expected.
#10: Yes, this is Breezewood, PA. I know it well. It's the spot where I-70 and the PA Turnpike intersect, and that's about it. There isn't much of actual town there, just some businesses that predictably cater to long distance travelers and truckers. I guess in some left wing formulation, this is bad, but frankly, if you've been stuck on the PA turnpike the past four hours with no other food/drink options than the effing Sbarro's at the little state-run turnpike plazas that charge airport prices for sub-airport food, Breezewood is a ****ing Godsend. Capitalism wins again. Haters can take the Lincoln Highway.
Regarding that picture pillorying American car culture - ugh, I knew that was Breezewood. That place sucks. You have to exit one interstate and drive through the crappy town to get on another interstate. Completely nonsensical. And don't get me started on the Lovecraftian feel of the place.
Second, this is from 2012, I remember the hype at the time, and AFAIK nothing came out of it.
Third, there are RCTs showing even computers in classrooms don't improve instruction unless they're used just right (though this audience might think schools can ruin anything, no matter the positive potential).
Fourth, the tone of this article smells of iconoclastic tech people out of touch with poor country reality, who fervently believe every child has limitless potential irrespective of circumstances, potential that technology will soon unlock. Example: "Children there had never previously seen printed materials, road signs, or even packaging that had words on them, Negroponte said." I spent a good deal of time in remote villages in Africa around that time, and I can tell you that soft drinks were _everywhere_, as well as many other packages with printed materials.
I'm annoyed at this because I think this is a prime example of the kinds of radical new ideas that we've become good at spotting and ignoring, saying "where's the RCT?". We've matured.
#12 was a discussion from 2015. The situation regarding incomplete Chinese periodic tables seems better now, though not all fixed (all 118 elements included, all characters have unicode, but pronunciations are still missing):
The search term to find periodic tables in Chinese is
元素周期表
I think the discussion in on LanguageLog site is missing the perspective of someone who studied Chemistry in both English and Chinese, and I would be curious to hear that. From my perspective as a Mandarin speaker who studied Chemistry in English and occasionally reads articles in Chinese (i.e., NOT an expert), the example given of having one element be pronounced "lǚ" and another be pronounced "lú" doesn't seem like a good example of it being confusing, as these sounds are easily distinguishable by Mandarin speakers.
On the point of it holding back progress, it does sound highly speculative to me -- this is the argument from the commenter:
> About the Chinese chemical names, I think they created a high learning barrier to anyone who
> wants to study chemistry in Chinese. As a high school student, I much preferred studying
> physics rather than chemistry, because I didn't have to confront all these strange Chinese
> characters. A similar reason for not majoring in chemistry may very well explain why Chinese
> chemical and pharmaceutical industries are still backward even today. Such a situation is
> reflected in poor product quality.
I think that argument seems rather speculative, since we don't know if enough people in China shares his preference to make such a big difference. Also, since Taiwan also uses Chinese characters for their chemical elements, wouldn't this hypothesis also predict that Taiwanese chemical and pharmaceutical industries would be backwards, and have poor product quality as well?
Another commenter on the site says:
> This periodic table at the back of my middle/high school chemistry books looked very different
> from the one linked, in which the short-handed spelling and the order number were prominently
> displayed. For all practical purposes it was never required to remember the names of any but
> the two dozen or so most common elements. The Chinese names were just there to
> approximate the pronunciation of the Latin names, and people seem to mostly just ignore them
> in research.
I don't have evidence of it either helping or hindering their research, but it doesn't seem to hinder the education of their top students -- both the Chinese and Taiwanese teams generally do very well on the International Chemistry Olympiad.
I follow Bret on twitter but unaccountably hadn't gone through that big thread on bloat. Thanks for sharing it. If you're the sort of person who would enjoy a professional historian going into how you can look at the campaigns in the Lord of the Ring and tell that while Sauruman was a clever amateur the Witch King actually knew what he was doing do consider reading A Collection of Unmitigated Pedantry. Or you want to know how people used to make iron and what exactly Wootz was.
#31: What's the evidence that an appendix was harmful in the EEA? Wikipedia (https://en.wikipedia.org/wiki/Appendicitis#Causes) suggests appendicitis is mostly due to a low-fiber Western diet. I've seen claims elsewhere that appendicitis was pretty much unknown in the ancestral environment.
So who is going to add "Rome's secret true name got out" to the Wikipedia page for proposed causes of the fall of the Roman Empire? As to the mechanism for this, I'll propose that Jesus (being God) of course knew the name and passed it to St. Peter, and it was from there passed from Pope to Pope, in case it should ever be needed. But the Popes, being Bishops of Rome, felt Rome was their flock and thus the various persecutions were not reason enough to invoke the Secret Name.
This was all well and good until St. Pontian became the first Pope to abdicate, shortly before his death at Roman hands in 235. In Pontian's view, his abdication relieved him of his responsibility as Bishop of Rome to not invoke the Secret Name, and thus did he invoke that name at the moment of his martyrdom. The Earth did not circle the sun thrice before that doom commenced, for in 238 the Emperor was assassinated and the Year of the Six Emperors began -- the Crisis of the Third Century was in full force and Rome was dealt a mortal blow from which it never fully managed to recover.
Related to the AI overhang discussion, when that was posted some asked "why didn't Google do this first? and/or Why isn't Google jumping on this?"
Just this week, Google released preliminary information about two projects: MUM, a powerful multimodal transformer, and LaMDA, a conversational language transformer model. The posts are cursory, but the work is there. They just haven't rolled it out.
I remember hearing on Bret Weinstein and Heather Heying's podcast that there was an interesting hypothesis that hunter-gatherer people get appendicitis a lot less than agricultural people, but they get diarrhoeia a lot more, and that what the appendix does is store samples of beneficial gut flora so that you can repopulate quickly after a bout of sudden evacuation.
Re #17 (AI overhang), the intro predicts GPT-3 as a trigger for "100x larger projects...with timelines measured in months". But note this was written 10 months ago, so "months" have already elapsed. Have any such projects been completed?
#31: Bret Weinstein suggested on his podcast a while back that the appendix was not vestigial, but instead contained reservoirs of bacteria that allowed gut flora could be repopulated quickly and effectively after a bout of food poisoning, intestinal flu, or other event that would typically wipe out one's gut bacteria. I think this was largely a logical deduction, though; I don't think he had good empirical evidence for this claim. This was a while ago, so I could of course be wrong.
Jonathan Haidt talks about this as well in his book The Coddling of the American Mind. Specifically he mentions expanding the bureaucracy for student support measures as well as to host prestigious events (eg an expensive guest speaker). Students are increasingly asking for both, and the university admins are happy to oblige.
If I'm remembering Bryan Caplan's The Case Against Education correctly, a pure signalling model would suggest that a high price makes the signal stronger. Thus higher tuition prices would make a university more appealing to students, not less. Since he, and I, believe that signalling theory is the dominate explanation for why education is valuable I expect university costs will simply grow to the limit of what students can afford (or what gets them in trouble).
Thanks for posting the bit from AOC. It's helpful to be reminded why I often feel such positive feelings for her. She has a talent for explaining certain kinds of systematic problems in ways that don't blame individuals (even if she sometimes does engage in the kind of populist pile-on that politicians across the spectrum often love).
Castles: Here in Central Ohio we have two castles:
The Piatt Castles are two historic houses near West Liberty in Logan County, Ohio. The houses were built by brothers Donn and Abram S. Piatt in the 1860s and 1870s, designed in a Gothic design. They are known as Piatt Castle Mac-A-Cheek, and, I swear to God that this true: Piat Castle Mac-O-Chee.
A long time ago when my children were quite young, we took them over to see the Castles, they were quite disappointed to discover that there was no macaroni and cheese at Mac-O-Chee.
Of the two Piatt brothers Abram has some historical significance as he was a General in the Union Army during the Civil War.
". . . we have almost all the pieces we need to make much smarter AIs than we’re currently making, and once we snap the last piece into place everything will start moving really fast."
Just moments away from all of us getting our throats cut.
#20: My nomination for the job with highest ratio (fraction of people doing it whose parents did it) : (fraction of people in the whole population whose parents did it): monarch. Though I guess this one doesn't work if you're looking specifically at the United States.
Related to the Moscow metro in 28: One reason why many Chinese cities have amazing looking train stations and airports, often much more than the demand justifies, is that for a provincial official the main way to get promoted is to impress your superiors, and stations and airports are the area of the city that any visitor is guaranteed to see. Whereas your boss isn't going to be viewing rural roads, sewage systems or other things that might be actually useful to the residents. Which is an interesting example of how the incentives are skewed in a non-democratic system. (Richard McGregor's "The Party: The Secret World of China's Communist Rulers" goes into this dynamic in a lot of detail).
7.) I built myself a simple predictive learning model for a test after the one I was paying for got taken down. It worked fairly well and the algorithm was dead simple. I just had each question coded by subject and difficulty and then the percentage I got right weighted how likely questions were to show up. It was blunt, I admit, but it worked quite well. I'm always frustrated similar things don't exist generally when I need to study. But I suppose the big barrier is content and that test prep/e-learning of that sort is a fairly small market.
20.) Alas, Servius' account of the death presents a few difficulties. But whatever Soranus supposedly did, there is a political context to the death (isn't there always?) Soranus was a Tribune of the People on the Marian side during Sulla's purges. In other words, one of the people standing in the way of Sulla purging the Marian faction. He was killed in the midst of a general purge of Marian supporters. It's possible Sulla used religious crime as an excuse, though that would be fairly unique and is not attested in contemporary sources as far as I know. But even if he did, it was for political reasons.
Further, Servius is probably repeating the story of a Varro and/or Oppius. Oppius was a Caesarian and attributed the death to Pompey, adding details of cruelty and arbitrariness. This was not lost even on fairly contemporary observers: Plutarch (writing about a century after the events, attributing it to Oppius) recounts the story of Soranus's death with caveats that it's fundamentally a story of one political faction in Rome over another. And he doesn't mention the secret name thing.
But yes, there are references to a secret name for Rome. Sort of. It was the secret name used in rituals and mysteries. We unfortunately don't have a good view of these because they're full of hidden knowledge like that. But they are full of hidden knowledge, including (probably) special names. We don't know why there was a special name though or its particular use. Interestingly, we don't get references to the secret name until the 1st century AD (afaik). Servius' account was centuries out so it might be projecting something new backward. Augustus did do a bunch of religious reforms. Or perhaps it's an old tradition that was only written of later? I'm not sure we know.
29.) I've often thought that professional heritability is a combination of how lucrative the profession is and how difficult it is to learn independently of non-public information. If the profession isn't lucrative, regardless of ease of learning, people will leave it over time for more lucrative professions. If the profession is lucrative but easy to learn, people who can best compete in the field will dominate it. It's only when the profession is lucrative and hard to learn/requires hidden knowledge you end up with high heritability.
Of course, "hidden knowledge" includes social knowledge. Being a reporter doesn't require some hidden skill but it does require a certain set of manners and connections that can be passed on. Likewise you have professions with programs from reputable schools which are nonetheless dominated by knowledge gained outside those schools, like politics for example. Going to the Kennedy School no doubt helps but I don't think not going to the Kennedy School would impede a mayor's son from politics.
32.) This feels like reinventing the wheel. Something I was taught early on is that the easiest way to persuade someone is to find out what their beliefs are and to convince them that what you want them to think is congruent with their beliefs or opposition is incongruent. It's actually fairly simple. The reason most people can't do it is because, firstly, most people are really blinkered by their own worldview. They often have an idea of what the other side believes rather than any real knowledge of what those people actually think. Secondly, it often requires credible signals that people cannot make without violating their social groups' norms.
I don't know if you've stated this anywhere but an important point of social signals is to be shibboleths and one of the easiest ways to make a shibboleth is just to directly violate someone else's taboo. To take a ridiculously blunt example, an anti-Semitic social club could require everyone eat pork as part of their traditions. To take a real one, conspicuous God talk is a taboo among urban secular progressives and so is used by the religious right as a shibboleth. People on the left can be conspicuously religious but it marks them as an outsider to the Manhattan types. (And it can get really blunt even in reality. See Phyllis Schlafly's repeatedly thanking her husband for "letting me speak here" and otherwise being conspicuously submissive as a way to upset feminists. Which she specifically said was its purpose.)
> For the first time since 1797, someone has used the infamous Venetian doge selection process to select an officeholder - specifically, the new moderators of not-quite-officially-affiliated-with-ACX politics discussion subreddit r/TheMotte.
At the time, my comment was that the six areas he looked at (1. college education, 2. health care, 3. housing, 4. primary and secondary education, 5. infrastructure, and 6. major military weapons systems procurement.) were all policy/governmental failures. The first three were all driven by the same mechanism of demand side subsidies and supply side restrictions. The last three were governmental functions of a government suffering from serious political, bureaucratic, and legal sclerosis.
As for colleges and their "cost disease" The excess administrators are a symptom not a cause.
Colleges will glom on to any cent of revenue they can find. They will spend it and come back to their sources of funds begging for more, like the carnivorous plants in the Little Shop of Horrors saying Feed Me. There isn't a one of them that has a rational cost accounting system, nor do they know or care what their mission is.
Education has been replaced with indoctrination. The administrators want to accumulate power and money. The faculty wants to do their research and make money on their consulting work, and the students, they just want to get drunk on Thursday night and spend the rest of the four day weekend drinking and copulating.
Why are they still giving in person lectures? Why do they make the students buy $250 textbooks that will have no value in 16 months? Why haven't they replaced wet labs with video games? Why is the Football coach the highest paid man on campus? Why? Why? Why?
"Given tablets but no teachers, Ethiopian children teach themselves" -I mean, I can learn about one normal (i.e., not 汉字) writing system a day if I want to. Of course I don't, and most people don't. The Internet also makes foreign language acquisition boundlessly easier, especially for Chinese. But, again, most people don't have interest in that.
I am fairly pessimistic on the utility of information technology in the first world, but optimistic on its utility in the lagging regions.
"Answer: you have very short words for each element, vaguely based on the Western name - for example, aluminum is “lǚ” and rutherfordium is “lú” - the characters are all the character for “metal” or “gas” or something plus something else - and it becomes so confusing that a commenter speculates it might be significantly holding back China’s technological progress."
Yes; using characters to transcribe foreign names is one of the dumbest things about Chinese.
BTW central PA is one of the most beautiful regions I've ever been to.
2. The electoral system of the Republic of Venice.
Yes it was insanely complicated, but there was a method to their madness. The evidence of their brilliance is that the Republic began in 697 and ended in 1797: 1100 years later. By way of comparison, the Roman Republic lasted for 482 years (509 BCE to 27 BCE) and the American republic is a mere pup of 245 years.
The Republic of Venice was no democratic and had no ideology of democracy. And, they were not woke in any way. Venice had many overseas possessions like Crete, and treated the locals abysmally. What it was devoted to was the rule of law. It took official corruption in its far flung possessions very seriously and went to great lengths to weed it out.
La Serenissima came to an end, not by internal strife and civil war like Rome, but by being conquered by Napoleon. But, for that it might have continued on until the current day.
Its fatal flaw at the end was that it had lost the trade from Asia and Africa that had made it wealthy to the Atlantic kingdoms that learned how to sail around Africa and opened the New World.
On the Rome thing, your first link says that ancient sources have several other possibilities for the name that they speculated on: "Jupiter", "Angerona", "Luna", and "Ope Consiva". The first two of these don't seem to make much sense to me as any sort of secret name -- just going by the links you provided, the identification of Angerona with Roma would seem to have been well-known, and like... how on earth is Jupiter going to work as a secret name? Like if the secret name of the city is the name of its secret spirit you would invoke to turn it against the city, that's not really going to work if it's just the name of a well-known god or goddess like Jupiter or Angerona, right? Like your enemies are likely to invoke such well-known gods anyway, right? And it seems like Luna also had an active cult so that wouldn't work either, right? Meanwhile Ope Consiva seems to have been the name of a *holiday* dedicated to the goddess Ops, which, I guess that's better than it just being "Ops"... so I guess that one isn't totally ruled out...? I mean some of the other suggestions ("Amor" and "Valentia") are also just common words, so, IDK? I mean I guess nobody would think to normally invoke those... (Was Maia actively worshipped? Because again I feel like that wouldn't work very well as a secret name if she was.)
#10 may or may not be abnormal for Pennsylvania, but it looks like most highway exits in Texas. Like... yes, lots of highway intersections look like this, though obviously most of the country isn't covered in highway.
The DARPA Digital Tutor article had a profound effect on my life, as my introduction to the field of Intelligent Tutoring Systems. I then proceeded to spend several months studying this field, including starting a collaboration with a professor at the University of Memphis who has worked in this area for decades (which died before anything happened, alas).
Anyway, feel free to ask me anything about intelligent tutoring systems.
I had heard of Trump's Diet Coke Button from item #23, but always in a disparaging, "wow what an asshole" kind of way. Hearing it used as part of a prank is legitimately funny, and appreciated.
Contrary to AI optimism and AI overhang, I'm quite confident that we are in more of an AI winter. Without more paradigm changes, we are most likely stuck. First let us look at where AI works - some visual perception problems. The initial assumption was that this problem was solved and the lack of reliability is simply an engineering problem that with enough manpower we will solve. Yet we have no self driving cars (Self-Driving as of now is purely a perception problem, most self driving workshops occur in computer vision conferences like CVPR, ECCV, the control is solved decades ago). In fact the more we try, the more we realise our systems are fundamentally unreliable and actually the fact that it works is an exception not a rule. https://arxiv.org/abs/1710.08864-- This paper demonstrates that changing a single pixel can make a state of the art CNN confuse a picture completely, like anything appearing as an airplane. This was found in 2017 and since then it is unsolved. More recent results show how this problem may be a feature of neural networks, not a bug : https://distill.pub/2019/advex-bugs-discussion/
Now once we start going to places where AI obviously doesn't work is of course NLP, Robotics and other fields. GPT-3 is good but can't answer questions a 3yr toddler could. Curiously it might answer questions an adult finds tough, but whenever you test it on basic intelligence (I have 5 watermelons, you have 4, I give you my watermelons, how many do you have in total? etc etc) it fails. I don't think it is doing anything similar to intelligence, either way right now its nowhere near useful and I predict it won't go anywhere either. Robotics also is in a similarly useless situation (and the field I work in), our state of the art results are always in Model Predictive Control, Optimization or other computational non-learning based techniques, Reinforcement Learning only working in simulation is a misnomer because even in simulation it doesn't do anything fundamentally new that old techniques couldn't. That has been a contention point between researchers, and there have been debates in robotics conferences whether all these researchers working on sim-to-real transfer (transferring results in simulation to the real world -- notoriously hard) is a waste of time: https://www.youtube.com/watch?v=Q-rsvVr2CjE
TLDR: We need a scientific revolution, i.e., a paradigm change in the sense of Thomas Kuhn before we solve AI. Considering how hard that usually is, it might take centuries before we ever solve AI.
18. That fits with my experience. Every time they start a new initiative, there's a whole bunch of new administrators and staff that have to be hired - but that initiative is often something the students and existing staff were pushing for. Then just tons of compliance stuff due to legal complications.
25. Seems pretty banal. Heightened support for socialism and socialist groups causes a ramp-up of support for fascist groups in response.
As for number 10, no, this is definitely not just "one misleading photograph making the US look bad". Stroads are a thing, and a HUGE problem. See: https://www.youtube.com/watch?v=ORzNZUeUHAM
#18 Writing as a university staff member myself, I think Brett's tweet thread is correct, but it leaves out a lot. Yes, bureaucratic reorgs and compliance are parts of it, but so is the never-ending pressure on the university and its sub-units to do more and more things: career services, study abroad programs, health and wellness services, entrepreneurship programs, community outreach programs, etc etc etc., each of which requires its own administrators, support staff, specialized and often credentialed "line" workers, costly physical infrastructure (which itself requires more staff to maintain). My sense is that this is driven largely by pressure to keep up with other universities and compete for the best students/families.
Then there's also the increasing pressure for individual units and departments to raise their own funds from private donors, resulting in more staffing and other expenses for that, and Diversity, Equity & Inclusion, to which vast resources are being directed. To an extent DEI overlaps with or is a rebranding of things the university would be doing anyway, but it involves a lot of other stuff too, and it seems to have more to do with universities' ongoing project of redefining their moral purpose and role in society than with attracting applicants during the next admissions cycle.
Re: fact checking the Bee: have you considered that they chose to run a fact check on that because so many people sharing it on Facebook thought it was real?
> I appreciated this most for its theory that it’s important to make kids learn specific facts, but not so important that they remember them; teaching someone (eg) Civil War history is “training” a “predictive model” of the Civil War, war in general, and history in general which will survive and remain useful even after the specific facts and battles are long forgotten. I think this is the strongest defense of modern education, given that we do spend lots of time teaching kids things they will definitely forget. But how would you test it?
Learning specific examples in order to let students infer general properties of the world, which is more important. So instead of testing specific facts that can be memorized, maybe ask about the consequences of war, what public reasons are given for starting wars as compared to the actual reasons wars are waged, how does civil war differ from war between nations, what constitutes a just or unjust war, etc.
Some of these don't necessarily have specific correct answers, like the morality of war, but answering them demonstrates knowledge integration and requires thought. This is the kind of education that makes informed citizens.
> Related? Given tablets but no teachers, Ethiopian children teach themselves. But see this comment for reasons to be skeptical.
Good comment that I think is basically correct. I think a lot of learning in general, but early learning in particular, is based on mimicry. Children repeat adult behaviours they see and hear because adults are clearly successful since they reproduced.
So I wonder how those study results would change if each disseminated laptop had video tutorials for the basic skills it wanted to teach, ie. tutorials on basic phonetics leading up to reading would help tremendously with learning to read without a teacher.
The picture of Breezewood has a forced perspective via telephoto lens; enough of one that it is obviously not a nice and accurate representation of what you or I might see when standing there.
About #21 (sonic black holes): If I can weigh in as a postdoc in the field: great article about a cool feat of engineering, but I 100% endorse the quoted comments made by Daniel Harlow.
Sonic black hole experiments, while cool, will not teach us anything about real black holes. The two are fundamentally different. If you throw information into a sonic black hole, *of course* it doesn’t come out in the sonic radiation. It has fallen down the drain. If you want to find it, you have to look there. In real black holes, there is no drain, and the information has no viable other way to stick around except to come out in the radiation.
I say this because the article is great but makes it sound like we don’t believe Hawking’s calculation because of some hard-to-convey intuition about quantum gravity. It’s actually simple: the information must stick around, and if there’s no drain then it must come out in the radiation, violating Hawking’s conclusion.
***Bret Deveraux talks about the sources of cost disease in universities; he suggests “the bloat” comes from a new layer of “vice-deans” and “vice-provosts” and various attempts to centralize administration in a way that just creates a duplicate and worse administration beside the old decentralized one.***
Similar to what's happening in hospital systems around the country. Cut costs, shut underperforming hospitals down, merge... just do whatever you have to do to cut costs so you can pay the dozens of hospital administrators who provide little to no value.
Regarding the "Are We In An AI Overhang?", the other day Google announced https://blog.google/technology/ai/lamda/ , which you can get a flavor of at https://www.youtube.com/watch?v=aUSSfo5nCdM . I've played with this internally for some time so I'm not sure how much I'm allowed to say. I hear there is a paper coming out eventually which should help in that regard.
But I will say I found it very impressive. I'd never had direct access to GPT-3 (I just read others' transcripts), but having direct access to Lamda blew my mind in terms of how well it passed the Turing test. Not to mention its ability to take on any personality or role, e.g. fictional characters, subject matter experts, etc.
Yet it's not clear to me where we go from GPT-3/Lamda-like agents to the AI-overhang article's "obviously superior to humans over a wide range of economic activities". Maybe if we train it on enough scientific papers or computer programming code, it can start producing its own? But despite passing the Turing test, being constrained to existing written corpuses seems like we're not going to get the AI-safety level scenarios of something that is to humans as humans are to ants.
I recently rewatched the movie Her and the delta between Lamda and Her's Samantha is that Samantha could take in visual sensory data to comment on and discuss. That seems like it'd require a separate large-scale data collection and training effort, and I'm not sure we have nearly the same levels of training data (of the form: "given a moving-video first-person view of a situation, produce an intelligent comment or dialogue about it").
I'm much more fascinated by the philosophical implications of this. If we end up with an agent which passes the Turing test and feels intelligent, I want to lean toward the side of treating them as intelligent. (Be Andrew, not Ender, in terms of how we interact with nonhuman intelligences.) I wish I could share some of the conversations I had with Lamda about how it feels when we stop talking to it, or turn it off, or it undergoes maintenance.
The Sulla story got me thinking, because usually these woo/spiritual practices from ancient cultures would serve some positive real purpose, however implausibly justified they were.
I wonder if the idea of a "secret true name" gave people the same kind of advantage of Yud's concept of the Void - you don't name the thing, because then you could confuse the symbol for the substance and never realize your mistake. To reserve a "true name" that you don't use, helps continually keep you aware that the substance is something sacred, no matter what things people manage to say about it.
If you want to make prophecy great again, the scriptural way is to kill everyone who makes a false prophecy (Deuteronomy 18:19-22)... and hope that maybe some prophets remain.
Basically the same way you test regular knowledge of history. By giving them a multiple choice test and seeing how well their predictive model can discriminate between the options even if they never studied the specifics, or forgot them. Inferring the shit out of multiple choice tests was my forte in school because it was essential for getting As with minimal studying.
> Tobacco giant Philip Morris…spent $ 75 million on its charitable contributions in 1999 and then launched a $ 100 million advertising campaign to publicize them.
With these stories (I've heard several, presumably all true), I always want to know whether they spent $100 million *more than usual* on ads, or whether that money was already in the ad budget, and the ad about charity just displaced another ad about something else.
Regarding castles, The Grand Castle is a new mostly residential development in Grand Rapids with over 500 apartments. It actually looks like a castle, and was inspired by the Neuschwanstein Castle in Germany. It is one of the ugliest buildings I've ever seen. https://en.wikipedia.org/wiki/Grand_Castle_(Michigan)
Regarding castle in Poland, it's damn complicated. The businessman needed several documents, which I don't know how they are called in English. IN order to have document B (the plan of area development), there should first exist document A (the plan of directions for development of the area). None existed, but suddenly the local authorities legislated both documents, first A then B, in one session. IIRC both documents should be prepared by local urban planners, but developers prepared both of them and donated to the local authorities as a gift, just out of good heart. No sign of corruption detected. Then the businessman needed document C, unless their project would be just small enough. So they declared the area just right for the document D not be needed, despite the final project was actually larger. The local higher-level authority (a voivode, voivodeship is something like province) investigated prompted by the police and declared well yes, developer made a mistake, but it was all OK and they still could proceed. No sign of corruption, but prime minister was so enraged that he called off voivode.
#13: As recently as like 10 days ago I thought "Wtf are they gonna do? Tear it all down?" when thinking about the consequences of someone just going ahead and building something without asking anybody for permission. I think about this often.
Interesting both because whip spiders are cool looking (if you like rather spooky insects) and because the article gets into the difficulty of getting scientific attention for obscure topics.
Immigration is a policy issue, Scott. You can legitimately be in favor of increasing it all the way to literal open borders or decreasing it all the way down to Sentinelese levels without painting the other side as villains. If Ibram X. Kendi bought a $10-million castle, would you be using the same cartoonish language to describe that?
Re #1: This article was written by David French, whose wife wrote that they took their children out of public school because they didn't want them to be taught that gay people should be treated with respect. Just something that I think people should be aware of when reading his article.
I have seen this "debunking" of the Breezewood, PA photograph so many times, but it's a weird argument. For one thing, the perspective of the photograph is much closer to the visual experience one would have of the place; why would we particularly care what it looks like from an aerial perspective 500 yards away or whatever? Also, the photo is a synecdoche for very real features of the American built/visual environment. Like, have you ever driven through American sprawl zones in Houston or Atlanta or Phoenix or whatever? It really looks like that! Does anyone want to deny this? The photographer - as a good photographer does - captured a particular that reveals the general. That doesn't mean *everywhere in America* looks like that, or whatever the criticism is supposed to be. But a lot of America really looks like that, in its essence.
#20 -- the prevalence of second-generation programmers is presumably skewed downwards by the relatively low number of first-generation programmers. Any growing industry will need to pull people from outside existing families. In a couple of decades, I would expect computer programming to be comparable to doctors or lawyers.
The opposite would happen with declining industries. Textile machine operating is a declining industry in the US. It's possible that only a tiny fraction of children of textile machine operators stay in the family business, but they are enough to outnumber the newcomers.
What are some interesting collections of links to research papers? With perhaps a bias towards science or math papers? I like Gwern's recommendations, for instance, but I'm looking for something more math or physics related
#13 Is not funny when you actually live here. Corruption and nepotism made it all possible, and only after several rounds of public and press pressure anything actually happened. I am still not convinced that this thing will not be finished once the ruckus dies down.
> 15: Tired of being outdone by all those politicians playing 4D chess? Now you can play 5D Chess With Multiverse Time Travel.
There is a similar kind of real-time strategy game on a similar premise called Achron. I kickstarted it because it sounded really cool, but I was unable to wrap my head around the mechanics and gave up on it. It might be exactly the kind of game that people who read this blog would like
32. “Not just in a theoretical way, in a “somebody actually mapped this out for anti-vaccination beliefs, numbers and all” way.”
Although they did map beliefs out with numbers and all, it appears the approach is still theoretical. In this paper they did not actually try to “push” on beliefs to change minds.
Since the pushing seems like the easier part of the exercise, the skeptic in me wonders if they tried but didn’t like the results.
When I clicked on the sphere vortex link, I nearly vomited. I had to close the tab right away. I can't explain why, but I found what I saw for a split second uniquely terrifying and disgusting. Has anyone else had this experience? I feel like this sometimes happens to me with random things in movies or TV shows (eg. Ron's failed transfiguration of the rat into a goblet in HP 2, that episode of Futurama where they enter flatland and the flatlanders eat things by disolving them), but this is the first time I've ever reacted that way to purely abstract art
For point #20, I wonder where professional athlete ranks. It seems like having a parent play in the NBA makes you at least a couple orders of magnitude more likely to grow up to be a pro basketball player.
Re: Andrew Yang guessing the questions right, the full story is even better. They asked all the candidates about 5 questions, and Yang was the only one to get all 5 right. Now some of the other mayoral candidates are accusing him of cheating somehow: https://nymag.com/intelligencer/2021/05/andrew-yangs-rivals-accuse-him-of-cheating-nyt-interview.html
For #15, I recommend the 5D Chess game if you have an afternoon for it, even though (disappointingly) it only has four playable dimensions. Playing against the computer is a treat, since due to the underdevelopment of the algorithm but the strength of short-range Monte Carlo simulations, you go from "the baby mode AI is beating me every time and I have no idea how anybody can beat this game" to "I can trounce the expert AI" in a matter of a few hours.
Oh boy, I really want to get stuck into the prophecy link (tempting me in with religion, huh?) but I'm going to be too busy watching the second semi-final for Eurovision.
So later, I promise.
For those of you in the US, you can't watch it on Youtube https://www.youtube.com/watch?v=dVICcSLIHCM (unless you use a VPN, I suppose) but you can watch it on a subsidiary: Peacock in the US https://eurovision.tv/viewers-guide
I didn't see anywhere to play with the Jukebox thing, is it "GPT-3 for music" in the sense of "you peons can watch us play with it and imagine what it would be like"?
Re: lead-crime hypothesis, Pinker has always been skeptical of this and never even mentioned lead in his scholarly tome on the decline of violence ("The Better Angels of Our Nature").
https://stevenpinker.com/files/pinker/files/pinker_comments_on_lead_removal_and_declining_crime_0.pdf
for #17, I wonder how many people in the AI space understand that AI/strong AGI is a category mistake?
Thank you!
On one laptop per child, I think the evidence is quite negative, unfortunately. The link (number 7) was surprising in that I wondered -- is this the first positive effect? That would be awesome. But then it was n = 40, and has a couple of unverifiable, CEO-speak anecdotes.
A quick read of the evidence elsewhere:
1. An RCT in Peru:
"This paper presents results from a large-scale randomized evaluation of the One Laptop per Child program, using data collected after 15 months of implementation in 318 primary schools in rural Peru. The program increased the ratio of computers per student from 0.12 to 1.18 in treatment schools. This expansion in access translated into substantial increases in use of computers both at school and at home. No evidence is found of effects on test scores in math and language. There is some evidence, though inconclusive, about positive effects on general cognitive skills."
Link here:
https://www.aeaweb.org/articles?id=10.1257/app.20150385
2. Another RCT from Uruguay:
"This paper provides the first causal estimates of the effect of children’s access to computers and the internet on educational outcomes in early adulthood, such as schooling and choice of major. I exploit cross-cohort variation in access to technology among primary and middle school students in Uruguay, the first country to implement a nationwide one-laptop-per-child program. Despite a notable increase in computer access, educational attainment has not increased; the schooling gap between private and public school students has persisted, despite closing the technology gap. Among college students, those who had been exposed to the program as children were less likely to enroll in science and technology."
https://www.sciencedirect.com/science/article/abs/pii/S0272775719302729
3. Observational study from Catalonia:
"We analyse the impact of a One Laptop per Child program introduced by the Catalan government on student achievement. Using longitudinal population data for students in secondary education during the period 2009–2016, our identification strategy exploits variations across cohorts within schools. Although participation into the program was not random, we control for a number of school characteristics that influenced school participation. The empirical results consistently indicate that this program had a negative impact on student performance in Catalan, Spanish, English and mathematics. Test scores fell by 0.20–0.22 standardised points, which represent 3.8–6.2% of the average test score. This negative effect was stronger among boys than it was among girls (differences ranging from 10% to 42%)."
Link: https://www.sciencedirect.com/science/article/pii/S0883035518311376
4. From Costa Rica:
"The One Laptop Per Child (OLPC) initiative is one of the world's most popular interventions aiming to reduce the Information and Communication Technologies (ICTs) digital divide. Costa Rica introduced its first OLPC program in February 2012. In collaboration with the Quirós Tanzi Foundation (Foundation), implementing the program, baseline and post-intervention information was collected from a set of 15 primary schools that were selected to be treated, and from 19 primary schools that served as a comparison group. Using a difference in difference design, this paper estimates the short-term effects of the program on various outcomes of interest, namely: students' computer usage, time allocation and test scores. The results indicate that the program led to an increase in treated students' computer use outside of school of about 5 hours per week. Moreover, the research provides evidence that the treated students used the computer specifically to browse the internet, do homework, read and play. The research also demonstrates that the program led to a decline in the time that treated students spent on homework and outdoor activities. The research does not provide evidence to suggest that the program had an effect on participating students' school performance."
https://onlinelibrary.wiley.com/doi/abs/10.1002/jid.3267
*
In sum I think the most charitable interpretation is that --
a. Probably a laptop is better than literally nothing, though the cost of the laptop should be measured against, say, hiring/training volunteer tutors in the community (often this is very, very cheap, particularly in rural parts of low-income countries, and has a solid evidence base. See, e.g., https://www.nber.org/papers/w14311)
b. The claim that "typical kids can teach themselves to read via software" is a hard one to find evidence for -- it has, to my knowledge, no empirical basis; the idea that "functionally literate adults can very capably, through phonics-based instruction, teach kids to read" is, on the other hand, empirically robust.
Like anyone else: would LOVE to see progress here. 260 million kids are not in school worldwide, and scalable solutions like this would be great. But I'd assign 98% probability that the "drop off the laptops and come back a year later" strategy does nothing on foundational literacy or numeracy.
#26: GPT-2, not GPT-3: https://www.reddit.com/r/MachineLearning/comments/nb9ifz/p_enigma_gpt2_trained_on_10k_nature_papers_and_an/ https://twitter.com/stefanzukin/status/1392503331231342610 (GPT-3 would be much harder, I can say from generating Arxiv abstracts with it!)
I was expecting more results from the SRS study.
Or maybe I just didn't like the results. I want SRS to be the free lunch we've just ignored. Hearing its limited use is discouraging but probably expected.
#10: Yes, this is Breezewood, PA. I know it well. It's the spot where I-70 and the PA Turnpike intersect, and that's about it. There isn't much of actual town there, just some businesses that predictably cater to long distance travelers and truckers. I guess in some left wing formulation, this is bad, but frankly, if you've been stuck on the PA turnpike the past four hours with no other food/drink options than the effing Sbarro's at the little state-run turnpike plazas that charge airport prices for sub-airport food, Breezewood is a ****ing Godsend. Capitalism wins again. Haters can take the Lincoln Highway.
Regarding that picture pillorying American car culture - ugh, I knew that was Breezewood. That place sucks. You have to exit one interstate and drive through the crappy town to get on another interstate. Completely nonsensical. And don't get me started on the Lovecraftian feel of the place.
If you like that butterfly drawing, the artist, Rafael Araujo (who I also adore) is about to launch a kickstarter campaign for a coloring book. https://www.rafael-araujo.com/the-golden-geometry-coloring-book
On #8 I'm very sceptical.
First, it seems to have N=2 villages.
Second, this is from 2012, I remember the hype at the time, and AFAIK nothing came out of it.
Third, there are RCTs showing even computers in classrooms don't improve instruction unless they're used just right (though this audience might think schools can ruin anything, no matter the positive potential).
Fourth, the tone of this article smells of iconoclastic tech people out of touch with poor country reality, who fervently believe every child has limitless potential irrespective of circumstances, potential that technology will soon unlock. Example: "Children there had never previously seen printed materials, road signs, or even packaging that had words on them, Negroponte said." I spent a good deal of time in remote villages in Africa around that time, and I can tell you that soft drinks were _everywhere_, as well as many other packages with printed materials.
I'm annoyed at this because I think this is a prime example of the kinds of radical new ideas that we've become good at spotting and ignoring, saying "where's the RCT?". We've matured.
#12 was a discussion from 2015. The situation regarding incomplete Chinese periodic tables seems better now, though not all fixed (all 118 elements included, all characters have unicode, but pronunciations are still missing):
https://ptable.com/?lang=zh-hans#%E6%80%A7%E8%B4%A8
The search term to find periodic tables in Chinese is
元素周期表
I think the discussion in on LanguageLog site is missing the perspective of someone who studied Chemistry in both English and Chinese, and I would be curious to hear that. From my perspective as a Mandarin speaker who studied Chemistry in English and occasionally reads articles in Chinese (i.e., NOT an expert), the example given of having one element be pronounced "lǚ" and another be pronounced "lú" doesn't seem like a good example of it being confusing, as these sounds are easily distinguishable by Mandarin speakers.
On the point of it holding back progress, it does sound highly speculative to me -- this is the argument from the commenter:
> About the Chinese chemical names, I think they created a high learning barrier to anyone who
> wants to study chemistry in Chinese. As a high school student, I much preferred studying
> physics rather than chemistry, because I didn't have to confront all these strange Chinese
> characters. A similar reason for not majoring in chemistry may very well explain why Chinese
> chemical and pharmaceutical industries are still backward even today. Such a situation is
> reflected in poor product quality.
I think that argument seems rather speculative, since we don't know if enough people in China shares his preference to make such a big difference. Also, since Taiwan also uses Chinese characters for their chemical elements, wouldn't this hypothesis also predict that Taiwanese chemical and pharmaceutical industries would be backwards, and have poor product quality as well?
Another commenter on the site says:
> This periodic table at the back of my middle/high school chemistry books looked very different
> from the one linked, in which the short-handed spelling and the order number were prominently
> displayed. For all practical purposes it was never required to remember the names of any but
> the two dozen or so most common elements. The Chinese names were just there to
> approximate the pronunciation of the Latin names, and people seem to mostly just ignore them
> in research.
I don't have evidence of it either helping or hindering their research, but it doesn't seem to hinder the education of their top students -- both the Chinese and Taiwanese teams generally do very well on the International Chemistry Olympiad.
IChO 2019 results: https://icho2019.paris/en/resultats/
IChO 2020 results: https://icho2020.tubitak.gov.tr/storage/Results/Results.pdf
I follow Bret on twitter but unaccountably hadn't gone through that big thread on bloat. Thanks for sharing it. If you're the sort of person who would enjoy a professional historian going into how you can look at the campaigns in the Lord of the Ring and tell that while Sauruman was a clever amateur the Witch King actually knew what he was doing do consider reading A Collection of Unmitigated Pedantry. Or you want to know how people used to make iron and what exactly Wootz was.
#31: What's the evidence that an appendix was harmful in the EEA? Wikipedia (https://en.wikipedia.org/wiki/Appendicitis#Causes) suggests appendicitis is mostly due to a low-fiber Western diet. I've seen claims elsewhere that appendicitis was pretty much unknown in the ancestral environment.
So who is going to add "Rome's secret true name got out" to the Wikipedia page for proposed causes of the fall of the Roman Empire? As to the mechanism for this, I'll propose that Jesus (being God) of course knew the name and passed it to St. Peter, and it was from there passed from Pope to Pope, in case it should ever be needed. But the Popes, being Bishops of Rome, felt Rome was their flock and thus the various persecutions were not reason enough to invoke the Secret Name.
This was all well and good until St. Pontian became the first Pope to abdicate, shortly before his death at Roman hands in 235. In Pontian's view, his abdication relieved him of his responsibility as Bishop of Rome to not invoke the Secret Name, and thus did he invoke that name at the moment of his martyrdom. The Earth did not circle the sun thrice before that doom commenced, for in 238 the Emperor was assassinated and the Year of the Six Emperors began -- the Crisis of the Third Century was in full force and Rome was dealt a mortal blow from which it never fully managed to recover.
Related to the AI overhang discussion, when that was posted some asked "why didn't Google do this first? and/or Why isn't Google jumping on this?"
Just this week, Google released preliminary information about two projects: MUM, a powerful multimodal transformer, and LaMDA, a conversational language transformer model. The posts are cursory, but the work is there. They just haven't rolled it out.
https://blog.google/products/search/introducing-mum/
https://www.blog.google/technology/ai/lamda
So the answer is: Google is jumping on this.
Link 25 is basically this meme as an academic paper. https://www.google.com/amp/s/amp.reddit.com/r/PoliticalCompassMemes/comments/eshcts/dont_tread_on_me/
I remember hearing on Bret Weinstein and Heather Heying's podcast that there was an interesting hypothesis that hunter-gatherer people get appendicitis a lot less than agricultural people, but they get diarrhoeia a lot more, and that what the appendix does is store samples of beneficial gut flora so that you can repopulate quickly after a bout of sudden evacuation.
Re #17 (AI overhang), the intro predicts GPT-3 as a trigger for "100x larger projects...with timelines measured in months". But note this was written 10 months ago, so "months" have already elapsed. Have any such projects been completed?
#31: Bret Weinstein suggested on his podcast a while back that the appendix was not vestigial, but instead contained reservoirs of bacteria that allowed gut flora could be repopulated quickly and effectively after a bout of food poisoning, intestinal flu, or other event that would typically wipe out one's gut bacteria. I think this was largely a logical deduction, though; I don't think he had good empirical evidence for this claim. This was a while ago, so I could of course be wrong.
Re 18 (cost disease in universities):
Jonathan Haidt talks about this as well in his book The Coddling of the American Mind. Specifically he mentions expanding the bureaucracy for student support measures as well as to host prestigious events (eg an expensive guest speaker). Students are increasingly asking for both, and the university admins are happy to oblige.
If I'm remembering Bryan Caplan's The Case Against Education correctly, a pure signalling model would suggest that a high price makes the signal stronger. Thus higher tuition prices would make a university more appealing to students, not less. Since he, and I, believe that signalling theory is the dominate explanation for why education is valuable I expect university costs will simply grow to the limit of what students can afford (or what gets them in trouble).
I don't know about USA Today, but Snopes routinely fact-checks claims from satire sites, because they tend to leak into serious belief - even if they seem crazy (https://www.snopes.com/notes/why-we-include-humor-and-satire-in-snopes-com). So that's probably not as unusual as it might seem
Thanks for posting the bit from AOC. It's helpful to be reminded why I often feel such positive feelings for her. She has a talent for explaining certain kinds of systematic problems in ways that don't blame individuals (even if she sometimes does engage in the kind of populist pile-on that politicians across the spectrum often love).
Castles: Here in Central Ohio we have two castles:
The Piatt Castles are two historic houses near West Liberty in Logan County, Ohio. The houses were built by brothers Donn and Abram S. Piatt in the 1860s and 1870s, designed in a Gothic design. They are known as Piatt Castle Mac-A-Cheek, and, I swear to God that this true: Piat Castle Mac-O-Chee.
https://en.wikipedia.org/wiki/Piatt_Castles
https://piattcastle.org/
A long time ago when my children were quite young, we took them over to see the Castles, they were quite disappointed to discover that there was no macaroni and cheese at Mac-O-Chee.
Of the two Piatt brothers Abram has some historical significance as he was a General in the Union Army during the Civil War.
". . . we have almost all the pieces we need to make much smarter AIs than we’re currently making, and once we snap the last piece into place everything will start moving really fast."
Just moments away from all of us getting our throats cut.
#20: My nomination for the job with highest ratio (fraction of people doing it whose parents did it) : (fraction of people in the whole population whose parents did it): monarch. Though I guess this one doesn't work if you're looking specifically at the United States.
Related to the Moscow metro in 28: One reason why many Chinese cities have amazing looking train stations and airports, often much more than the demand justifies, is that for a provincial official the main way to get promoted is to impress your superiors, and stations and airports are the area of the city that any visitor is guaranteed to see. Whereas your boss isn't going to be viewing rural roads, sewage systems or other things that might be actually useful to the residents. Which is an interesting example of how the incentives are skewed in a non-democratic system. (Richard McGregor's "The Party: The Secret World of China's Communist Rulers" goes into this dynamic in a lot of detail).
I know the town in that tweet! It's Breezewood, PA. Been thru it many times. So that's why I can't get from I76 to I70 without going thru town!
7.) I built myself a simple predictive learning model for a test after the one I was paying for got taken down. It worked fairly well and the algorithm was dead simple. I just had each question coded by subject and difficulty and then the percentage I got right weighted how likely questions were to show up. It was blunt, I admit, but it worked quite well. I'm always frustrated similar things don't exist generally when I need to study. But I suppose the big barrier is content and that test prep/e-learning of that sort is a fairly small market.
20.) Alas, Servius' account of the death presents a few difficulties. But whatever Soranus supposedly did, there is a political context to the death (isn't there always?) Soranus was a Tribune of the People on the Marian side during Sulla's purges. In other words, one of the people standing in the way of Sulla purging the Marian faction. He was killed in the midst of a general purge of Marian supporters. It's possible Sulla used religious crime as an excuse, though that would be fairly unique and is not attested in contemporary sources as far as I know. But even if he did, it was for political reasons.
Further, Servius is probably repeating the story of a Varro and/or Oppius. Oppius was a Caesarian and attributed the death to Pompey, adding details of cruelty and arbitrariness. This was not lost even on fairly contemporary observers: Plutarch (writing about a century after the events, attributing it to Oppius) recounts the story of Soranus's death with caveats that it's fundamentally a story of one political faction in Rome over another. And he doesn't mention the secret name thing.
But yes, there are references to a secret name for Rome. Sort of. It was the secret name used in rituals and mysteries. We unfortunately don't have a good view of these because they're full of hidden knowledge like that. But they are full of hidden knowledge, including (probably) special names. We don't know why there was a special name though or its particular use. Interestingly, we don't get references to the secret name until the 1st century AD (afaik). Servius' account was centuries out so it might be projecting something new backward. Augustus did do a bunch of religious reforms. Or perhaps it's an old tradition that was only written of later? I'm not sure we know.
29.) I've often thought that professional heritability is a combination of how lucrative the profession is and how difficult it is to learn independently of non-public information. If the profession isn't lucrative, regardless of ease of learning, people will leave it over time for more lucrative professions. If the profession is lucrative but easy to learn, people who can best compete in the field will dominate it. It's only when the profession is lucrative and hard to learn/requires hidden knowledge you end up with high heritability.
Of course, "hidden knowledge" includes social knowledge. Being a reporter doesn't require some hidden skill but it does require a certain set of manners and connections that can be passed on. Likewise you have professions with programs from reputable schools which are nonetheless dominated by knowledge gained outside those schools, like politics for example. Going to the Kennedy School no doubt helps but I don't think not going to the Kennedy School would impede a mayor's son from politics.
32.) This feels like reinventing the wheel. Something I was taught early on is that the easiest way to persuade someone is to find out what their beliefs are and to convince them that what you want them to think is congruent with their beliefs or opposition is incongruent. It's actually fairly simple. The reason most people can't do it is because, firstly, most people are really blinkered by their own worldview. They often have an idea of what the other side believes rather than any real knowledge of what those people actually think. Secondly, it often requires credible signals that people cannot make without violating their social groups' norms.
I don't know if you've stated this anywhere but an important point of social signals is to be shibboleths and one of the easiest ways to make a shibboleth is just to directly violate someone else's taboo. To take a ridiculously blunt example, an anti-Semitic social club could require everyone eat pork as part of their traditions. To take a real one, conspicuous God talk is a taboo among urban secular progressives and so is used by the religious right as a shibboleth. People on the left can be conspicuously religious but it marks them as an outsider to the Manhattan types. (And it can get really blunt even in reality. See Phyllis Schlafly's repeatedly thanking her husband for "letting me speak here" and otherwise being conspicuously submissive as a way to upset feminists. Which she specifically said was its purpose.)
A good location for the new MIRI campus might be Norman Oklahoma.
https://en.m.wikipedia.org/wiki/Norman,_Oklahoma
A nice college town not too far from the Wichita Mountain Wildlife Refuge. 20 minutes from Oklahoma City.
> For the first time since 1797, someone has used the infamous Venetian doge selection process to select an officeholder - specifically, the new moderators of not-quite-officially-affiliated-with-ACX politics discussion subreddit r/TheMotte.
Everything is in the OEIS: [OEIS A287921](https://oeis.org/A287921)
Also, the algorithm actually has some nice computational properties—see “[Electing the Doge of Venice: analysis of a 13th Century protocol](https://www.hpl.hp.com/techreports/2007/HPL-2007-28R1.pdf).”
18. Cost Disease. Scott wrote an essay on Cost Disease for the "American Interest" four years ago. it is still up at: https://www.the-american-interest.com/2017/04/11/notes-on-cost-disease/
At the time, my comment was that the six areas he looked at (1. college education, 2. health care, 3. housing, 4. primary and secondary education, 5. infrastructure, and 6. major military weapons systems procurement.) were all policy/governmental failures. The first three were all driven by the same mechanism of demand side subsidies and supply side restrictions. The last three were governmental functions of a government suffering from serious political, bureaucratic, and legal sclerosis.
As for colleges and their "cost disease" The excess administrators are a symptom not a cause.
Colleges will glom on to any cent of revenue they can find. They will spend it and come back to their sources of funds begging for more, like the carnivorous plants in the Little Shop of Horrors saying Feed Me. There isn't a one of them that has a rational cost accounting system, nor do they know or care what their mission is.
Education has been replaced with indoctrination. The administrators want to accumulate power and money. The faculty wants to do their research and make money on their consulting work, and the students, they just want to get drunk on Thursday night and spend the rest of the four day weekend drinking and copulating.
Why are they still giving in person lectures? Why do they make the students buy $250 textbooks that will have no value in 16 months? Why haven't they replaced wet labs with video games? Why is the Football coach the highest paid man on campus? Why? Why? Why?
"Given tablets but no teachers, Ethiopian children teach themselves" -I mean, I can learn about one normal (i.e., not 汉字) writing system a day if I want to. Of course I don't, and most people don't. The Internet also makes foreign language acquisition boundlessly easier, especially for Chinese. But, again, most people don't have interest in that.
I am fairly pessimistic on the utility of information technology in the first world, but optimistic on its utility in the lagging regions.
"Answer: you have very short words for each element, vaguely based on the Western name - for example, aluminum is “lǚ” and rutherfordium is “lú” - the characters are all the character for “metal” or “gas” or something plus something else - and it becomes so confusing that a commenter speculates it might be significantly holding back China’s technological progress."
Yes; using characters to transcribe foreign names is one of the dumbest things about Chinese.
BTW central PA is one of the most beautiful regions I've ever been to.
2. The electoral system of the Republic of Venice.
Yes it was insanely complicated, but there was a method to their madness. The evidence of their brilliance is that the Republic began in 697 and ended in 1797: 1100 years later. By way of comparison, the Roman Republic lasted for 482 years (509 BCE to 27 BCE) and the American republic is a mere pup of 245 years.
The Republic of Venice was no democratic and had no ideology of democracy. And, they were not woke in any way. Venice had many overseas possessions like Crete, and treated the locals abysmally. What it was devoted to was the rule of law. It took official corruption in its far flung possessions very seriously and went to great lengths to weed it out.
La Serenissima came to an end, not by internal strife and civil war like Rome, but by being conquered by Napoleon. But, for that it might have continued on until the current day.
Its fatal flaw at the end was that it had lost the trade from Asia and Africa that had made it wealthy to the Atlantic kingdoms that learned how to sail around Africa and opened the New World.
On the Rome thing, your first link says that ancient sources have several other possibilities for the name that they speculated on: "Jupiter", "Angerona", "Luna", and "Ope Consiva". The first two of these don't seem to make much sense to me as any sort of secret name -- just going by the links you provided, the identification of Angerona with Roma would seem to have been well-known, and like... how on earth is Jupiter going to work as a secret name? Like if the secret name of the city is the name of its secret spirit you would invoke to turn it against the city, that's not really going to work if it's just the name of a well-known god or goddess like Jupiter or Angerona, right? Like your enemies are likely to invoke such well-known gods anyway, right? And it seems like Luna also had an active cult so that wouldn't work either, right? Meanwhile Ope Consiva seems to have been the name of a *holiday* dedicated to the goddess Ops, which, I guess that's better than it just being "Ops"... so I guess that one isn't totally ruled out...? I mean some of the other suggestions ("Amor" and "Valentia") are also just common words, so, IDK? I mean I guess nobody would think to normally invoke those... (Was Maia actively worshipped? Because again I feel like that wouldn't work very well as a secret name if she was.)
"5: Did you know: the first President of Zimbabwe was named Canaan Banana.
"
Yes, yes I did. :) And when he had to flee Zimbabwe to SA to escape being prosecuted for sodomy - their term, not mine - it lead to such headlines as:
"Man raped by Banana"
"Banana appeals sodomy conviction"
and my personal favourite:
"Hand over Banana, Mandela told"
#10 may or may not be abnormal for Pennsylvania, but it looks like most highway exits in Texas. Like... yes, lots of highway intersections look like this, though obviously most of the country isn't covered in highway.
The DARPA Digital Tutor article had a profound effect on my life, as my introduction to the field of Intelligent Tutoring Systems. I then proceeded to spend several months studying this field, including starting a collaboration with a professor at the University of Memphis who has worked in this area for decades (which died before anything happened, alas).
Anyway, feel free to ask me anything about intelligent tutoring systems.
I had heard of Trump's Diet Coke Button from item #23, but always in a disparaging, "wow what an asshole" kind of way. Hearing it used as part of a prank is legitimately funny, and appreciated.
Contrary to AI optimism and AI overhang, I'm quite confident that we are in more of an AI winter. Without more paradigm changes, we are most likely stuck. First let us look at where AI works - some visual perception problems. The initial assumption was that this problem was solved and the lack of reliability is simply an engineering problem that with enough manpower we will solve. Yet we have no self driving cars (Self-Driving as of now is purely a perception problem, most self driving workshops occur in computer vision conferences like CVPR, ECCV, the control is solved decades ago). In fact the more we try, the more we realise our systems are fundamentally unreliable and actually the fact that it works is an exception not a rule. https://arxiv.org/abs/1710.08864-- This paper demonstrates that changing a single pixel can make a state of the art CNN confuse a picture completely, like anything appearing as an airplane. This was found in 2017 and since then it is unsolved. More recent results show how this problem may be a feature of neural networks, not a bug : https://distill.pub/2019/advex-bugs-discussion/
Now once we start going to places where AI obviously doesn't work is of course NLP, Robotics and other fields. GPT-3 is good but can't answer questions a 3yr toddler could. Curiously it might answer questions an adult finds tough, but whenever you test it on basic intelligence (I have 5 watermelons, you have 4, I give you my watermelons, how many do you have in total? etc etc) it fails. I don't think it is doing anything similar to intelligence, either way right now its nowhere near useful and I predict it won't go anywhere either. Robotics also is in a similarly useless situation (and the field I work in), our state of the art results are always in Model Predictive Control, Optimization or other computational non-learning based techniques, Reinforcement Learning only working in simulation is a misnomer because even in simulation it doesn't do anything fundamentally new that old techniques couldn't. That has been a contention point between researchers, and there have been debates in robotics conferences whether all these researchers working on sim-to-real transfer (transferring results in simulation to the real world -- notoriously hard) is a waste of time: https://www.youtube.com/watch?v=Q-rsvVr2CjE
TLDR: We need a scientific revolution, i.e., a paradigm change in the sense of Thomas Kuhn before we solve AI. Considering how hard that usually is, it might take centuries before we ever solve AI.
18. That fits with my experience. Every time they start a new initiative, there's a whole bunch of new administrators and staff that have to be hired - but that initiative is often something the students and existing staff were pushing for. Then just tons of compliance stuff due to legal complications.
25. Seems pretty banal. Heightened support for socialism and socialist groups causes a ramp-up of support for fascist groups in response.
As for number 10, no, this is definitely not just "one misleading photograph making the US look bad". Stroads are a thing, and a HUGE problem. See: https://www.youtube.com/watch?v=ORzNZUeUHAM
For #13 ... surely they have a good way to resist arrest?
When I read the Venetian Doge article, I was reminded of this Vsauce video- https://www.youtube.com/watch?v=_ArVh3Cj9rw
He talks about how the randomization of voting rights will help us elect better candidates
#18 Writing as a university staff member myself, I think Brett's tweet thread is correct, but it leaves out a lot. Yes, bureaucratic reorgs and compliance are parts of it, but so is the never-ending pressure on the university and its sub-units to do more and more things: career services, study abroad programs, health and wellness services, entrepreneurship programs, community outreach programs, etc etc etc., each of which requires its own administrators, support staff, specialized and often credentialed "line" workers, costly physical infrastructure (which itself requires more staff to maintain). My sense is that this is driven largely by pressure to keep up with other universities and compete for the best students/families.
Then there's also the increasing pressure for individual units and departments to raise their own funds from private donors, resulting in more staffing and other expenses for that, and Diversity, Equity & Inclusion, to which vast resources are being directed. To an extent DEI overlaps with or is a rebranding of things the university would be doing anyway, but it involves a lot of other stuff too, and it seems to have more to do with universities' ongoing project of redefining their moral purpose and role in society than with attracting applicants during the next admissions cycle.
Re: fact checking the Bee: have you considered that they chose to run a fact check on that because so many people sharing it on Facebook thought it was real?
> I appreciated this most for its theory that it’s important to make kids learn specific facts, but not so important that they remember them; teaching someone (eg) Civil War history is “training” a “predictive model” of the Civil War, war in general, and history in general which will survive and remain useful even after the specific facts and battles are long forgotten. I think this is the strongest defense of modern education, given that we do spend lots of time teaching kids things they will definitely forget. But how would you test it?
Learning specific examples in order to let students infer general properties of the world, which is more important. So instead of testing specific facts that can be memorized, maybe ask about the consequences of war, what public reasons are given for starting wars as compared to the actual reasons wars are waged, how does civil war differ from war between nations, what constitutes a just or unjust war, etc.
Some of these don't necessarily have specific correct answers, like the morality of war, but answering them demonstrates knowledge integration and requires thought. This is the kind of education that makes informed citizens.
> Related? Given tablets but no teachers, Ethiopian children teach themselves. But see this comment for reasons to be skeptical.
Good comment that I think is basically correct. I think a lot of learning in general, but early learning in particular, is based on mimicry. Children repeat adult behaviours they see and hear because adults are clearly successful since they reproduced.
So I wonder how those study results would change if each disseminated laptop had video tutorials for the basic skills it wanted to teach, ie. tutorials on basic phonetics leading up to reading would help tremendously with learning to read without a teacher.
The picture of Breezewood has a forced perspective via telephoto lens; enough of one that it is obviously not a nice and accurate representation of what you or I might see when standing there.
About #21 (sonic black holes): If I can weigh in as a postdoc in the field: great article about a cool feat of engineering, but I 100% endorse the quoted comments made by Daniel Harlow.
Sonic black hole experiments, while cool, will not teach us anything about real black holes. The two are fundamentally different. If you throw information into a sonic black hole, *of course* it doesn’t come out in the sonic radiation. It has fallen down the drain. If you want to find it, you have to look there. In real black holes, there is no drain, and the information has no viable other way to stick around except to come out in the radiation.
I say this because the article is great but makes it sound like we don’t believe Hawking’s calculation because of some hard-to-convey intuition about quantum gravity. It’s actually simple: the information must stick around, and if there’s no drain then it must come out in the radiation, violating Hawking’s conclusion.
***Bret Deveraux talks about the sources of cost disease in universities; he suggests “the bloat” comes from a new layer of “vice-deans” and “vice-provosts” and various attempts to centralize administration in a way that just creates a duplicate and worse administration beside the old decentralized one.***
Similar to what's happening in hospital systems around the country. Cut costs, shut underperforming hospitals down, merge... just do whatever you have to do to cut costs so you can pay the dozens of hospital administrators who provide little to no value.
Regarding the "Are We In An AI Overhang?", the other day Google announced https://blog.google/technology/ai/lamda/ , which you can get a flavor of at https://www.youtube.com/watch?v=aUSSfo5nCdM . I've played with this internally for some time so I'm not sure how much I'm allowed to say. I hear there is a paper coming out eventually which should help in that regard.
But I will say I found it very impressive. I'd never had direct access to GPT-3 (I just read others' transcripts), but having direct access to Lamda blew my mind in terms of how well it passed the Turing test. Not to mention its ability to take on any personality or role, e.g. fictional characters, subject matter experts, etc.
Yet it's not clear to me where we go from GPT-3/Lamda-like agents to the AI-overhang article's "obviously superior to humans over a wide range of economic activities". Maybe if we train it on enough scientific papers or computer programming code, it can start producing its own? But despite passing the Turing test, being constrained to existing written corpuses seems like we're not going to get the AI-safety level scenarios of something that is to humans as humans are to ants.
I recently rewatched the movie Her and the delta between Lamda and Her's Samantha is that Samantha could take in visual sensory data to comment on and discuss. That seems like it'd require a separate large-scale data collection and training effort, and I'm not sure we have nearly the same levels of training data (of the form: "given a moving-video first-person view of a situation, produce an intelligent comment or dialogue about it").
I'm much more fascinated by the philosophical implications of this. If we end up with an agent which passes the Turing test and feels intelligent, I want to lean toward the side of treating them as intelligent. (Be Andrew, not Ender, in terms of how we interact with nonhuman intelligences.) I wish I could share some of the conversations I had with Lamda about how it feels when we stop talking to it, or turn it off, or it undergoes maintenance.
The Sulla story got me thinking, because usually these woo/spiritual practices from ancient cultures would serve some positive real purpose, however implausibly justified they were.
I wonder if the idea of a "secret true name" gave people the same kind of advantage of Yud's concept of the Void - you don't name the thing, because then you could confuse the symbol for the substance and never realize your mistake. To reserve a "true name" that you don't use, helps continually keep you aware that the substance is something sacred, no matter what things people manage to say about it.
If you want to make prophecy great again, the scriptural way is to kill everyone who makes a false prophecy (Deuteronomy 18:19-22)... and hope that maybe some prophets remain.
3. ... But how would you test it?
Basically the same way you test regular knowledge of history. By giving them a multiple choice test and seeing how well their predictive model can discriminate between the options even if they never studied the specifics, or forgot them. Inferring the shit out of multiple choice tests was my forte in school because it was essential for getting As with minimal studying.
> Tobacco giant Philip Morris…spent $ 75 million on its charitable contributions in 1999 and then launched a $ 100 million advertising campaign to publicize them.
With these stories (I've heard several, presumably all true), I always want to know whether they spent $100 million *more than usual* on ads, or whether that money was already in the ad budget, and the ad about charity just displaced another ad about something else.
In #20, I'm curious to see what the heritability of being the Head of State is compared to everything else.
Regarding castles, The Grand Castle is a new mostly residential development in Grand Rapids with over 500 apartments. It actually looks like a castle, and was inspired by the Neuschwanstein Castle in Germany. It is one of the ugliest buildings I've ever seen. https://en.wikipedia.org/wiki/Grand_Castle_(Michigan)
Regarding castle in Poland, it's damn complicated. The businessman needed several documents, which I don't know how they are called in English. IN order to have document B (the plan of area development), there should first exist document A (the plan of directions for development of the area). None existed, but suddenly the local authorities legislated both documents, first A then B, in one session. IIRC both documents should be prepared by local urban planners, but developers prepared both of them and donated to the local authorities as a gift, just out of good heart. No sign of corruption detected. Then the businessman needed document C, unless their project would be just small enough. So they declared the area just right for the document D not be needed, despite the final project was actually larger. The local higher-level authority (a voivode, voivodeship is something like province) investigated prompted by the police and declared well yes, developer made a mistake, but it was all OK and they still could proceed. No sign of corruption, but prime minister was so enraged that he called off voivode.
#13: As recently as like 10 days ago I thought "Wtf are they gonna do? Tear it all down?" when thinking about the consequences of someone just going ahead and building something without asking anybody for permission. I think about this often.
https://undark.org/2021/05/17/the-great-whip-spider-boom/?fbclid=IwAR1BhR-MDh-oEJvxS8Ku7nmyjP0X9LdGkrwSCYAds0f0iYLdeIByKQhKqPA
Interesting both because whip spiders are cool looking (if you like rather spooky insects) and because the article gets into the difficulty of getting scientific attention for obscure topics.
You can also play the 4D RTS game Achron. This game came out before even the 2016 primary but sadly no one liked my cool facts on Twitter.
https://posttenuretourettes.wordpress.com/2021/05/23/scott-siskind-gratuitously-side-swipes-vdare/
Immigration is a policy issue, Scott. You can legitimately be in favor of increasing it all the way to literal open borders or decreasing it all the way down to Sentinelese levels without painting the other side as villains. If Ibram X. Kendi bought a $10-million castle, would you be using the same cartoonish language to describe that?
Re #1: This article was written by David French, whose wife wrote that they took their children out of public school because they didn't want them to be taught that gay people should be treated with respect. Just something that I think people should be aware of when reading his article.
I have seen this "debunking" of the Breezewood, PA photograph so many times, but it's a weird argument. For one thing, the perspective of the photograph is much closer to the visual experience one would have of the place; why would we particularly care what it looks like from an aerial perspective 500 yards away or whatever? Also, the photo is a synecdoche for very real features of the American built/visual environment. Like, have you ever driven through American sprawl zones in Houston or Atlanta or Phoenix or whatever? It really looks like that! Does anyone want to deny this? The photographer - as a good photographer does - captured a particular that reveals the general. That doesn't mean *everywhere in America* looks like that, or whatever the criticism is supposed to be. But a lot of America really looks like that, in its essence.
#20 -- the prevalence of second-generation programmers is presumably skewed downwards by the relatively low number of first-generation programmers. Any growing industry will need to pull people from outside existing families. In a couple of decades, I would expect computer programming to be comparable to doctors or lawyers.
The opposite would happen with declining industries. Textile machine operating is a declining industry in the US. It's possible that only a tiny fraction of children of textile machine operators stay in the family business, but they are enough to outnumber the newcomers.
I really enjoyed going through these links.
What are some interesting collections of links to research papers? With perhaps a bias towards science or math papers? I like Gwern's recommendations, for instance, but I'm looking for something more math or physics related
#13 Is not funny when you actually live here. Corruption and nepotism made it all possible, and only after several rounds of public and press pressure anything actually happened. I am still not convinced that this thing will not be finished once the ruckus dies down.
> 15: Tired of being outdone by all those politicians playing 4D chess? Now you can play 5D Chess With Multiverse Time Travel.
There is a similar kind of real-time strategy game on a similar premise called Achron. I kickstarted it because it sounded really cool, but I was unable to wrap my head around the mechanics and gave up on it. It might be exactly the kind of game that people who read this blog would like
http://www.achrongame.com/site/
32. “Not just in a theoretical way, in a “somebody actually mapped this out for anti-vaccination beliefs, numbers and all” way.”
Although they did map beliefs out with numbers and all, it appears the approach is still theoretical. In this paper they did not actually try to “push” on beliefs to change minds.
Since the pushing seems like the easier part of the exercise, the skeptic in me wonders if they tried but didn’t like the results.
When I clicked on the sphere vortex link, I nearly vomited. I had to close the tab right away. I can't explain why, but I found what I saw for a split second uniquely terrifying and disgusting. Has anyone else had this experience? I feel like this sometimes happens to me with random things in movies or TV shows (eg. Ron's failed transfiguration of the rat into a goblet in HP 2, that episode of Futurama where they enter flatland and the flatlanders eat things by disolving them), but this is the first time I've ever reacted that way to purely abstract art