iiinova — where your digital self keeps evolving

Ragnar {Groot Koerkamp} @curiouscoding.nl · 11/09/2026

sponsored by Gibson's Finest (tm)

2201

Reposted by kamila

Amazing paper by Crick et al., in which they postulate that there are 20 common amino acids because the maximum size of a "comma-free code" on triples of ACTG is 20. I get to cite it because it turns out these codes give rise to optimal sampling (minimizer) schemes :D www.pnas.org/doi/10.1073/...

pnas.org

PNAS

Proceedings of the National Academy of Sciences (PNAS), a peer reviewed journal of the National Academy of Sciences (NAS) - an authoritative source of high-impact, original research that broadly spans...

2183

Are you aware of how oppressive and stupid the scrapers owned by LLM companies are?

lists.gnu.org/archive/html...

230

What an ignorant response. people.kernel.org/monsieuricon...

140

kamila @iczelia.net · 09/09/2026

My website is currently caving in under the galore of LLM bots. I receive 250 million junk requests a day, that's about 3000 requests per second. I am about to get very, very upset.

2332

github.com/iczelia/bzip... Why do I hate being a CS researcher and programmer? Because of people like this. They end up being coddled as if they were geniuses; when in reality, the underling reality should be much different.

4231

kamila @iczelia.net · 06/09/2026

I think that the worst part about LLMs is that people, who otherwise have no idea about the specific field that i work in, start arguing with me like a meat proxy and wasting my time.

2220

kamila @iczelia.net · 04/09/2026

en.wikipedia.org/wiki/LHA_(fi... -rping

kamila @iczelia.net · 03/09/2026

github.com/iczelia/balr... balrogg losslessly recompresses Ogg Vorbis and Opus files. Archives are typically 8-12% smaller than .ogg files and 3-8% smaller than .opus files.

github.com

GitHub - iczelia/balrogg: Demonically compacting Vorbis/Opus recompressor.

Demonically compacting Vorbis/Opus recompressor. Contribute to iczelia/balrogg development by creating an account on GitHub.

0365

kamila @iczelia.net · 30/08/2026

you also forgot citric acid

kamila @iczelia.net · 29/08/2026

Deutsche Bahn set me up like Erwin Rommel today

260

kamila @iczelia.net · 28/08/2026

Hey, I know one of the committee members :-)

010

kamila @iczelia.net · 28/08/2026

I have no hopes for academia surviving AI. not because the AI is so good, but because `scientists' are rushing to dismantle the system using it for personal gain. that said, i am still looking for reasons to jump the ship.

kamila @iczelia.net · 28/08/2026

when are we getting disclosures for reviewer comments? judging by your bio you have probably submitted to the journal who rejected our paper with slop :-).

H(iczelia.net | {APL, data compression}) = 0

000

i think that zsh is best for overall use but csh is extra swag

whoops, i should have read further in the thread.

010

but most data structures are made for and with GCs in mind to begin with

000

in a garbage collected language with conservative escape analysis you get best of both worlds: what can be proven statically doesn't contribute to the GC heap and can be strength reduced, implied ownership must be deduced so you also lose reloads in hot loops thanks to restrict.

100

also if you want to go this way you have to write your own collections, otherwise Drops e.g. for map keys or values will take up time if the map doesn't recognise that they have the same lifetime because they are in the same arena. but then the key or value can't be stolen without copy.

Perhaps. an arena must have lifetime of max(o1, o2, ...); the lifetime of objects in a gengc nursery depends on runtime behaviours. because of the long lifetime there are long pauses until deallocation with higher peak memory usage.

... at which point you have reinvented a small, brittle GenGC.

010

An arena can't promote, unless you do it manually.

200

All of these 1mil objects would be quickly - in O(1) - dropped if they were in an arena or a Cheney GC nursery. However the author didn't use either. A simple arena has lifetime drawbacks over a nursery.

100

I have a new 88x31 button! iczelia.net/webring/

6654

Rust: Dropping heavy things in another thread can make your code 10000x faster | Hacker News

> because arena allocations are soooo haaaard and no one uses them. Because it's true. news.ycombinator.com/item?id=2336.... Garbage Collection is a necessary but insufficient component of a good stack; only really achievable on the higher level as instrumenting machine code for GC is hard.

news.ycombinator.com

100

700 followers ! Welcome.

270

it looks like ai slop, there's also probably trademark infringement issues waiting to pop up.

kamila @iczelia.net · 26/08/2026

I help out with autoconf also now!

2290

Reposted by kamila

Marko Bajlovic @marko.tech · 26/08/2026

State of SaaS

673745890

kamila @iczelia.net · 26/08/2026

Java's generics are fake. No static dispatch, no monomorphisation, only erasure...

Human stupidity is less appalling than Artificial Disinteligence

040

Pangram has a very low FPR. I also invite you to read the mind-numbing paragraph below the blanked out part...

151

I doubt so. I don't actually have any useful connections that would help me find a job, sadly.

000

gabby @fullmoon.id · 22/08/2026

I am thrilled to announce that our most recent paper has been rejected from the best journal in the field thanks to low effort, small context window, and completely nonsensical AI slop of Reviewer #2.

714317

Reposted by kamila

any messaging app that asks me to restart my app to update every single time i use it is a very unserious app

2291

kamila @iczelia.net · 21/08/2026

75/25 chance of me leaving academia after a M.Sc degree.

180

kamila @iczelia.net · 19/08/2026

i can look in the uds infobib. we have everything. i checked two years ago and they have the original BCH paper from some obscure french journal from the 50s.