HNHacker News
TopNewBestAskShowJobs

john-h-k

384 karma · joined March 23, 2025

https://johnk.dev
submissionscomments
john-h-k··on The language brain matters more for programming than the math brain? (2020)
I agree, but also via some napkin math, the chance of getting those different results is something like 40%.

(Ie, if you sample the same signal twice for the numbers they did in the study, there’s a 40% chance it’ll be >0.04 away from the original sample, as numeracy was from language)

john-h-k··on The language brain matters more for programming than the math brain? (2020)
> I certainly don’t dispute the empirical validity of the findings from the study

You absolutely should. Tiny sample size and poor statistical method. It is p-hacking plain and simple

john-h-k··on The language brain matters more for programming than the math brain? (2020)
Has anyone here actually read the paper? Heavily clickbait headline, completely exaggerating the study results, which themselves are not very rigorous (to be generous)

Tiny sample size - 36 people completed

Numeracy has R^2 = .27 Language has R^2 = .31

They then run stepwise regression to determine variance contributions, seemingly ignoring their earlier results, and this leads to almost no contribution from numeracy. Why? Because they have ~10% shared variance and stepwise regression is greedy - will just take whatever you give it first.

I can't mention this part enough. If you got a second, very similar language test, and added it to the model, you would _also_ find it has almost no unique variance added.

Every thing they measure is incredibly noisy and they do not once attempt to deal with this. Human based reviewers, time-to-completion, etc.

p-value for "language learning is more significant than numeracy" on the values they give (Steiger test) gives 0.772. Utterly insignificant.

Also, to beat the point home, just think about the argument here: * Numeracy contributes 27% of variance

* Language skills contribute 31% of variance

* After regression step, numeracy contributes only 2% of unique variance. Because you added a correlated variable!

john-h-k··on Minimum Viable Blog
Thats an awesome setup and very similar to the one I have. But damn triple-digit-ms does actually sound shockingly slow for what is effectively just heavy text processing. Is there a huge amount of text?
john-h-k··on The Brief Origins of May Day
> If you sell your time for money you are working class.

This would put a trader earning $3m/yr as working class. That’s fine I guess, but then we need another distinction, for the difference between that person and a minimum wage worker.

Now in all the conversations where we would say “working class” we use this new term, because we almost always mean things relating to income level rather than labour vs owner.

And so all we’ve done is changed the name for “working class”

john-h-k··on Mac app launches slowed by malware scan (2024)
I’ve got a personal project compiler I built and it’s hit by this very hard. Testing involves (naturally) generating lots of executables. Running it in a Linux docker container takes around ~1s for all 500 tests. macOS by default takes around a _minute_, and even with the workarounds I’ve found (“allow untrusted software to be run by iterm2”) it takes 5-8 seconds.

It’s a pretty niche use case but it’s deeply frustrating

john-h-k··on Shoes on at home or shoes off? If you care about your health, it's a no-brainer
I was mid way through my evening floor-lick, took my normal 5m HN break, and I see this. Will make sure people take their shoes off!
john-h-k··on Programming languages should have a tree traversal primitive
Excellent time to mention rust's `ControlFlow` enum which allows doing this exact sort of thing https://doc.rust-lang.org/std/ops/enum.ControlFlow.html

Not as ergonomic as a direct tree-iterator, but I can't see of an elegant way to introduce that in an imperative language while keeping the forking/recursion aspect clear

john-h-k··on Ask HN: CS degrees, do they matter again?
I don't think is apple/google. This sounds like the kind of thing small companies do. I would be very surprised if that was FAANG

I am not a fan of leetcode interviews myself, but I don't think the FAANG process is that bad. I am interested in any specific examples of authors of projects being rejected from big-tech companies for a role where they would work on their own thing

john-h-k··on Ask HN: CS degrees, do they matter again?
He wasn't rejected for a role working on homebrew. He was rejected from google for an unrelated role

> answer by himself: https://www.quora.com/Whats-the-logic-behind-Google-rejectin...

john-h-k··on Show HN: I built a hardware processor that runs Python
I can't tell if this is joke but will assume not. It's because the .net version is needed for some reason. There are not processors that run .net bytecode, primarily because they would be slower and worse (and again, don't exist)
john-h-k··on Ask HN: CS degrees, do they matter again?
Is this referring to homebrew? On the one hand i get it, but on the other hand reversing a binary tree isn’t some crazy leetcode grind. Most people can figure it out in a few mins and i think its fair to reject someone who can’t
john-h-k··on Show HN: I built a hardware processor that runs Python
The tl;dr (I spent lots of time investigating this) is that it just fundamentally isn’t a good bytecode for execution. It’s designed to be small on disk, not hardware friendly
john-h-k··on Show HN: I built a hardware processor that runs Python
I’d be willing to bet my net worth that they don’t
john-h-k··on Computer Architects Can't Find the Average
I don't think it is an argument in favour of it.

arith-mean = E[x] , the first moment of x geo-mean = exp(E[log x]) , so log geo-mean = first moment of log x

They are both equivalent in amount of information preserved, but arithmetic preserves additive structure whereas geo preserves multiplicative structure

john-h-k··on Still standing
Absolutely. Point is we have to judge it in context. People saying horrible things for no reason on a message board is fundamentally different to causing a stampede in a cinema.

I agree that we need laws to make "yelling fire and causing stampedes" illegal. I do not agree that "free speech requires an intent to express a point of view" is the correct way of implementing this.

john-h-k··on Still standing
> To practice free speech requires an intent to express a point of view

Excellent, now we can ban speech we don’t like by just saying it doesn’t actually express a point of view

john-h-k··on OpenVSX, which VSCode forks rely on for extensions, down for 24 hours
Strongly agree. There's a new page on the site about helix-mode in other editors (although there are a limited number of entries): https://docs.helix-editor.com/master/other-software.html

I myself have built zsh bindings for helix as that was my main pain point: https://github.com/john-h-k/helix-zsh

john-h-k··on OpenVSX, which VSCode forks rely on for extensions, down for 24 hours
Try the helix tutor itself. Install, then do `hx --tutor` (binary is also called helix rather than hx on a few package managers).
john-h-k··on OpenVSX, which VSCode forks rely on for extensions, down for 24 hours
Lots of vim/emacs mentions so I feel obliged to mention Helix (https://helix-editor.com/). Used neovim for _years_, tried Helix for a few weeks and never looked back
john-h-k··on Is 1 Prime, and Does It Matter?
Can you elaborate on this?

My understanding is you can specifically and formally define the natural numbers with addition and multiplication, although multiplication means the language is no longer decidable.

You can define natural numbers with just addition ( Presburger arithmetic ) and it’s decidable.

Im not sure how undecidable <=> “will define things that are similar to natural numbers but are not” but maybe I am missing something

john-h-k··on What do I think about Lua after shipping a project with 60k lines of code?
I fork the editor and write them in myself. This way I can boast about my tiny config while ignoring the tens of hours of hacking the editor itself
john-h-k··on Googler... ex-Googler
If someone much richer and with a better career than me got punched in the face i would still see that as suffering. I don’t know why someone’s situation being better means they can’t be suffering
john-h-k··on Bored of It
Maybe the tool knows nothing. But it allows me to learn niche things often much faster than via a web browser. So it has to value for me.

I think there’s lot of dangers and problems with it and frankly I’d probably be happier if it was never invented. But even then I can still see the value it has

john-h-k··on Span<T>.SequenceEquals is faster than memcmp
Yes, this is correct. The span itself - the (ptr, len) pair - is on stack (by default) but the data is almost always on the heap, with stackalloc being the most notable exception
john-h-k··on A Debugger is a REPL is a Debugger
`b line` and `expr expr` both work fine in lldb. I agree the ergonomics are less excellent, but that doesn't seem a reason to completely abandon lldb/gdb to me. Maybe I am missing something
john-h-k··on What went wrong with the Alan Turing Institute?
Avoiding conflicts of interest is key, but lacking interests is worse imo. Having someone with experience in the role has value and I generally think his AI policy is excellent
john-h-k··on Watchdog Monitors AI Companies Breaking Safety Promises
AI safety shouldn’t stymie the obvious value of AI, but the claim that AI safety is utterly irrelevant seems just as counterproductive as claiming we need to stop all AI research right now
john-h-k··on Problems with the heap
as there seems to be some confusion, this is my interpretation:

atop is (for some reason) touching memory of processes it monitors.

atop is touching this in an insecure way. An executable can cause atop to corrupt its memory.

this has high potential (although not guaranteed) for allowing RCE within atop via a correctly crafted process that atop monitors.

atop is often run in root, and so this otherwise meaningless RCE becomes privilege escalation, which is bad

either this is correct, or cunningham’s law will bring out the correct interpretation

john-h-k··on Optimizing by 1700x by not being silly
> yes, we design some pieces of RavenDB with CPU microcode for performance

This is… Oddly worded? Unless I’m missing something major (possible), they just mean “writing assembly with microcode in mind”, which is pretty standard for anyone writing optimised assembly, but they’ve written it as if they’re etching new microcode into the CPU

← PreviousPage 6 of 7Next →