HNHacker News
TopNewBestAskShowJobs

benkuhn

1,366 karma · joined May 12, 2015

https://www.benkuhn.net/
submissionscomments
benkuhn··on Rooms can be as bright as the outdoors
Ah, this does seem quite plausible--my local power utility doesn't seem super competent, and my landlord doesn't seem particularly excited about maintaining my house :)
benkuhn··on Rooms can be as bright as the outdoors
While LEDs do produce some waste heat, efficient ones convert 50%+ of the input power to visible light, so it's less than you might think (compare to ~10% for incandescents, iirc). The bulb I have doesn't have a perceptible effect on my room temperature (although I guess it probably does cause the thermostat to turn on a tiny bit less often).
benkuhn··on Rooms can be as bright as the outdoors
If a 120-degree cone is good enough for you, try searching for "high bay lights"? E.g.: https://www.amazon.com/s?k=high+bay+lights

If you're willing to pay more for much better color rendering than mine, my coworker got excited about these: https://store.yujiintl.com/collections/high-cri-led-lights/p...

Note that it's 1/3 the lumens of my corn bulb, but the smaller angle means the lux may be about the same (or you might need more--I haven't tried these).

benkuhn··on Launch HN: Embrace (YC S19) – Spot and fix bugs in mobile apps
Looks really interesting!

How does it behave with slow/unreliable networks? Do you buffer events if they don’t go through immediately? What’s typical data usage for the SDK?

benkuhn··on Database IDs Have No Place in URIs (2008)
This argument seems backwards.

As far as I know, no database has ever had any of the problems he mentions with IDs not round-tripping properly during backup/restore. (Please correct me if I'm wrong! But why do I think this is true? Because any database that doesn't roundtrip primary keys would break all foreign keys on backup/restore. Databases with broken backup/restore tend to either fix it real fast, or go away.)

On the contrary, it's much easier to keep IDs stable over time, than any other field. The problem with using a different field as an identifier (e.g. a URL component) is that almost any property with an objective meaning, might change over time or become non-unique. This is why it's a best practice not to use, e.g., user email as a primary key.

For instance, Stack Overflow allows users to edit the title of a post. Right now when that happens, the canonical URL changes (to include the new slug), but the old URL is easily 301'd to the new one because it includes the ID. If the numeric ID was not in the URL, you'd be stuck with either a misleading slug, or having to maintain a list of former slugs/redirects for each post. Plus, you'd have to hack your slug-generation algorithm to ensure the generated slugs to be unique, and the easiest way to do that is... by appending an incrementing ID to the end of the slug.

The point about avoiding cruft like .aspx in URLs is well taken, but unfortunately it's in direct tension with the point about keeping them stable given that "meaningful" non-cruft tends to change over time!

(And if you don't want someone to guess your metrics from IDs, you can use a random non-numeric ID instead!)

benkuhn··on Software Projects and Heroes: Lessons Learned from GitHub Projects
Everyone who's interpreting this in the context of full-time work / the traditional notion of "10x programmer" (i.e. ~every toplevel commenter) should read the abstract:

> A "hero" project is one where 80% or more of the contributions are made by the 20% of the developers... We identify the heroes developer communities in 1100+ open source GitHub projects.

(emphasis mine)

Of course nearly every Github project is going to have heroes--we call them "maintainers." This study is totally unsurprising and totally non-generalizable to professional teams working on proprietary software.

The use of the term "heroes" in the abstract is massively misleading and tendentious, IMO.

benkuhn··on QUIC and HTTP/3: Too big to fail?
> the median global network and device are both quite slow. This means a lot of the network gains from QUIC are potentially (largely) undone by the slower hardware.

The rest of the article made reasonable points but this is a really bad argument.

The cited claim that congestion control was application limited 50% of the time used bandwidth of either 10Mbps or 50Mbps. Meanwhile, the median mobile device is on a low-end (~15th percentile) 3g connection, I would guess around 1Mbps or less; 45% of devices were on 2g in 2017 according to the author's own citation.

I'm writing this from Senegal, currently on great internet with a 350ms RTT; mobile RTT often hits 1s. At those speeds I really couldn't care less how much time my phone spends decrypting if you can save me roundtrips!

benkuhn··on HTTP-over-QUIC will officially become HTTP/3
It looks like the opposite to me?

TCP minimum header is 20 bytes.[1] QUIC "short header" (after key negotiation) looks like it can go down to 2 bytes for non-roaming connections with a 1-byte packet number, if the server allows the omit_connection_id, and 10 bytes otherwise.[2] Add the UDP header (8 bytes[3]) and you're at 10-21 bytes. So if you both supply a connection ID and a 4-byte packet number you're just barely at the same transport overhead as TCP; otherwise it's less.

(The QUIC "long header" is longer but that's a fixed cost and shouldn't matter too much for overhead.)

Or am I missing something?

[1]: https://en.wikipedia.org/wiki/Transmission_Control_Protocol#... [2]: https://tools.ietf.org/html/draft-ietf-quic-transport-08#sec... [3]: https://en.wikipedia.org/wiki/User_Datagram_Protocol#Packet_...

benkuhn··on Small company or big company?
Thanks! This is fair and useful criticism. I will update the post.

In the case of the particular person I originally wrote this for he wasn't looking at FAANG, and my impression is that most large companies are not competitive with them. In your experience, is that part still accurate?

benkuhn··on Macbook Pro frying USB peripherals
I have this problem often when I’m in Africa. Sometimes the case shocks me very lightly as well. I’ve always assumed it’s a problem with lack of grounding and/or dirty power. (Had it with non-macs as well)
benkuhn··on Developer Survey Results
It's a little disappointing to see so many developers--maybe on average the most comfortable job in the world right now--prioritizing comforts like compensation, using their favorite languages, or remote work, over helping to improve the world.

Over 45% of men prioritized one of the above three, whereas a mere 6.5% prioritized "How widely used or impactful the product or service I'd be working on is." Revealingly, "How much impact I could have by working there" isn't even an option. Sounds like the survey authors didn't even consider that people might actively care about improving the world.

So many developers have comfort to spare! Maybe spend a little bit more optimization power on trying to accomplish something important?

benkuhn··on Dustin Moskovitz pours funds into high-risk research
For anyone else who was confused by the headline: it's "high-risk" research in the sense of "speculative" or likely to have null results, not in the sense of e.g. the lethal virus research that was on here a few days ago.
benkuhn··on Unsupervised Machine Translation Using Monolingual Corpora Only [pdf]
I think you missed something.

Previously, machine translation required being trained on a bilingual corpus, that is, a corpus of the same set of sentences in eg English and French. These corpora are pretty hard to come by and expensive to produce.

The paper describes a technique to use two monolingual corpora instead, i.e. one set of sentences in English and a different set of sentences in French. That's way easier to find.

It's far from just a definitional trick.

benkuhn··on Tesco Bank halts online payments after money was taken from 20K accounts
Wait, that's still awful! It allows you to crack each character individually. For instance, a 10-letter password requires 26^10 ~= 1.4e14 attempts to test every option if you only have a hash of the full password, but only 10*26 = 260 attempts to test every option for every individual character.
benkuhn··on Machine Learning and Ketosis
I mean, the uncertainties in the middle are less interesting than the uncertainties at the extremes. I'm sure you can't say anything useful about melon, but it would be interesting to know if (e.g.) "sleep" was consistently that big. He did regularize somewhat (`--l2 1.85201e-08`) but unclear whether that's enough regularization.

Basically I'd love to see some actual diagnostics I guess :)

PS I believe the "relevance" is just the coefficient divided by the biggest coefficient (the `RelScore` column in the printout in the README file).

benkuhn··on Machine Learning and Ketosis
It's astonishing (and really awesome) that they were able to extract such a strong signal from daily weight swings! This makes me a lot more optimistic about the possibilities of quantified-self-type stuff.

Three things I'm really curious about:

- How significant are the estimates of lifestyle factors? Do you have p-values? If you bootstrap resample, how much do the rankings change at the extremes?

- How much cognitive overhead did it impose to collect the data for this? Did you put a lot of effort into designing the tags beforehand or making sure you weighed yourself at a consistent time?

- It looks like the predicted delta from going from a "nosleep" day to a "sleep" days is about 1.4 pounds (sleep coef minus nosleep coef). That seems fishy, or at least like it will stop working fairly soon because you can't actually lose 1.4 pounds/day sustainably. Is it possible there's something weird going on with the data or those variables don't have the obvious meanings?

benkuhn··on We built voice modulation to mask gender in technical interviews
I would really have liked to see them check whether the gender-blinding actually worked. Even if someone's voice is modulated, they may have gendered behavior patterns that could influence someone's performance ratings. I wouldn't be surprised if I could guess the gender of a voice-modulated person pretty accurately from other cues.

To combat this, they could have asked the interviewer during the performance assessment whether they thought the applicant was definitely male/probably male/unsure/probably female/definitely female. Then you could use voice modulation as an instrument for perceived gender and get a better estimate of the true effect of perceived gender when controlling for actual gender.

Hope they do something like that for the next round of experiments!

benkuhn··on I apparently got 50% better at my job last month
Thanks for pointing this out; I think "procrastination" was a lazy label for that time category on my part and doesn't actually reflect what I'm doing.

As I mentioned in a parent (cousin-once-removed?) comment, the RescueTime statistics capture every moment I spend in front of my computer, not just when I'm in work mode. My actual work pattern is more like

- Wake up

- Bike 2 minutes to my office

- Work for 9-10 hours

- Bike home

- Do the things that I labeled "procrastination" (mostly reading things on the Internet)

This leaves me working a pretty normal schedule. I actually find I'm able to focus very well at work, since I'm really excited about my job and have a private office that screens out most distractions.

benkuhn··on I apparently got 50% better at my job last month
I evidently need to make this more clear in the article, but RescueTime captures all time I spend in front of the computer, not just work time. My actual work time is about 50h/week, which is probably about average for software people in my area. In fact, it's probably on the low end if you factor in that everyone else spends 10h/week commuting into San Francisco.

I find my job incredibly exciting and rewarding, so I don’t find that amount of work burdensome. YMMV, of course :)

benkuhn··on Why squared error?
Thanks for the kind words! I actually agree that L1 error is often more practical, and didn't mean the post to come off as prescriptive. (In fact, part of my motivation for writing it was that L1 error seemed to have much better practical properties, and I was curious why people used L2 despite this!)
benkuhn··on Does donation matching work?
> As an aside, I think people don't fully consider the effectiveness of earning more money as a form of altruism. Since around 30% will be taken by the government, and used to fund various social services, the choice to earn more money is really the choice to earn money and donate 30%.

I'm super sympathetic to this idea, but I think generally there are even more effective places to donate extra earnings than the government! For one thing, a lot of government programs aren't very efficient for various political reasons. For another, even the worst-off people in the US are largely better off than the people that you could help by giving to organizations that work abroad. (Personally, I'm a fan of GiveWell's[1] approach to finding the most effective places to give.) But I agree that trying to earn money in order to fund this sort of thing is an under-appreciated idea, and in fact exactly what I'm trying to do :)

[1]:http://www.givewell.org/

benkuhn··on Does donation matching work?
Yeah, I definitely think donation matching is generally a good deal from the charity's perspective.

When I wrote this I was more thinking from the perspective of someone who has a large fixed donation and is deciding what to do with it. In that case, they might want to back a matching campaign, but if they're trying to maximize the impact of their donation, they would want to know whether such a campaign would actually draw out more donations.

← PreviousPage 2 of 2