lambdaphagy

I think it's fine to build NLP models with any desired property you like, including "not leaking vulgar opinions that offend the courtly manners of your society." Makes plenty of sense.

But I do wish people would be more frank and self-aware about these purposes, rather than fig-leafing it as "ethics" or "fairness". By the time you've deliberately omitting the US Congressional Record on grounds of problematicity, it's worth asking yourself whether you're making your bot polite or doing damnatio memoriae.

Bias in prediction, rather than parameter estimation, is a perfectly well established sense of the term. In particular, people doing language modeling are practically never concerned with identifiability, because you can't pick out one weight out of a trillion parameter model and say what it ought to be in the limit of infinite data.

Whenever someone uses the term "bias" in ML in anything other than its statistical sense (E[\hat y - y]), it's helpful to mentally replace it with "opinions I disagree with" and see if the argument still makes sense.

Not a teacher but know a few well.

The professional culture of teachers tends to flow downward from ed schools, and is pretty different from the professional culture of programmers. If you think of a tech company and then roughly invert the proportions of engineers vs. office managers and HR, you will have some sense of it. To first order, it will be a lot less autonomy than you're probably used to, and at a much more concrete level of direction. There is a sharp delta in the amount of paperwork, sometimes understandably and sometimes not. An admin that will run interference for you on those fronts, or simply not interfere with you, can make a big difference in QoL points. "I teach for free and they pay me to do the rest of the job [of which there is an awful lot]" is a pretty common sentiment.

I had some very good teachers for whom I'm grateful. In retrospect, they gave up a lot in order to teach. Best of luck with your decision.

I think elementary is supposed to mean "self-contained" or "using the basic methods appropriate to the field", not necessarily "easy".

Two friends of mine have cured jaw tightness by taking methylfolate.

It may not work for you, but it's cheap to try.

Another point of caution here is that CDC adjusts flu mortality rates for under-reporting in an attempt to estimate how many people actually died of the flu, as opposed to how many people have influenza listed on their death certificates.

COVID-19 mortality rates, however, are almost always reported according to narrower inclusion criteria (either positive COVID-19 dx or "presumptive COVID-19"), so there is a risk of an apples-to-oranges comparison here. One ought either count death certificates or estimate the total mortality burden, but not both.

Probably because it's hard to trust public health experts in a crisis whose very existence has already demonstrated a failure in their authority.

If I wanted to reduce the risk of acquiring or transmitting an airborne infection back in February, I would wear a mask and avoid enclosed spaces without waiting for CDC's permission. Indeed, I would ignore their advice to the contrary, which stood until April (!)

It would be really nice if we could defer to competent technocratic experts, but they earn their trust by having credible track records, not by collecting credentials.

I would find it more surprising if journalists didn't share certain demographic regularities. As an exercise, I looked up the educational backgrounds of the current NYT masthead on wikipedia. Here are the results:

Columbia

Harvard, Harvard

UMich

N/A (the art design guy)

Notre Dame

Princeton

Yale

Columbia

Tufts

Stonyhurst, Oxford

UVA

Are those results out of line with what you were expecting? I can't say that they are for me. If anything, I consider it exceptionally broadminded of The Gray Lady to hire an ND grad.

Please recall that I never claimed this was a good or a bad thing. In some sense I regard domination by elites as inevitable, but I do think it's preferable to be explicit about it, and failing to recognize "the media" as a coherent social category tends to frustrate that recognition.

The political affiliation of journalists is a well-studied topic. Polls consistently find that Democrats outnumber Republicans and liberals outnumber conservatives by wide margins, ranging from 4-1 to 20-1 [!] depending on survey methodology (https://papers.ssrn.com/sol3/papers.cfm?abstract_id=3279453). A study which compared rates of citation of various think tanks by media outlets, vs. by Congress, found that every media outlet surveyed except Fox and the Washington Times stood to the left of the Congressional median (https://www.jstor.org/stable/25098770?seq=1).

Again, I'm not making any normative claims about any of this. I'm just surprised at the resistance to taking it for granted.

OP mentioned "the most popular news outlets". I expect that there's pretty good intersubjective agreement about which entities are in or out.

It's true in one sense that there's no such thing as "the media". If "the media" is just "the set of people who write things about current events that other people read", then I, too, am a member of the media. Groovy.

But this is an isolated demand for rigor. If you look at who's employed by "the most popular news outlets" (which can be relatively objectively determined through subscription numbers), you cannot help but notice that they are a rather distinct class. They come from particular schools, live in particular cities, have particular points of view including suppositions so basic that they often do not even realize that they hold them, understand themselves to be accomplishing particular professional and social objectives through their work, and so on and so on. None of these qualifications are strictly categorical, but as a statistical matter they are so robust as to be tedious. Which is all fine, so far as it goes, but the concept of "the media" is "a thing" inasmuch as it allows you to concisely name patterns and regularities that exist in the world. And there are an awful lot of patterns and regularities to be noticed.

Incidentally, Orwell made precisely the same point (about a distributed, decentralized press that coordinates through shared cultural background and sense of what's simply Not Done) in a proposed preface to the first edition of Animal Farm. The free, decentralized, totally-not-a-thing British media declined to print it.

Sarah Constantin's essay on plow cultures and hoe cultures seems apposite here: https://srconstantin.wordpress.com/2017/09/13/hoe-cultures-a...

(Note the very first photo in the wikipedia article)

It's not as clear to me that this development is fortunate. The customs of social dance have evolved over centuries, and I'm leery of doing away with things simply because we no longer understand them.

I learned to contra dance in a groovy, gender-free group, and I found it frustrating for several reasons.

Contra dance is designed for bodies that fit together, statistically, as men's and women's bodies do, and it was far more often the case that a figure designed for a man and a woman would become physically awkward when attempted with two men.

Many of the figures involved four or eight partners, and it's honestly confusing for beginners when you lack a visual reference point for whom you're supposed to dance with next.

Dance is a pretty mystical thing, and I don't find the question "why should the sex of your dance partner matter?" to be that compelling. It matters because I want to dance with women who want to dance with men. I don't begrudge anyone for setting whatever ground rules suit them, but I wish the STRICTLY GENDERFUL, PROBLEMATIC AND PATRIARCHAL dance groups would advertise themselves a little more loudly.

There is a strongly negative relationship between women's education and fertility. Is this, in the long run, also good for the taxpayers?

Testosterone levels in the West have declined by about a factor of two over the past few decades (https://academic.oup.com/jcem/article/92/1/196/2598434). Maybe society is indeed sick.

The cognitive modules that govern dominance and hunting are completely different things, in us as in all hunting species. We don't actually treat prey as we treat conspecifics we wish to dominate. Neither do cats or mantis shrimp.

Apologies, and thank you for the correction. My comments betray a lack of intellectual curiosity and I will strive to harmonize them going forward.

I'd love to respond substantively to this post, and in particular to your figures, but hn has threatened to ban me for discussing this topic. I suppose each of us can only make of that what we will. Best wishes!

Well, the victimization surveys tell us that for every category of crime that matters, there are large disparities in base rates that (a) predate policing decisions and (b) recommend focusing on black neighborhoods, because that's where the crime is. The idea that higher arrest rates are a self-reinforcing statistical artifact has no support from the data. The notion that "crime is where you look for it" is specifically undercut by victimization surveys as well as by murder, a crime that is largely impossible to conceal from the state indefinitely. If the baseline disparities are already so high to begin with, there's hardly any variance left over to be explained by self-reinforcing patrol strategies.

Put another way, if you took the first-order strategy you'd come up with by looking at the body count, and then layered the most moustache-twirlingly racist jaywalking policy that you could think of on top of it, the two strategies wouldn't look that much different.

That's all before questioning why it is a bad thing for neighborhoods to be policed for misdemeanors. I lived in West Baltimore for years and I can assure you that my neighborhood was, if anything, chronically underpoliced.

That's not what I'm saying.

Misdemeanors are not a predictor of crime, they are crime. Further, many types of misdemeanors are quality of life issues that have outsize impacts on the people who live in those neighborhoods. Go to a town hall meeting in a bad neighborhood and see what people complain about. It's often not so much burglaries as the guys who are chronically drinking and breaking bottles on the corner at 2 am. It's not terribly enlightened to criticize the police for going after QOL violations you would never tolerate in your own neighborhood.

As for whether police attention creates feedback loops, we can check by looking at crime victimization surveys, which generally show the same patterns of racial disparities as arrest records. For that matter, the black / white murder gap has been stable at about 6-8:1 for as long as we've been keeping track, and you can't change that by selectively hassling certain people for weed.

This isn't about overfitting. Race is almost exactly the opposite problem: the signal in the data is so strong that you can't help but pick it up, even after you censor everything you can think of. If you try to predict any kind of consumer behavior from a racially heterogeneous dataset you will end up finding something that correlates with race, because practically everything correlates with race.

It doesn't uniquely pick out the Bible, except to the extent that Western civilization was shaped far more thoroughly by the Bible than by Homer.

The Bible is data. It's a collection of texts with its own [secular] field of study.

Peterson's thesis is something like the following: that the early biblical stories constitute a kind of dominant eigenvector of human cognition and experience, what you get when you iteratively retell a story over a hundred generations.

To dismiss this as anecdotes or fiction is very much to miss the point of what Peterson is trying to do.

How many generalizations about human behavior whatsoever do you think are "deterministically" true? How many as robust statistical tendencies?

"I always kept my cell neat, and they never called me Palle the Clean. I always kept out of fights, and they never called me Palle the Peaceful.. But you murder four police ONE TIME..."

What good is a social policy if you have to implement it at national scale for an unlimited time to see if it works as advertised?

Restriction of range is a potential concern, but not all studies are subject to that criticism. Bruce Sacerdote's study of Korean adoptees randomly assigned to American families found, for example, that the adoptive families represented perhaps the middle 90% of the SES spectrum. Obviously you will rarely see children adopted into the lowest percentile, but you can compare children taken from the lowest percentile to children who are raised natively in that percentile, and you typically find small or nil differences for most outcome measures in adulthood.

Plenty, of smart, dedicated people have been trying for decades to devise interventions to permanently boost IQ among children raised in poor (for the first world) conditions, and unfortunately not much has turned up. Instead, there is a familiar pattern of small pilot studies that show initial promise, but whose gains tend to wash out when the program scales up to a less than one-to-one child/child psychologist ratio, or even when the children are followed up on in adulthood.

The gains from permanently boosting IQ among low-IQ children would be so great that it might still be worth banging our heads against that wall on the off chance that we find something that might work. But we should keep in mind the base success rate of past interventions.

"Upbringing" can be a slightly misleading way of stating that summary, as most people think of upbringing as comprising environmental factors that parents can influence. But factors that parents can influence should be largely shared between siblings, and studies typically find that the effect of shared environment is quite small in adulthood. IQ researchers typically find that 60-80% of the variance in IQ is explained by genetics, and most of the remainder to "non-shared environment", a grab bag of factors not shared between siblings that we might as well call "luck".

The above is a slight simplification, but I think it's a fair summary of the current consensus among intelligence researchers, given the word count.

This is frustratingly accurate. When I was learning I wanted to tell people that I didn't speak English, only... halting Dutch with an American accent.

If.