stabbles

5 ms·

61.
by stabbles 5mo ago
Sure, in the days of Markov chains you could already generate nonsense in the style of Shakespeare, so it shouldn't be surprising you could also do the inverse. But the LLM will trigger on a typo you've made only once, and argue &
62.
by stabbles 6mo ago
After yesterday's outage they admitted that their elasticsearch index for issues/prs lost data. They seem to have changed the primary source of data in the issues and pull requests tabs (w/o filters applied) from the underlyi
63.
by stabbles 6mo ago
Is "migration to azure" or "microsoft acquisition" a cause or a symptom? I'm wondering to what extent the natural life cycle of SaaS products comes down to: the company grows, the old guard with good technical taste
64.
Tell HN: GitHub PR page no longer lists all PRs
3 points by stabbles 6mo ago|3 comments
65.
by stabbles 6mo ago
I was surprised to find that this sentence > Plan prices aren’t changing did not continue with an em-dash followed by something profound that is changing. Plan prices aren't changing -- the value you get out of it is.
66.
by stabbles 6mo ago
Other than TPUs they're also planning for 960,000 Rubin GPUs [1] which can do 33 teraflops fp64 each, so over 30 classical exaflops, and with emulation it could be more than 100 exaflops. [1] https://blogs.nvidia.com/bl
67.
by stabbles 6mo ago
Another one I noticed is "or maybe I hallucinated that" instead of "or maybe I dreamed that". Researchers will be horrified to learn that even talk about LLMs affects people's vocabulary.
68.
by stabbles 6mo ago
Oh no, LLMs threaten our individuality ⸻ what will we do?!
69.
by stabbles 7mo ago
Are grammatical errors and typos fashionable now? Reading this post it seems the anti-thesis in the LLM era is not to edit at all, but rather write down a stream of consciousness to make it "personal".
70.
by stabbles 7mo ago
My experience with RISC-V so far is that the chips are not much faster than QEMU emulation. In other words, it's very slow.
71.
by stabbles 7mo ago
You're glancing over the fact that mathematics uses only one token per variable `x = ...`, whereas software engineering best practices demand an excessive number of tokens per variable for clarity.
72.
by stabbles 7mo ago
The other day someone commented on this site that in the age of agentic coding "maintaining a fork is really not that serious of and endeavor anymore." and that's probably the case. I'm sure continuously rebasing "r
73.
by stabbles 7mo ago
That's overstating things. The biggest piece of infra is PyPI, to which uv is only an interface. They do distribute Python binaries, but that's not very impressive.
74.
by stabbles 7mo ago
https://files.worldhappiness.report/WHR26.pdf
75.
by stabbles 7mo ago
You would think that Finland's unemployment rate (10%+) would influence its ranking, but that's not the case at all.
76.
by stabbles 7mo ago
You really need dedicated types for `int64` and something like `final`. Consider: class Foo: __slots__ = ("a", "b") a: int b: float there are multiple issues with Python that prevent optimizati
77.
by stabbles 7mo ago
That was how the Mojo language started. And then soon after the hype they said that being a superset of Python was no longer the goal. Probably because being a superset of Python is not a guarantee for performance either.
78.
by stabbles 7mo ago
The TL;DR: code should be easy to audit, not easy to write for humans. The rest is AI-fluff: > This isn't about optimizing for humans. It's about infrastructure > But the bottleneck was never creation. It was always verifica
79.
by stabbles 7mo ago
I wouldn't call "ChatGPT says" an equivalent of LMGTFY. The former is people in awe with the oracle, the latter is people tired of having to look something up for others.
80.
by stabbles 7mo ago
For well-intended open source contributions using GenAI, my current rules of thumb are: * Prefer an issue over a PR (after iterating on the issue, either you or the maintainer can use it as a prompt) * Only open a PR if the review effort is
81.
by stabbles 7mo ago
Sure, naming is important, but the RSS icon was well known. It was part of the Firefox address bar.
82.
by stabbles 7mo ago
Does Nix do one layer per dependency? Does it run into >=128 layers issues? In Spack [1] we do one layer per package; it's appealing, but I never checked if besides the layer limit it's actually bad for performance when doing
83.
by stabbles 8mo ago
It's not just that, it's also a filesystem layout issue. If you install everything in `/usr` or `<venv>/lib/pythonX.Y/site-packages` you cannot have two versions / variants of the same package instal
84.
by stabbles 8mo ago
There was a nice talk recently at FOSDEM about Reactant and Oceananigans.jl: https://fosdem.org/2026/schedule/event/TYF8ZD-accelerating_s...
85.
by stabbles 8mo ago
Back in the day I had trouble convincing my C++ friends to give Julia a try, because Julia's garbage collector was a showstopper. But if you follow these performance tips related to pre-allocation, in-place mutation, avoiding temporary
86.
by stabbles 8mo ago
Ah yeah, `/dev/fd/xxx` works :) somehow thought that was Linux only.
87.
by stabbles 8mo ago
Would be nice if `curl` had something to dump headers to a third file descriptor while outputting the response on stdout.
88.
by stabbles 8mo ago
The .run file is a shell script with a compressed archive appended: MS_dd "$0" $offset $s | eval "gzip -cd" | UnTAR t Where MS_dd()
89.
by stabbles 8mo ago
Just today I found that rust-coreutils makes installing cuda toolkit impossible, related to use of `dd`: https://forums.developer.nvidia.com/t/cuda-runfile-wont-extr...
90.
by stabbles 8mo ago
Missed opportunity to call it `git ciao`

More ›