PokerPirate
12hr ago
Maybe the only thing I really can ask is what you can do to help someone who's clearly overloaded, but can't stand it when something isn't done the way they would do it, and doesn't know how to explain what they want?
I had testicular cancer ~decade ago. I was the same way during chemo. I didn't like it when other people did "stuff" for me (don't like how they do it, don't want to be a burden, don't want other people's pitty, etc). But I very much appreciated a good game of chess. From other people I've talked to in similar situations, "fun" stuff is can be more well received than "help" stuff.
PokerPirate
13hr ago
In what world are 2 tenured Stanford professors considered upper middle class ?
This world. Upper class is the people who have names on buildings, not the people who work in the buildings.
PokerPirate
13hr ago
former math professor
upper class? hahahahahahaha
PokerPirate
13hr ago
It’s ok to be bad at school because the only thing you need to do to have an OK life in America is to… not fuck up.
This is so well said. I should print it out and put it in my office for students to read.
I wonder how much of the culture war can be charitably phrased as: On the one side we have people striving to try to make themselves/America better and on the other side people worrying that this striving is going to fuck things up. (It's not too hard to find examples that work with both red/blue as either side.)
PokerPirate
13hr ago
I will answer your questions with two questions of my own. (The questions are semi-rhetorical in that I think they shed light on the answers to your questions, but also I would genuinely really like answers and I haven't seen any good answers.)
Theoretical Q.
I overall like the forecasting trend in the rationalist community. I find the idea of quantifying bias and uncertainty to be a valuable exercise that I have benefited from personally. I have a theory-level concern, however, that I've never seen properly addressed.
I internally model forecasting as: there exists a probability distribution over all possible futures, and the job of the forecaster is to approximate this distribution. In practice, forecasters do this by assigning probabilities to a bunch of events and then scoring themselves based on what actually happens (like you describe in your OP).
So here's my question: How confident can we actually be that your scoring algorithms are stable and consistent? I'm using these words in the technical sense from statistics. To see an example of how everything can go bad: Let's say you're trying to predict the number of people who die in 2026. If the true distribution of deaths/year is gaussian, you can use standard formulas for computing the mean and get a good estimate with error bars. But if the true distribution is Cauchy, the mean is undefined, and there is provably no way to accurately estimate this mean because it doesn't exist. The Cauchy distribution looks essentially identical to the Gaussian distribution, and it is extremely difficult to determine whether you are actually sampling from one or the other in practice. In practice, people who work under the Gaussian assumption will look like they're doing very well by the metrics superforecasters use until suddenly they have a disaster (see e.g. the 2008 financial collapse). Similarly, a 40% return over 5 years is "trivial" to achieve if you allow yourself to have a very high risk of ruin. Just invest in the S&P500 with 5x leverage.
So what are the actual, philosophical and statistical assumptions about the universe that superforecasters are relying on?
Practical Q
I work professionally with North Korea. I put in a lot of time studying their culture, geopolitics, language, etc in order to make my professional work more effective. I've long thought about how to quantify this work both to make my work even more effective and to convince other people that I am an expert on this topic. How do I go about as a practical matter starting to forecast on a very niche topic like this?
My impression is that most forecasters work very generally and basically try to eek out an edge over the general populace by (like you mention) not being fooled by basic statistical fallacies. This lets forecasters make more level-headed judgements about a wide range of topics, most of which are well-established questions that normies also think about (who will win the election? will an epidemic cause a downturn in the economy? etc.)
But I am interested only in a very narrow domain where there are basically no established questions to ask. With regards to North Korea, the basic questions might be:
- Will Kim Jong Un die this year? (Almost certainly no; without looking it up, I'd guess the actuarial tables put him as <5% chance of death.)
- Will the North and South declare war? (Also almost certainly no; I'd put it <1%.)
- Will the North and South have a military skirmish? (Happens 1-2 times per decade, so let's say 20%)
But these are all super basic questions that anyone moderately politically aware could reasonably answer. There's no opportunity for me to develop my skill with questions like this, and there's not a "large enough n" for me to meaningfully test my skill. So I need to develop more detailed questions if I want to really improve my forecasting ability. But how? Some more detailed questions could be:
- Will the North develop a new fully domestic cell phone in 2026? (I'd say 75% probability since they've been developing them the past few years. But then what exactly counts as "new" and what exactly counts as "fully domestic"?)
- What will the price of rice be in Jan 2027? (It's currently 1 kilo/1800 won. I predict it will be <2200 in 1 year with 75% probability. Either a bad crop this year or more economic sanctions from the US could increase the price substantially, and I'll say that the union of those two events is about 25% probable.)
But how do I go about actually creating good questions like this? You especially want the questions to be correlated with the "basic"/important questions above, but it's not at all clear to me that the ability to predict food prices is at all related to the ability to predict whether and how large of a military conflict there will be.
One last aside: You don't mention the intelligence community at all. This is where calibrated predictions are rewarded more than narrative, and this is where people who actually want to work as superforecasters work. Some "fun" reading if you haven't already seen them are:
-
"Psychology of Intelligence Analysis"
-
"A Tradecraft Primer: Structured Analytic Techniques for Improving Intelligence Analysis"
-
"Analytic Culture in the U.S. Intelligence Community"
These are all declassified CIA publications you can get from cia.gov. Most of my questions/frustrations expressed above are things that I've thought about from reading these works and talking to the people who use them professionally.
PokerPirate
2d ago
That's a crazy video. I can't believe that the cop kept his finger off the trigger (either for the taser or the pistol). I know all about trigger discipline, but I'm shocked that police would keep their finger off the trigger when a potentially armed suspect is digging through their pants like that. The extra couple hundred milliseconds to move the finger into the trigger guard could really make a difference.
PokerPirate
2d ago
I wrote a response to @sarker that also responds to your part A.
Not my problem. You're complaining about the existence of consumer credit.
Yes, I'm complaining about the existence of consumer credit (at least as it's practiced today). But even more so I'm also complaining about the "not my problem" attitude.
I do in fact care about the welfare of my fellow countrymen. I even care about the financially illiterate and irredeemably midwit among us. Every fancy financial scheme that exists makes these midwits feel like suckers for not taking advantage of it, and so they try to take advantage of it and get their lives wrecked because they're not equipped for it.
PokerPirate
2d ago
I disagree with Patrick McKenzie. I think are disagreement is probably at a philosophical level that I don't want to go into, so instead I'll share a personal story:
I teach a data science practicum course for economic majors where financial institutions "hire" our students. One of the projects our students were contracted for was literally optimizing the advertisement of credit card rewards programs to attract low income consumers who would not default on the loans but would carry a high interest balance. Another project was optimizing the fee schedule to extract the most money as possible from overdue payments on these cards from low income consumers.
I've sat at the table with the men and women who run these programs. I've asked them how they justify it to themselves. They fully acknowledged that some people were ruining their lives, but they did not have any moral qualms and said "everything we do is legal and fully regulated". So I think the folk that run these programs are every bit as evil as the worst communist propaganda would have you believe.
(I refused to work with these companies, but other professors chose to work with them.)
PokerPirate
2d ago
Except for all the people who get into massive credit card debt who these programs are actually trying to target and where the credit card companies make all their profit.
PokerPirate
3d ago
At least in a couple cases it would also be irresponsible for them to break up their extant lives in the U.S. to go over and maybe die for a regime change.
Just like it's irresponsible for the parents in the US military to disrupt their child's development to go on deployment.
PokerPirate
3d ago
they are pretty vehement that it'd be worth significant amounts of death to remove the existing regime
But notably not their death. If they actually thought this way---from a revealed preferences angle---they'd be out guerilla-ing.
PokerPirate
3d ago
I'm good with the idea that the we can't always have a Wild West and that part of what made the West fun was the taming of it, and so we can't have it again.
But I'm not cool with the idea that destroying the commons is okay when I do it in a classy way but not when those shlups do it in a low class way. It's either a commons that needs preserving or a resource that needs exploiting.
As an aside, my impression is that there is still a lot of finds to be had at estate sales (at least in CA). I think the real reason the thrift store market has dried up is not because of people buying the good stuff from the thrift stores, but because the suppliers have stopped sending the good stuff to the thrift stores. People now find the good stuff at the estate sale, and so the left over junk that gets donated has much less signal to noise. I suspect the higher prices at the thrift store are also related to garden variety inflation, where it is magnified tremendously by not being part of the official basket of goods tracked.
PokerPirate
3d ago
Mrs. FiveHour, when between jobs, made tens of thousands of dollars buying at Goodwill or Poshmark and arbitraging to Ebay or TheRealReal.
It's weird to hear you lamenting the decline in thrift stores when you actively destroyed what made them special :/
PokerPirate
3d ago
Here's an example of the correspondence between Bayesian/frequentist interpretations for linear regression: https://stats.stackexchange.com/questions/283238/is-there-a-bayesian-interpretation-of-linear-regression-with-simultaneous-l1-and.
PokerPirate
6d ago
Fun Unicode story time.
In 2015, I went to North Korea to teach computer science. One of the things I taught was how to integrate their computer systems to the internet, and one of the major challenges is the lack of compatibility between Unicode and their internal character sets.
In North Korea, when they use Unicode (which is rare actually), they use these private-use code points for special characters for the Kims. The Kims are thought to be so special that their names are always written in a fancy calligraphy, and the Norks don't want to rely on HTML to provide this fancy calligraphy (because that might not always be available), and so they do this calligraphy at the font level. The "advantage" of this is that you can differentiate between an "ordinary" peasant Kim Il Sung (of which there actually were some) and "the" Kim Il Sung at the font level in every computer program. (The North Koreans didn't invent this idea, but rather borrowed it from how some Arabic encodings treat Muhammad and his sayings.) Anyway, this caused problems for US diplomats when we would receive documents from the North, convert them to Unicode, but then all references to any Kim would appear as square boxes and diplomats didn't know whether the document was talking about Kim Il Sung, Kim Jong Un, or Kim Jong Il. So I went to Korea to help sort this mess out.
Below is (part of) a memo I wrote for the North's ministry of education that outlines some of the other problems that the North has had with the Unicode standard.
Technical Problems
The Committee for Standardization of the DPRK (CSK) submitted a memo to the Unicode Consortium in 1997 that lists three difficulties in working with Unicode in the DPRK. None of these problems have been fixed in the last 25 years. The problems are:
-
The official name of the Korean language script in Unicode is "Hangul" (see Section 18.6 of the Unicode 14.0 standard). Hangul is the ROK's name for their script, and the DPRK prefers the name "Choseongul". The DPRK suggested that the name "Korean characters" be adopted as a politically neutral term.
-
The DPRK and ROK use a different sorting order for their alphabets. The ROK order for consonants is
ㄱ ㄲ ㄴ ㄷ ㄸ ㄹ ㅁ ㅂ ㅃ ㅅ ㅆ ㅇ ㅈ ㅉ ㅊ ㅋ ㅌ ㅍ ㅎand the DPRK order is
ㄱ ㄴ ㄷ ㄹ ㅁ ㅂ ㅅ ㅈ ㅊ ㅋ ㅌ ㅍ ㅎ ㄲ ㄸ ㅃ ㅆ ㅉ ㅇFor example, in the ROK, the word 까치 (magpie) comes alphabetically before the word 나비 (butterfly), but in the DPRK the word 나비 comes alphabetically before 까치.
The Unicode standard orders Korean characters according to the ROK-ordering, and so by default all sorting done in any programming language will sort Korean words in the ROK-preferred way. A special extension called a collation algorithm is required to sort according to the DPRK-ordering.
As of 2022, the current list of collation algorithms does not have an entry for the DPRK-dialect of Korean, and so it is currently impossible in any programming language to sort text alphabetically accoding to the DPRK-ordering.
-
The DPRK internally uses the KPS9566 character set. This character set contains several characters that the Unicode Consortium does not want to support. For example, it contains political characters representing the Workers Party of Korea, and 4 distinct versions of the character 김 (one for normal text, and one each for Kim Il Sung, Kim Jong Il, and Kim Jong Un).
This lack of support for certain characters used by the DPRK prevents documents produced in the DPRK from being opened in tools like Microsoft Word, and even programming languages like Python and R cannot work with these documents. This lack of compatibility adds considerable friction to negotiations, since diplomats between the DPRK and the United States cannot easily exchange documents.
There is at least one more problem with the Unicode standard for the DPRK not listed above:
- The current Unicode standard does not support transliteration of Korean into Latin characters using the DPRK's preferred Romanization system, and instead only supports the McCune–Reischauer system. Furthermore, transliterations into non-Latin alphabets are not supported at all, despite the importance of transliterating into Cyrillic. A 2018 UN report on romanization describes a good history of the many Romanization systems for Korean.
Historical Basis
The ROK has been actively and publicly developing their systems for encoding Korean text since the earliest days of the internet. KAIST first developed the KSC5601 encoding method in 1974, and actively worked with companies like IBM and Microsoft, and standards organizations in the US and Europe to ensure widespread support for this standard. The ROK issued an official Request for Comments (RFC) on the encoding in 1993 via RFC1557 to suggest that KSC5601 be the standard format for exchanging Korean emails. When the Unicode Consortium was first founded in 1991, ROK programmers were well positioned to contribute to the developing standard. They had the detailed technical knowledge of developing many of their own internal encodings, they had experience interacting with diverse technical committees, and they had the English communication skills for communicating in the Unicode Consortium's working language.
In contrast, the DPRK has severely lagged the ROK in this area. It's not known when the DPRK first developed their own Korean encoding, but the DPRK's KPS9566 encoding was first published internationally in 1997 and officially registered with the Internaional Standards Organization (ISO) in 1998. It wasn't until August 1999 that the DPRK began discussions for enabling Unicode compatibility. The DPRK submitted an official statement to the Unicode Consortium outlining their difficulties adopting the Unicode standard (summarized above), but since they entered this discussion 8 years after it began, the technical decisions had already been made. In order to not break backwards compatibility, the Unicode Consortium issued a statement that they could not implement the changes requested by the DPRK.
Fun Fact: There are 7 emojis in the current Unicode standard that were added at the request of the DPRK. The DPRK originally suggested that the HOT BEVERAGE emoji ☕ should be called the HOT TEA emoji, but an American suggested the emoji be renamed so that Americans could use it to represent coffee. The DPRK delegation agreed, and so the emoji was renamed. This is an example of technical experts working on narrow technical problems being able to work together in a way that diplomats can't.
PokerPirate
7d ago
I'm a white 40yo male who grew up in a middle class CA suburb. It was explicitly taught to me in school that if a police officer pulls you over and you put your hands out of view, you will be shot (because you could be reaching for a gun).
PokerPirate
8d ago
describing them as "making people disappear" is not an unfair characterization.
When Pinochet "made people disappear" in Chile, they had cement blocks tied to their feet and were thrown into the Pacific Ocean from a helicopter. Then the government never acknowledged that this happened.
These death flights have been used by proper dictators to "make people disappear" all over the world. The Trump method of "disappearing" people is very different, and using the same word to describe them is an obvious motte-and-baily.
PokerPirate
8d ago
I don't think it's reasonable to expect cops to engage in that kind of self-risk to avoid shooting people, but I think cops should aspire to as a matter of personal virtue.
This is well said. I find it quite rare for people to make this distinction between what should be legal and what should be matters of personal virtue. I'd wager all the culture war battles talked about here could be profitably thought about through this lens.
PokerPirate
9d ago
Universities are allowed to essentially act as courts for their employees and students, with far more power over them than a private-sector employer has in almost any field,
I don't see how universities have more power over their employees than any other employer. They can fire (and sort of blackball) an employee, but so can any other employer.
Students, on the other hand, lose an insane investment if they get kicked out. So I agree there.
PokerPirate
10d ago
As a machine learning researcher, I don't find anything in the article that is outright wrong, but I also don't see any insights that are new or useful to me.
One of the fundamental results of machine learning theory states (very informally) that every learning algorithm has both a bayesian and a frequentist interpretation. So the quote "[LLMs] aren’t Bayesian by design; they are Bayesian by geometry" is certainly true, it is also true of every other possible learning algorithm. Basically everything else in the article strikes me as the same sort of tautology.
PokerPirate
10d ago
I think most adults have forgotten how to have fun.
And this isn't a new problem. When Jesus said "unless you turn and become like children, you will never enter the kingdom of heaven", I think he was telling his followers that they need to have fun like children.
One of the main reasons that adults can't have fun is because we worry too much. This worry causes us---and this tweeter in particular---to focus on "productive things" instead of "fun things". But fun things done right are actually highly productive! One of main evolutionary purposes of play is to teach us new skills: Building legos is fun because we learn new techniques that can help us build bridges. Basketball is fun because it improves our body's conditioning. Reading is fun because it expands our imagination.
I am a father of 4 young kids (7m, 4m, 3m, 2f). It's obviously exhausting at times, and I don't begrudge any parent who complains about the exhaustion. But most adults I know are just genuinely not fun people. They are either too addicted to their "productivity" or wireheaded by social media/tv/literal drugs. But kids are untainted. They still know deep in their bones how to have fun, and are constantly seeking it out.
PokerPirate
10d ago
All parents talk about children being exhausting, and it's not any sort of secret. I am the father of 4 young children. The easiest way to start up a conversation with another parent (at a park/school/cub scouts/whatever) is to observe that their children have a lot of energy and commiserate with them. I've easily had this conversation with >100 different people.
PokerPirate
28d ago
This is a non sequitur. Good treaties define Schelling points, and Schelling points do not need enforcement.
As I mentioned before, no one has a military powerful enough to enforce the CWC on the US and yet the US is voluntarily complying because it makes the US a safer place (and increases trade/commerce) to allow these foreign inspections.
PokerPirate
28d ago
C-nile greybeard
Worth reading the post just for this pun.
My young kids have just gotten into minecraft, and this seems like something they'd love. We actually play luanti (an open source clone) so there's no chance of connecting to your server, but maybe I'll have to tinker around with something like this. There's been a lot of cool work in reinforcement learning with minecraft, but I never did anything with that since it sounds too compute intensive for me, but stuff with llms would be much easier/cheaper.
More options
Context Copy link