Yeah, well — she’s dead. She died when my mom was much younger than I am today, and my mom died a while back.

Pretty much everyone else in the Western world is getting fatter, and it’s primarily because of the poor food choices that most people make. Whether it’s fast food from McDonald’s or a frozen dinner, the result is pretty much the same.

Even the higher quality frozen dinners from Whole Foods aren’t as good for you as fresh food, but they are better than most of the stuff you get just about anywhere else.

I have a general rule in life; if I see one person making a decision I disagree with, I might be comfortable criticizing that person for their actions. When I see a bunch of people making a similar set of decisions I disagree with, I start wondering if there’s a separate driver for shared group behavior. When I see entire populations doing things I disagree with, I am very certain that I’m missing whatever is driving that behavior.

Back to the subject at hand, I suspect that the answer is a mix of cost, stress, and purposeful design. Good food is expensive while the middle class is losing buying power, huge swaths of the population are over worked (stress makes you hungry), and there are massively powerful corporations purposefully attempting to make their products more addictive in order to win an arms race for massive profits.

Blaming the individuals on the receiving end is highly misplaced when it’s happening to the entire country at the minimum.

Finally I can take NumPy seriously now that it's been published.

Peter and his former student Dan McCarthy have done a lot of work on CLV calculations.

The output of a machine learning algorithm is code that no one was clever enough to write. In light of Kernighan's law, that should give you pause.

Is there ever a right time to relax meat safety rules?

NEXP is nondeterministic exponential time, right? In that case we know that P /= EXPTIME by the time hierarchy theorem and EXPTIME is contained in NEXP. We also have a direct result that NP /= NEXP (reference given in https://complexityzoo.uwaterloo.ca/Complexity_Zoo:N#nexp).

Is there a non-paywalled version?

You get a certain number of free articles with registration.

You think it's bad now? Wait until you see what happens in 2026.

As many of you remember, the bottom dropped out of the US economy in 2008. As a result, many people delayed having children for a few years (or just decided not to). That means that the freshman class of 2026 is going to be significantly smaller than that of 2025, and that's going to shut down a large number of small and financially strapped colleges.

It's difficult to imagine exactly what's going to happen afterwards, but we can be confident that things will be different.

Hmm, that's an interesting perspective I never considered. I wonder if it gets easier to get admitted to (med/law/top undergrad) schools if there is a population drop.

Probably in 2030 we're going to see a similar but smaller hit to those sorts of schools. What effect that has is anyone's guess right now.

Not everyone on HN knows about C++11 threading. A "hello world" is of interest to some people.

Concentration of measure. If you have a quantity that depends on a large number of random variables but not too strongly on any small subset of them, it tends to behave like a constant. That's the intuition behind the law of large numbers, the central limit theorem, a bunch of concentration inequalities, and model averaging.

And, of course, there is nothing that says this would work for intermediate layers, since the sample dimensions may get there from any input.

What works is averaging similar networks and averaging your networks a lot of times.