This maybe points to another theory (which may be entirely wrong, I'm just guessing!): honeybees die because they aren't supposed to attack each other. Like they can't be aggressively selfish because they'll just die in the process.
Honeybees do attack other colonies' honeybees. Africanized honeybees definitely do it. As someone else points out the barbs don't get stuck in insects, but do get stuck in mammals (and presumably birds too?).
> Africanized honey bees are typically much more defensive, react to disturbances faster, and chase people further (400 metres (1,300 ft)) than other varieties of honey bees.
My bees will chase me about 200 yards, and probably more if I had to go further to go inside a building (they don't like dark places). They lay ambushes, too. They'll wait outside the building I go into and will attack again if I go out.
Well, they used to anyways. Since switching to top-bar hive bodies they're much nicer.
I had this experience with a wasp nest near my house. I figured "live and let live" until one day I walked out my door and a wasp flew directly over and stung me without provocation. So I got some insecticide and got rid of the nest.
I also thought it was interesting that it was actually several BASIC programs with data passed back and forth by stuffing it in specific memory locations.
I notice if you look at Madrid it includes all of Spain and none of Portugal, and similarly from Lisbon. I assume this is because the schedules don't line up, because it wouldn't really make sense in terms of physical distance.
About 20 years ago I was in Portugal and remember it was a TGV project that didn't take off. It was supposed to connect to Madrid I believe. I imagine that politics and money created this issue.
That keeps being discussed and filling pockets from politicians and their close friends.
We could already have good connections with Alpha Pendular trains, it would be a matter to extend and improve existing infrastructure, not build a new one from scratch.
Same goes for the never ending story of the new Lisbon airport, it will make very happy everyone involved into its construction.
"Langsmith appeared popular, but we had encountered challenges with Langchain from the same company, finding it overly complex for previous NonBioS tooling. We rewrote our systems to remove dependencies on Langchain and chose not to proceed with Langsmith as it seemed strongly coupled with Langchain."
I've never really used Langchain, but setup Langsmith with my own project quite quickly. It's very similar to setting up Langfuse, activated with a wrapper around the OpenAI library. (Though I haven't looked into the metadata and tracing yet.)
Functionally the two seem very similar. I'm looking at both and am having a hard time figuring out differences.
not too shocking if you've read into this scene. even non-sexual parasocial interactions can have some scarily powerful effects. Add the idea that someone is forming a romantic connection and they will give you their entire paycheck.
I like JSON5 and have used it some. When GPT was younger and I was parsing its JSON output directly, JSON5 was forgiving in useful ways.
The one thing I really wish it had was some form of multi-line string, mostly so I could use it with line diffs. Also sometimes for really simple property editors it's nice to put the serialization in a textarea, and that works fine for everything but multiline strings.
(I suppose you can escape newline characters to put a string on multiple lines, but I find that rather inhumane and fragile)
"IQ is Gaussian" – it was pointed out somewhere, and only then became obvious to me, that IQ is not Gaussian. The distribution is manufactured.
If you have 1000 possible IQ questions, you can ask a bunch of people those questions, and then pick out 100 questions that form a Gaussian distribution. This is how IQ tests are created.
This is not unreasonable... if you picked out 100 super easy questions you wouldn't get much information, everyone would be in the "knows quite a lot" category. But you could try to create a uniform distribution, for instance, and still have a test that is usefully sensitive. But if you worry about the accuracy of the test then a Gaussian distribution is kind of convenient... there's this expectation that 50th percentile is not that different than 55th percentile, and people mostly care about that 5% difference only with 90th vs 95th. (But I don't think people care much about the difference between 10th percentile and 5th... which might imply an actual Pareto distribution, though I think it probably reflects more on societal attention)
Anyway, kind of an aside, but also similar to what the article itself is talking about
This is a subtle aspect of intelligence measurement that not many people think about.
To go from an IQ of 100 to 130 might require an increase in brainpower of x, and from 130 to 170 might require 3x for example, and from 170-171 might be 9x compared to 100.
We have to have a relative scale and contrive a Gaussian from the scores because we don’t have an absolute measure of intelligence.
It would be a monumental achievement if computer science ever advances to the point where we have a mathematical way of determining the minimum absolute intelligence required to solve a given problem.
> It would be a monumental achievement if computer science ever advances to the point where we have a mathematical way of determining the minimum absolute intelligence required to solve a given problem.
While that would be nice, it's likely a pipe dream :( There's a good chance "intelligence" is really a multi-dimensional thing influenced by a lot of different factors. We like pretending it's one-dimensional so we can sort folks (and money reinforces that one-dimensional thinking), but that means setting ourselves up for failure.
It doesn't help that the tests we currently have (e.g. IQ) are deeply flawed and taint any thinking about the space. (Not least because folks who took a test and scored well are deeply invested in that test being right ;)
There is no "the IQ test". The most prominent ones are Stanford-Binet and Wechsler.
That, I think is the first problem. There isn't a single agreement what IQ is or how to measure it. There isn't a single one for good reasons, because they all measure slightly different things. But that means that fundamentally any single IQ scale is likely flawed. (Wechsler acknowledges this. SB sorta does as well, but hides it well)
But if we're looking for a second at Stanford Binet :
It's hard to administer. Scoring requires subjective judgment. It's sexist. It uses language and situations that don't apply to current times. It's highly verbal. The normative sample is questionable (though SB-V has gotten better)
And because I've had this discussion before: I'm not saying IQ tests are completely meaningless. Yes, there's some signal there. But it's so deeply flawed signal that building rigorous science on top of it is just hard to impossible.
>It would be a monumental achievement if computer science ever advances to the point where we have a mathematical way of determining the minimum absolute intelligence required to solve a given problem
For a huge number of problems (including many on IQ tests) computer science does in fact have a mathematical way of determining the minimum absolute amount of compute necessary to solve the problem. That's what complexity theory is. Then it's just a matter of estimating someone's "compute" from how fast they solve a given class of problems relative to some reference computer.
You're right - we can get closer and closer to an absolute measure by looking at many brains and AI's solving a problem, and converging to maximum performance given a certain amount of hardware by tweaking the algorithm or approach used.
But I think proving that maximum performance is really the ultimate level, from first principles, is a much harder task than looking at a performance graph and guesstimating the asymptote.
IQ scores have proven highly correlated to educational achievement, occupational attainment, career advancement, lifetime earnings, brain volume, cortical thickness, health, longevity, and more.
To the point of being accurate predictors of these things even when controlling for things like socioeconomic background.
It's used because it works as a measuring tool, how the tests are constructed is largely irrelevant to the question of if the outcome of the test is an accurate predictor of things we care about.
If you think you have a better measuring tool you should propose it and win several awards and accolades. No one has found one yet in spite of many smart people trying for decades.
I'm not saying the ranking is necessarily wrong, but that turning the ranking into a distribution is constructed. And it MIGHT be a correct construction, but I am less confident that is true.
The distribution implies something like "someone at 50% is not that different than someone at 55%" but "someone at 90% is very different from 95%". That is: the x axis implies there's some unit of intelligence, and the actual intelligence of people in the middle is roughly similar despite ranking differences. That distribution also implies that when you get to the extremities the ranking reflects greater differences in intelligence.
The distribution implies that a score of 100 means you did better than half the population, and that a score of 130 means you did 2 standard deviations better than the population ie. better than 95% of other people. We have no objective measure of IQ so we use relative rankings. If you used a uniform distribution for iq everyone currently above 145 would have 99 out of 100 IQ. Normal distribution is useful when you want to differentiate points in the tails
It does seem like you should assume the accuracy of the result decreases as you get away from the norm of an IQ test, though I have no idea if it's been validated. But particularly if there are mistakes on the test questions or any kinds of ambiguity in any of the questions, it seems like you'd expect that.
Like if you have two different IQ tests and someone takes one, and gets 100, if 100 is normed to the 50th percentile, maybe you have 95% confidence that on the next test they're also getting 100 +/- 2.5. But if they get 140, that's normed to like 99th percentile, maybe your 95% confidence interval for the next test is 140 +/- 12.5. (I really don't know, I just suspect that the higher the percentile someone gets, the less confidence you'd have and mostly know stats from physical and bio science labs, not from IQ or human evaluation contexts.)
The GP is saying that IQ tests are deliberately calibrated and normalized to produce a Gaussian output, and that the input is not necessarily a Gaussian distribution of any particular quantity.
This doesn't say anything in particular about whether it's useful, just that people should be careful interpreting the values directly.
Exactly. This is a criticism of the article where it says that HR has a good reason for assuming employee performance would be Gaussian, since IQ is Gaussian.
If IQ is a good predictor of employee performance, then it does follow that employee performance would be Gaussian. It doesn’t matter that IQ was “made” to be Gaussian.
Not necessarily. A "good predictor" could still result in non-Gaussian performance for at least two reasons:
1. The prediction could be on a relative rather than quantitative basis. If IQ(A) > IQ(B) always implies Perf(A) > Perf(B), then the absolute distributions for each could still be arbitrary.
2. A "good predictor" in the social sciences doesn't always mean that it explains a large part of the variance. If IQ quantitative correlates with observed performance on some scale, but it explains only 25% of the variance, then the distributions could still be quite different. Furthermore, if you're making this kind of quantitative comparison you must also have quantitative performance measurement, whereupon its probability density function should be much easier to directly estimate.
I think IQ is useful in aggregate (for example, a finding that exposure to local toxins reduces a cities' performance on IQ by 10 points), but not useful an an individual level (e.g. you have an IQ of 130, so we can say with certainty you will earn $30,000 more per year). It's similar with MRI scans of ADHD: they find brain differences at a large scale, but you can't use a MRI to diagnose ADHD.
Individual test-retest variability is high. It's only a valid measure of anything much below 100.
Consider a test of walking speed which each time you take it gives results of (2, 3, 6, 2, 3, 5, 7, 3) etc. -- does this measure some innate property of walking speed? No.
Yet, if it were < 1, it would measure having a broken foot.
The entire field of psychometrics is pseudoscience, as is >>90% of research with the word "heritability" in it.
The levels of pseudoscience in these areas, statistical malpractice, and the like is fairly obscene. Nothing is reproducible, and it survives only because academia is now a closed-system paper mill where peer citation is the standard of publication and tenure.
A discussion of statistical malpractice is difficult on HN, consider how easily fooled these idiots are by statistics. Researchers motivated to get into psychology are not rigorous empirical statisticians, instead they are given stats GUIs into which they put data and press play. These are the most gullible lot you'll ever find in anything called science.
The world would be better off if a delete button could be pressed on the whole activity. It's a great tragedy that it continues.
If it was really “pseudoscience” you would present the experiment that demonstrates it’s obviously false rather than name calling (asserting a label with a negative connotation).
The reality is not so clear and you have to contest with decade long studies in support. Maybe those studies have flaws, but it’s not a vacuum.
I have already stated I don’t believe IQ is intelligence.
There is no experiment which proves its false. This is the problem with pseudoscience, it's "not even wrong".
Psychometrics presents summaries of data as if they are properties of reality. As-if taking a mean of survey data meant that this this mean was a property of the survey givers.
This applies only in extremely controlled experiments in physics, and even then, somewhat rarely.
All one has to do to show the entire field is pseudoscience is present a single more plausible theory than "mean of data distribution = innate property", and this is trivially done (eg., cf. mutualism about intelligence).
There is a "positive manifold" of results across test we call "intelligence tests", this is a property of the test data. What "IQ" does is take the mean of this and call it a property -- no such property exists.
Consider athleticism: across all sporting activites there's a positive correlation of ability. Call it "athleticism". But people do not have "athleticism". If you break there leg some of these correlations disappear, and some survive. Atheticism is a result of a very large number of properties of people, which arises out of highly complex interactions.
Heritability measures the correlation of traits with genes. We have ~20k genes, and we share 90% with mice, almost no genes code for traits. The vast majority of trait-gene correlations are caused by geographical (and cultural) mating patterns. So scottish accents are nearly 100% heritable, since nearly all people with one share some genes; and nearly all people without one do not have at least some of these genes. So much for "heritability" -- the use of this statistic, outside of extremely narrow biological experiments where corrlation is the result of causing genes to corrlete (by design) -- is pseudoscience.
And so 1) there is no trait "intelligence"; and 2) all claims to trait-intelligence-gene correlation are confounded by massive non-genetic factors beyond causal control.
And so: psychometrics is pseudoscience. The vast majority of its popular results are by frauds, charlatans and just plain idiots. I have no pleasant words for them: I call them by their name. Fraud is rampent, and even if it weren't, given causal-physiological semantics to factor correlations is pseudoscience.
It wasnt very hard to see, many of the citations in these famous works of IQ research were conducted under extreme causal confounders (eg., black people in aprethid south africa, people selected for the IQ test by an IQ test, etc.).
There isnt any kind of scientific research which can today establish "IQ" as a property of people --- we have no experiments which can control for known, massive, confounders. We cannot breed generations of people with deterministic genetic variation, deterministic childhoods, (etc. etc.). This kind of science is as impossible today as microbiology was to the greeks.
If you're willing to agree "IQ" is in the same realm "Athleticism" in terms of realism and heritability, then I have nothing more to say. There would be no question studying IQ is incredibly valuable.
Sure, maybe there is no part of the human which is the IQ, and it's a merely a summary of other factors being expressed. I don't think IQ researchers ever claimed otherwise.
Are you familiar with "entropy"? Isn't that a statistical summary of a configuration of atoms? Wow! Emergent properties, with no physical existence are a big part of science.
I think this would be more accurate without the "at best"; I think IQ is widely considered to be a useful diagnostic measure, misapplied to prediction in generalized populations.
I didn't know that about how IQ tests are formed.
Would that mean that there could be some sliver of the population that could score in the top %'s on the 1000 question test but due to the selection of questions, scored average on the final IQ exam?
If so, that'll be my excuse next time I have to take an IQ exam. I just got the wrong distribution.
Sum of N independent similarly distributed variables (questions), will tend to be normally distributed, that the central limit theorem, no need to manufacture anything.
Indeed. The whole premise of the activity is that they are highly correlated.
The imposition of a normal distribution is done ad-hoc at the population level. All it says is that if scores were normally distributed, then "people would be so-and-so comparable".
Almost all assumptions of this method are false.
Any time anyone mentions the central limit theorem in applied stats is a warning sign for pseudoscience. If reality existed at the end of the CLT, it would be in heat death.
> and then pick out 100 questions that form a Gaussian distribution. This is how IQ tests are created.
You missed an extremely important final step. People's scores on those 100 questions still aren't going to form a Gaussion distribution. You have to rank-order everyone's scores, then you assign the final IQ scores based on each person's ranking, not their raw score.
It would form a gaussian distribution if you pick the questions carefully enough.
If you rank-order scores and fit to the distribution after the fact, the questions are nearly irrelevant, as long as you have a mix of easy, medium and hard questions.
It's worse, because every test is obviously bounded, and it's absurd to not expect some noise to be there.
Join those two, and the test only becomes reasonable near the middle. But the middle is exactly where the pick of questions makes the most difference.
All said, this means that IQ is kinda useful for sociological studies with large samples. But if you use it you are adding error, it's not reasonable to expect that error not to correlate with whatever you are looking at (since nobody understands it well), and it's not reasonable to expect the results to be stable. And it's really useless to make decisions based on small sample sizes.
That’s not how IQ tests are made as can be found by reading how they’re actually made via Google scholar. And it would be spectacularly hard to do what you describe.
How they’re actually made is a batch of questions thought to take some form of reasoning are curated, then ALL of those questions are used in the test. It is an empirical fact the percentages of decent sized groups of people will score a bell curve, in exactly the same way humans do on hard calc exams, on hard writing items, on chess problems, and across a bewildering amount of mental tasks, none of which are preselected and fidgeted with to fake a Gaussian.
A simple example: see how many simple arithmetic problems people can do in fixed time. What do you find? Gaussian. No need to fiddle with removing pesky problem. Do reading. Do repeat this sequence for length. Just about any single class of questions has the same bell curve output in human mental ability. The curve may bend based on some inherent difficulty, say addition versus calculus, but there will be a bell curve.
Now take plenty of types of questions to address various wobbles in people’s knowledge, upbringing, culture, etc, giving a host of bell curves per category (and those also correlated by individual). Then the sum of gaussians is gausdian. All IQ tests do is shift the mean score to be called 100 (normalized) and the std dev to match a preset amount of people so such tests can be compared over time.
And the empirical evidence is these curves do strongly correlate over time, so scaling a test to align with this underlying g factor is well founded.
This latter fact, that score on one form of intelligence seems to transfer well to others, forms the basis of modern intelligence research on the g factor. IQ tests correlate well with this g factor. And across all sorts of things the results are bell curves.
For anyone wanting to hear all this and a ton more, Lex Fridman has an excellent interview with a state of the art intelligence researcher at https://www.youtube.com/watch?v=hppbxV9C63g. The researcher goes into great depth on what researchers do know, how they know it, what they don’t know, and what has been proven wrong. This is all there.
People may find that manufactured or "oh IQ is just made up and there is no measure of intelligence". But I find beauty in the way that IQ tests create and reconfigure a distribution across a multi-dimensional vector or dimensional space. It figures out what we need in the general case, and allows us to use and reason with it, without ever having to do the grunt work or arguably impossible task of finding out an actual measure of intelligence or some way to untangle the way a brain works.
That's a problem with it: its high legibility masks the complex (and deceptively muddy) math underneath it. Cosma Shalizi's "Statistical Myth" essay is a good dive into this; the "general factor" underneath all the different IQ tests is more or less a statistical inevitability, reproducing even with totally random tests.
Yes, this has always bothered me. IQ doesn't easily correspond to any measurable real-world quality.
For example, if we would postulate that height is gaussian, we could measure people's heights and just ordering them we could create a gaussian distribution. Then we could verify the hypothesis of height being gaussian by mapping the probability distribution function's parameter to a linear value (cm) and find that these approaches line up experimentally.
We could do the same thing with any comparable quantity and make an order of them and try to map them to a gaussian distribution, but we would have no knowledge if what we were making actually corresponded to a linear quantity.
This is a serious issue, as basically making any claim like 'group A scores 5 points higher than group B' is automatically, mathematically invalid.
I think your comment about an easy test having everyone in the “knows a lot” category hints that the reverse (a hard test) would be Pareto distributed.
I have given Threads a good try, and recently when Bluesky activity started up I restarted using Bluesky (it didn't stick for me the first time). The technology doesn't really matter that much, as long as it's basically competent. It's only the social network itself.
I'm not sure there's anything in any of the products that makes one better than the other (except Mastodon is actively obtuse). It's just a matter of who joins and how they interact. People on Bluesky act like people on Twitter used to, but maybe (hopefully) without as much rage-baiting. Though seeing some classic Twitter personalities translating their snarky and meta commentary to Bluesky, I'm finding it doesn't really work... the medium is exactly the same, but the vibe isn't.
Threads feels like a text Instagram, because so many of its users came from there. It can be entertaining, but it feels ephemeral, and the algorithm promotes a kind of low-brow broad content that doesn't make me feel good after consuming it. Somehow it feels like trying to make a social network out of someone else's comment thread... like it's never really meant for us.
X feels pretty shitty, not like Twitter. It's a lot of self-promotion bullshit, and doubling down on rage bait. Using it is also an expression of fealty to someone who in his vanity is actively hurting this nation. Threads isn't an expression of fealty to Zuckerberg... it's all filtered through the capitalistic process that mostly removes direct ideology. It might suck or be great, but it's not a person. X is a person. There's no way to separate the two.
Bluesky feels like what we make of it. There's not a lot of algorithm putting its thumb on the scale.
The repo doesn't say much... I thought maybe the docs would justify "world domination" in some fashion, but they are rather dry: https://haltp.org/posts/owl.html
Is there something that describes what is notable about this Lisp dialect?
I haven't looked at it, but it's purely functional (so no destructive operations such as set!). It can't be called Scheme, because it's a subset of RnRS Scheme; that's why Racket isn't called PLT-Scheme any more. I can imagine this as a teaching tool (though the FAQ says the error messages aren't good), or perhaps usable as an extension language.
I'm going to look at it as a scripting tool that compiles to C.
I used to work for a company whose internal communication often claimed “world domination” as its ultimate goal. I just looked at revenue estimates for its market sector, this company isn't in the top 5, and is far behind the leader. Let's just leave Owl's world domination goal as aspirational.
It's the other way around for Racket, where R5RS and R6RS are subsets :-)
IIUC Racket's new name came about from, basically, brand confusion: to avoid being (mis)understood as "yet another implementation of Scheme" rather than as the thing-in-itself it had become.
