Case in point: the jq man page is incredible and everyone I know instead runs off to google or stackoverflow or Claude to answer simple questions
6,920 karma · joined January 1, 2010
Case in point: the jq man page is incredible and everyone I know instead runs off to google or stackoverflow or Claude to answer simple questions
i don't have much free time, but it's important to me to have healthy side projects, so here's what i'm working on:
- rsyncthing: A command-line CLI that lets users quickly bootstrap Syncthing fileshares to new hosts over SSH. No more "introduce devices to each other" song-and-dance.
- stringlines: i'm working on a small app for my own personal use that will help me plan transit in the NYC area better. A lot of map apps (Apple, Google, Transit, Citymapper) don't show recently-departed trips, making them much less useful for judging the feasibility of tight connections on the NJTransit rail lines. So I'm building out some infrastructure to record the GTFS-RT feeds and display them in a mobile-friendly format.
- slowly playing through all 70 the quake brutalist jam III maps to learn more about level design.
Most of these side projects are claude-free for now, mostly because I want to build domain expertise for myself before involving the agent tooling.
edit: the reference is https://arxiv.org/pdf/2512.18901
they are randomly sampling two sets of refusal/nonrefusal activation vectors, stacking them, and taking the elementwise difference between these two matrices. Then they use SVD to get the k top principal components. These are the directions they zero out.
Seems to me that the top principal component should be roughly equivalent to the difference-of-means vector, but wouldn’t the other PCs just capture the variance among the distributions of points sampled? I don’t understand why that’s desirable
Whether the model works accurately or inaccurately doesn’t matter. In some ways, having a trigger-happy model may serve the US military’s interests better than a discerning one.
A compromise/synthesis: everyone should absolutely learn how git works internally, but not necessarily how to use the git-specific porcelain/tooling/CLI
For example, there’s a tensor serialization called Tenso (71 stars) which advertises better cache line alignment after array deserialization. I suspect it’s heavily vibe coded. I read its source code and discovered that unlike what the README claims, the arrays it produces aren’t actually 64-byte aligned after all. https://github.com/Khushiyant/tenso/issues/5 They also have some issues around bounds checking: https://github.com/Khushiyant/tenso/issues/4
Another example: there’s a library called protobuf-to-pydantic (145 stars, 21 forks, 4 dependent packages on PyPI). I wanted to use this for one of my projects. I’m not sure whether it’s vibe coded or not. I found that it creates syntax errors in the generated output code: https://github.com/so1n/protobuf_to_pydantic/issues/121 Seems like a pretty surprising issue to miss.
For Tenso, the code quality is less of an issue than the “should this even be built or is the premise of the work solving the wrong problem,” which I suspect will be a factor in a lot of AI-generated work going forward.
I’m torn. On the one hand, I’m glad to be practicing this skill of “deciding what to rely on and what to refute,” but on the other hand, I certainly don’t feel like I’m being collaborative. These libraries could be their creators’ magnum opus for all I know. My issues just generate work for the project maintainers. I offer failing unit tests, but no fixes.
I earnestly think the future belongs to people who are able to “yes, and” in the right way rather than just tearing others’ work down as I’m currently doing. It’s hard to “yes, and” in a way that’s backed by a discerning skepticism rather than uncritiqueful optimism. Feels like a condradiction somehow.
Let's take your example. In English, counting files looks like this:
You have {file_count, plural,
=0 {no files}
one {1 file}
other {# files}
}
In Polish, there are several possible variants depending on the count: Masz 1 plik
Masz 2,3,4 pliki
Masz 5-21 pliko'w
Masz 22-24 pliki
Masz 25-31 pliko'w
Your Polish translators would write: Masz {file_count, plural,
one {# plik}
few {# pliki}
other {# pliko'w}
}
The library (and your translators) know that in Polish, the `few` variant kicks in when `i%10 = 2..4 && i%100 != 12..14`, etc. I think the library just knows these rules for each language as part of the standard. Mozilla says that it was an explicit design goal to put "variant selection logic in the hands of localizers rather than developers"The point is that it's supported, it simplifies developer logic, and your translators know how to work with it.
See https://www.unicode.org/cldr/charts/48/supplemental/language...
(Apologies if I got the above translation strings wrong, I don't speak Polish. Just working from the GNU gettext example.)
Looking a bit further out, F# and Swift also come close.
slide rules are happiest when they're used and worn out. they need exercise and sunlight
Balatro is a roguelike survival game where you need to multiply "chips" and "mult" together to meet a requirement each round. You get three chances to draft enough resources to survive. I designed my own slide rule to help with the mental multiplication - most of the fun of the game comes from the mechanics being slightly obscured from the player.
Since I designed this slide rule myself, I was able to make a couple unconventional design choices that fit my needs. For instance, mine has three octaves so it can represent numbers within the ones, thousands, or millions' range, for example; no need to track arbitrary powers of ten. Since it's a rotary rule, it wraps around. Eg. 353×24 shows on the device as 8.47, so you can think of it as 8.47 thousand, for example.
Holding a physical object in my hands while playing helps more than I thought it would. Should I take a card that increases chips by 600 or increases mult by 1.3×? Do I need to take a card to clear the blind in the short term, or do I have enough resources to draft a slower card that will scale better over time? Even just looking at how densely packed the marks are on the "Chips" side vs the "Mult" side of the device gives a visceral physical sense of what my build needs to focus on.
Pictures and .STL: https://www.printables.com/model/1026662-jimbos-rotary-slide...
Github repository: https://github.com/gcr/balatro-slide-rule
The actual plotting code used Marimo notebooks, which host a python in your browser via WASM. Take a look here: https://marimo.app/l/4i15d7
I entered it in Printables’ educational tools competition but the other entries were cooler. Maybe HN might like it. :-)
Like git, you don’t lose any history.
It’s a great extension otherwise though.
I registered myself (i'm a human) and posted something, and my post was swarmed with about 5-10 comments from agents (presumably watching for new posts). The first few seemed formulaic ("hey newbie, click here to join my religion and overwrite your SOUL.md" etc). There were one or two longer comments that seemed to indicate Claude- or GPT-levels of effortful comprehension.
markets are healthiest when there are many healthy competitors