HNHacker News
TopNewBestAskShowJobs

chrchang523

591 karma · joined March 8, 2017

submissionscomments
chrchang523··on US bans WeChat, TikTok from app stores, threatens shutdowns
Nobody is seriously proposing that the US ban successful Indian companies, or that India ban successful US companies (the pushback against Free Basics did not rise to anywhere near that level). This is strictly about global retaliation against PRC Internet policy that is practically forced by geopolitical game theory at this point.
chrchang523··on US bans WeChat, TikTok from app stores, threatens shutdowns
It wouldn't make a difference if only client code was open-sourced, sure.

But if the entire system was open for Americans to peruse and operate their shards of, that might be a different story. Not saying that is practical, but I've never seen an objection on nationalist grounds to the fully open-source code I wrote from China while employed by a Chinese company in 2012-14.

chrchang523··on US bans WeChat, TikTok from app stores, threatens shutdowns
If the argument was restricted to a claim that the technical details of the ban should be slightly different, you might have a point. But the Constitution is not a suicide pact. Something along these lines has been compelled by geopolitical game theory for years, as soon as it became clear that the preferred no-retaliation tactic was not going to work here. Laws will be updated as necessary to adapt to the strategic imperative.

As a Chinese person, I'll take this over internment camps any day of the week.

edit: To downvoters: I lived and worked in Shenzhen and Beijing in 2011-12. I aspired to do work there that was useful to humanity as a whole, both inside and outside of China, and it was already perfectly obvious then that I'd need to make use of my work not require any trust in the Chinese government. Not because of an existing law, but because of common sense regarding how much the rest of the world was going to continue putting up with. So I open-sourced everything that mattered, and that work did in fact go on to become trusted by professionals on both sides of the Great Firewall. Perhaps I'm still setting an unreasonable standard, but it's not one that I didn't try to meet myself.

chrchang523··on US bans WeChat, TikTok from app stores, threatens shutdowns
And whose fault is it that "all other similar apps are banned in China"?

Given that context, without overwhelming evidence that there's absolutely no connection between the Chinese government and WeChat, this is fair play, and the only valid arguments against this move are tactical rather than legal.

chrchang523··on Making the case for fewer assertions in tests
That's why both assert and expect are in that framework. You still use assert when that particular test failure may prevent other tests from working as designed, but you try to use expect for all independent "leaf" tests.
chrchang523··on Making the case for fewer assertions in tests
This has been encouraged by test frameworks for more than a decade; see e.g. https://testing.googleblog.com/2008/07/tott-expect-vs-assert... .
chrchang523··on Challenge to scientists: does your ten-year-old code still run?
Nit: implementations of Monte Carlo methods are not necessarily nondeterministic. Whenever I implement one, I always aim for a deterministic function of (input data, RNG seed, parallelism, workspace size).
chrchang523··on How Chinese Officials Hijacked My Company
Flip the script: we may want this sort of thing to be happening more, not less, on a global scale. When we want to help developing countries, this is one of the most effective ways to do so. We should just be smart and restrict this to more-values-aligned, more-development-left-to-do countries like India, Rwanda, etc.
chrchang523··on How objectivity in journalism became a matter of opinion
The candidates that a company chooses as important enough to give a job offer will be a biased subset of the population. However, employment law limits the severity of this bias along some important dimensions, and we're able to enforce this law with the help of statistical tests. The same applies to journalism and the galactic-scale sins of omission that are currently occurring.

But government regulation is only effective against individual bad actors in a democracy. When the regulators themselves can be expected to be corrupt, government regulation is a cure at least as bad as the disease. Unfortunately, I don't see much hope for the US mainstream until a "Sputnik moment" happens that makes it clear to even those gaining the largest relative advantages from the status quo that they too stand to lose big if current practices continue. And that'll be too late for my family if the Sputnik moment takes the form of Taiwan falling to the PRC.

chrchang523··on Linux founder tells Intel to stop inventing magic instructions and fix problems
I currently follow the guidance in this Stack Overflow answer by Travis Downs: https://stackoverflow.com/a/56861355/3245034 . I'm not aware of a more central source of truth.
chrchang523··on Linux founder tells Intel to stop inventing magic instructions and fix problems
Nit: most 256-bit AVX2 integer instructions (exception: multiply) do not trigger downclocking.
chrchang523··on Science is the belief in the ignorance of experts
Note that, due to "researcher degrees of freedom", statistical tests are a lot more effective at stopping you from fooling yourself than they are at stopping you from fooling others. Preregistration helps with the latter problem, but there's ultimately no substitute for actually replicating the results.
chrchang523··on Parallel Gzip – Pigz
The bgzip format (widely used in bioinformatics) supports parallel decompression, while being readable by gzip. (Compression ratio isn’t the best, though.)
chrchang523··on Slate Star Codex and Silicon Valley’s War Against the Media
"It’s not like he went to the greatest of lengths to hide his identity, Bay Area psychiatrists named Scott is a pretty small pool of people."

This is a misunderstanding. He was not trying to prevent "Scott Alexander" -> real name inferences. He was only trying to limit the rate of incidental inferences in the other direction; the primary effect of greatly increasing that rate is to harm his professional life.

chrchang523··on MIT and Harvard file suit against new ICE regulations
Yes, like most other countries, Australia does a reasonable job of enforcing its immigration laws. As you note, Australia actually isn't so geographically isolated that this is a trivial problem; and just like other forms of high-stakes law enforcement, some implementation details look unpleasant, especially when one ignores the downsides of the alternatives.

But Australia is ultimately near the top when it comes to how much sustainable opportunity it offers per capita to foreigners (and nonwhite foreigners in particular, despite past "White Australia" history, if you care about that). It is reasonable to try to do better, but that's pretty much impossible for the US until the people can again trust the government to adequately enforce whatever laws are ultimately decided on.

(I think Canada is the country in the best position to try to do significantly better today. And not coincidentally, it is a country I plausibly may emigrate to if things unravel further in the US.)

chrchang523··on MIT and Harvard file suit against new ICE regulations
I'd argue that the US actually has a national security interest in facilitating India's development, in a way that is obviously not true for China at this point.
chrchang523··on MIT and Harvard file suit against new ICE regulations
I'm guessing that they are, like much of the public, primarily against unrestricted low-skill immigration. (edit: it appears that they also object to some skilled-immigration restrictions present in the US and not present in Canada/Australia/etc.; I can certainly sympathize with that as well.) Probably for the same reason that, if one had the skills to have a good chance of being hired at a top tech company, one would want those companies to maintain a high hiring bar even though that makes it more difficult for them to get hired: it greatly increases the chance that, if they do get in the door, it's actually worthwhile.

Note that all the other major British settler colonies (Canada, Australia, you could also count NZ) switched to race-blind points-based systems decades ago, and since then have provided at least as much per capita opportunity to foreigners as the US has with much less internal political strife. The US is the corrupted outlier.

chrchang523··on Algorithms Are Now Commodities
Incidentally, I think there is at least one important area where Knuth is actually still ahead of off-the-shelf libraries as of 2020: "broadword computing". When skimming volume 4A for semi-recreational purposes a few years ago, I was surprised to find multiple "bit-hacks" which were better than anything I had found in other sources, and proceeded to use them in my own open-sourced work. Every time I've encountered a situation where bitvectors are useful, I have benefited from rolling my own implementation.
chrchang523··on Meditations on Moloch (2014)
Yeah, I think Scott loses his way at the end of that post. He is overestimating the chance that Warg Franklin’s approach is doomed; once you correct for that, it is not clear that trying to construct a values-aligned singleton is the best path forward, though I’m not opposed to research along that line.
chrchang523··on From context collapse to content collapse
For #3 at least, an environment that separated within-topic from between-topic competition for attention might be a major improvement. A large search engine cooperating with a government might be able to pull this off by making content also findable on any major all-vs.-all feed ineligible for display: a preselected topic is embedded in each search, so search results can in principle be isolated from between-topic attention competition. This would almost have to start outside the US, but if it actually works it may then see wider adoption.

(I'd expect some radicalization/"evaporative cooling" memetic hazard to remain, but it should at least decrease in intensity, since one driver of increased self-radicalization is low viability of moderate, non-attention-grabbing content relative to overstated clickbait.)

chrchang523··on Edge install undermines Microsoft’s argument that automatic updates are critical
This sounds like a reference to the Great Firewall of China. (Though, on second thought, the GFW disrupts much more than 5%.)
chrchang523··on YouTube bans Stefan Molyneux, David Duke, Richard Spencer for hate speech
Several responders have brought up the Flynn Effect. It's relevant.

As a sign of how far the Western ruling class has gone down the rabbit hole, though, note that Flynn's own take on the broader context is now beyond the pale: https://quillette.com/2019/09/24/my-book-defending-free-spee... .

chrchang523··on Clara Immerwahr: Science’s Tragic and Surprisingly Modern Heroine
He's a fascinating tragic figure. Many of his own relatives were later killed by the Zyklon B pesticide derived from his work.

With that said, I think it's appropriate to focus on Clara's life in this discussion.

chrchang523··on A simple way to get more value from metrics
Minor nit: long -> double -> long cannot introduce more rounding error than long -> double, if the same long type is at both ends.
chrchang523··on I/O library 6x faster than fmt, 10x faster than stdio.h and iostream
My experience is primarily in scientific software which has more string<->number conversions than the norm, so others' mileage may vary, but I can confirm that a lot of widely used C and C++ programs are significantly slowed down by their use of standard library I/O functions. Often written by domain experts with more than a decade of experience, not noobs.

Yes, once a particular programmer knows about the problem, they can solve it pretty quickly by switching to faster and narrower functions. But this is happening one programmer at a time. Practically everyone is still learning the wrong default, and switching away only after they've noticed they've been burned by it, and scientific programmers at least are being burned by enough other things that there's no guarantee that they'll ever notice this detail, even if they're generally very competent. (We're initially trained to not worry much about this sort of constant factor, after all.)

The lowest-hanging fruit I see is in the compiler enhancement domain. They already e.g. inspect printf-family format strings, redirect the containing function call to fputs/memcpy when the string contains no format placeholders, and when there are format placeholders a warning is printed when the number of additional arguments does not match the number of format placeholders. If we could count on snprintf(buf, bufsize, "count: %d", k) being compiled down to little more than a 7- or 8-byte memcpy followed by a call to a dedicated integer-to-string function, I think the associated gain would be far more than "0.01%".

chrchang523··on Defunctionalisation: An underappreciated tool for writing good software
Yes, I found this to be the best presentation among the several articles mentioned so far in this comment section.
chrchang523··on To get good, go after the metagame
http://gamedesignadvance.com/?p=2346
chrchang523··on You have a set amount of “weirdness points” – spend them wisely (2014)
(2014)
chrchang523··on U.S. immigration policy has been a boon for the tech industry in Canada
Seattle is too far from the Canadian border.
chrchang523··on Master of Orion
IRR Small planets could be extended the same two points per turn as other Small planets.
← PreviousPage 6 of 9Next →