jmoss20

> …‘‘computer science’’ is not a science and…its significance has little to do with computers. The computer revolution is a revolution in the way we think and in the way we express what we think. The essence of this change is the emergence of what might best be called procedural epistemology – the study of the structure of knowledge from an imperative point of view

-- SICP

I studied both in undergrad. They're more similar than different.

"sleeping on the streets" and "living in Brooklyn Heights" are hardly the only two options here.

To be fair, I don't think the rationale is really "if they have nothing to hide...". More something having to do with whether privacy is something that should come along with the legal arrangement of, say, an LLC.

> just that the blanket argument "X shouldn't get to be secret because I don't think they have a legitimate reason" doesn't differentiate between these two cases.

Not only these cases -- that argument won't differentiate between any cases ;-).

Better I think to make sure we really understand the arguments being made. Good chance the real argument isn't quite _that_ bad.

It's only a problem if you want to use Bayes to take some credences + evidence and prove them irrational (indep. of prior).

Bayes is great for verifying self-consistency: given some priors and some evidence, it produces exactly one set of credences. If you've somehow got a different set, you've gone wrong somewhere (and can be Dutch-booked).

What it won't give you is a full theory of rationality--but IMO this is not a problem with Bayes in particular. No theory will. There /must/ always be some free variable that prevents landing at exactly one set of credences. All theories that disagree come with very strange (and not very believable) implications.

I imagine at least part of this is that specs (and documentation generally) just suck now. It's nigh impossible to sort out which bits are CYA legalese, and which bits are "no, actually, do this or else <terrible outcome>".

That obviously isn't the problem with airplane manufacturing, and maybe not for car mechanics either. But it's totally endemic in the consumer world.

"Do not operate while driving" on car HUDs. "Do not consume if pregnant" on perfectly safe OTC medications. "Do not continue to ride after a crash" on bike frames.

It's not surprising most of this is just ignored now -- there's no information content. The documentation is nothing more than a list of things for which the manufacturer would like not to be liable, and the marginal cost of adding to that list is ~0. It will grow until we run out of room in the manual / space on the packaging.

"Store between 68 and 75 F." Or what? Is that a "must follow or else death", or a "it may reduce efficacy 0.5%" or a "we've never run a sufficiently powerful study under any other conditions, but there's no theoretical reason it should matter"? It matters quite a bit to me which!

I don't see how we can hope to have good-faith communication under such a heavy threat of litigation. I would not be surprised if /that/ turns out to be relevant to the Boeing issue, even if the rest is unrelated.

It's a matter of economy-- if you spent all day measuring obvious things you'd never get anything done.

Clearly which things you choose to measure should be a function of how certain you are about them, and how much you stand to lose if you get them wrong.

Agreed with the general point, but it doesn't apply here. Memory locality can be objectively measured (e.g. with last level cache miss counters), and parent comment is correct besides -- it's usually plain to see in the code.

There are mysterious boogiemen in performance optimization, but this isn't really one of them.

The problem is exactly that the improvements are not most of what you're paying for, at least in many locales where this sort of policy would make a difference.

How certain are you that the upper tail is not impacted just as much, if not more?

The claim here is that the military families are /so much more stable/ that it explains their kids (collectively) performing better than every other state in the country.

Obviously the military provides some sort of economic lower bound. But it also applies a pretty harsh upper bound, and has all sorts of other effects that you would expect to push the mean down.

Yes, in my experience rural areas around bases tend to be more well-off than rural areas not around bases -- the base stimulates the local economy quite a bit, if nothing else. (Otoh, the revolving door population is not great for stability.)

But FWIW I do not think the effect is even close to strong enough to explain the results in the article.

Yeah, I had a similar upbringing, and this is my feeling as well. There's a massive trade-off here, but I do feel like the military kids I knew (and run into now) are radically more adaptable and sociable.

Another upside, in retrospect: you end up getting to see, up-close, a huge range of the social/cultural/political landscape.

It's hard to demonize an outgroup much when you at times were that outgroup -- or were at least, in the abstract, some outgroup. You end up forced to confront (deep-down, maybe mostly unconsciously) the arbitrariness and...malleability of a lot of things. You end up with a lot of tolerance. I'm thankful I had that experience, even if it was at times not particularly fun.

Yeah, this is along the lines of my comment. (Am a military brat, attended some DoD schools, also attended a school in FCPS.)

The educational rigor across different districts is massive. Many military kids get to sample from it a lot of times (possibly even with a bias for FCPS in particular!); other kids get to sample it ~once. There is probably some huge reversion to the mean at play here.

This sounds familiar -- I wonder if we were there at the same time.

I grew up on military installations and attended some DoD schools. Some things unmentioned/underemphasized:

Families PCS (move) extremely often -- sometimes every school year, frequently every few years. Some places have DoD schools "on base", some do not, with students instead attend the local public schools. Some of those public schools are majority military kids, some are not.

DoD schools may have a consistent curriculum (not sure), but public schools across states/countries certainly do not. Constant moves mean students get a fractured, redundant curriculum. (Comically, I recall learning about the "Explorer" in History class no less than three times.)

Some bases are located in well-off areas with great public education, many others are not. Students might find themselves one year learning algebra, the next back to basic multiplication. Schools tend to be stubbornly inflexible and will not make accommodations on their own. Extremely attentive and pushy parents may get weak accommodations (e.g., letting students moving full grade levels up/down; something difficult to explain later), but it's rare.

Added to this is impact of constant social upheaval + stress of parents deployment, lack of lasting friendships, etc.

This is all to say -- you would not expect this population of kids to do well academically! The fact that they seem to (as measured in these tests at DoD schools) should be really surprising, and probably has little to do with the DoD schools themselves. They're after all only responsible for part (often a small part) of these kids' education.

---

My main guesses at the real drivers here are:

1. (As mentioned in the article) It's a different world on base. Parents have a massive stake in their children's behavior -- and the students know this. No one wants their parent to get an earful from their CO, and it does happen. (This is most pronounced at DoD schools, but also extends off base.)

Drug and alcohol use is exceedingly rare, due to the above + how serious an offense it is on base.

It's true also that there's a modest baseline of economic + social support. Maybe not as much as the article suggests, but it's not nothing.

2. Simple reversion to the mean. The DoD schools are full of kids with a really diverse set of educational experience. Maybe some of the good experiences are even a bit "sticky" -- habits and skill learned transferring over to new environments, maybe even bad ones. Maybe it's not surprising that that population wins vs. the baseline (where kids only get a homogenous, mostly-good or mostly-bad experience). - If the good skills and habits are "contagious", maybe DoD schools even help spread them across this population.

3. The tests are mostly measuring the lower portion of the distribution. Well-off schools will have most students clipping the top end of the measurement. Many DoD students attend those schools! (At least for a time.)

This is going to seriously amplify (2), but also (1) and other things to the extent that they improve (or remove from the sample) the worst-off students.

Because the model (=="dictionary") is 70B floats -- 280GB naively, 40-70GB aggressively quantized (which might reduce compression rate). If your file is big enough that the marginal compression win over other methods makes this space-effective, sure. But that's a very narrow case.

> The existence of even the most basic objective truth, such as 1+1=2

You might be surprised that even this is a relatively (sorry) controversial view. Many (most?) practicing mathematicians do not hold it.

Even if it were, then you get the is-ought gap -- the existence of objective / analytic / verifiable / what have you facts doesn't obviously imply anything about moral "facts".

Whoever sold them those call options, at the very least.

> It seems like almost every day I run into some Cruise or Waymo (mostly Cruise) vehicle stuck in the middle of the street blocking traffic

...really?

I've seen this maybe once or twice, total. Is there really some part of town where this is a daily occurrence?

Really?

Maybe my (large, relatively well respected state school) lectures were uniquely bad. But I found MIT lectures to be leagues ahead of what we got. They covered more content, went deeper and were faster paced. The lecturers were more talented and engaging, and the problem sets were harder and more efficient. It really was night and day.

Quantization is hardly a "finding of the open source community". (IIRC the first TPU was int8! Though the tradition is much older than that.)

The Presidio is not dense.

This does indeed improve efficiency significantly: https://arxiv.org/abs/2207.07061

Which makes you wonder what would happen if you gave ChatGPT two streams of output: one for "speech" and one for inner-monologue-style "thinking".

For most of our "hmm, let me think"'s I'm not sure what we do is significantly more complicated than that. We just get to hide the inner monologue.

What if I hook some (very hypothetical) actuators up to the rock?

Surely the /concept/ doesn't depend on us having the tech to make such actuators...

The problem with this view is Hilary Putnam's "a rock implements every finite state automata". (Nicely explained, with a rebuttal in [0].)

It's an attractive idea that some thing has agency (or consciousness, or whatever property you want to define) to the extent it does some sort of computation. But then you have to go figure out what it means for something to do that computation. That turns out to be extremely difficult; most things can be interpreted as computing most other things.

[0]: https://philpapers.org/rec/CHADAR

Social media may be global, but other countervailing factors may be local.

You most certainly were not seeing hippies endorsed by the state department in the 60s. (Though, that we look back on the hippies of the 60s fondly does say something about the direction we've charted since then.)

No, 512 still fries skylake.

It's a UX thing. The "auto-close" behavior makes the policy obvious.

And presumably the maintainers are not bound in any way to not re-open an especially good/important bug. It just makes makes non-patron support opt-in, instead of opt-out.