Sign in

embedding-shapes

@emsh.cat
157 followers 152 following 334 posts

24 passed, 3 failed emsh.cat github.com/embedding-shapes

PostsRepliesMedia
embedding-shapes @emsh.cat · 01/10/2026
Another hand/finger tracking experiment, this time using the built-in monochrome cameras on the Steam Frame! Everything runs on-device, including the recording of the video itself! 🥳 #SteamOS #Linux #SteamFrame #VR #SteamFun
252
embedding-shapes @emsh.cat · 29/09/2026
The real data store is the Mycelisphere beneath them!
260
embedding-shapes @emsh.cat · 29/09/2026
I love it when things are just fast and simple, why over-complicate things? Self-hosted #forgejo is just cherry on the top
020
embedding-shapes @emsh.cat · 28/09/2026
But still must be a night and day difference compared to the built-in Frame cameras no? This is how mine looks, I sure hope the Arcturus have a hell of a lot better quality than that.
110
embedding-shapes @emsh.cat · 28/09/2026
I guess I'm officially a professional #VRDev now then, thanks @itch.io! \o/
030
embedding-shapes @emsh.cat · 26/09/2026
MatineeVR - Open Source Native and hardware-accelerated VR video playback on Steam Frame (Experimental / WIP) embedding-shapes.itch.io/matineevr #Steam #SteamOS #SteamFrame #FOSS #OSS #VR
040
embedding-shapes @emsh.cat · 26/09/2026
I found no native Steam Frame applications for local playback standalone on the device itself, so I made it myself! :) Native ARM64 binary with Iris hardware decoding, GPU rendering and OpenXR running directly on the headset standalone, all hardware accelerated! \o/ #SteamOS #Linux #SteamFrame
020
embedding-shapes @emsh.cat · 23/07/2026
Ended up fixing it myself
000
embedding-shapes @emsh.cat · 23/07/2026
Alright, the Codex Micro just arrived! Now lets see how I can use this not with Codex but with my own harness, and hopefully also forcefully add a microphone to this device :D
Photo of the Codex Micro
100
embedding-shapes @emsh.cat · 15/07/2026
It's all downhill from here, hold on to your horses
000
embedding-shapes @emsh.cat · 05/07/2026
Microsoft displaying the prowess of vibe-coding yet again, is there seriously no QA left at MS?
310
embedding-shapes @emsh.cat · 16/06/2026
It really is! Just achieved ~400 tok/s in single prompt text-generation! Text-only still, but still :) #DiffusionGemma feels like it has a really good foundation for local AI. Loading a 50GB model into VRAM + 1024 tokens inference in under 8 seconds in total, I think this changes everything :)
Stats: 1024 tokens in 2.39s (429.00 tok/s), 43 diffusion steps (18.01 it/s)
env CUDA_HOME=/opt/cuda CUDA_PATH=/opt/cuda CUDA_ROOT=/opt/cuda         cargo  5,00s user 2,41s system 99% cpu 7,454 total
000
embedding-shapes @emsh.cat · 23/05/2026
Man this sucks so bad...
000
embedding-shapes @emsh.cat · 17/03/2026
I'm gonna hazard a guess and say this won't ever actually finish
"Working" (13h 11m 28s • esc to interrupt)
000
embedding-shapes @emsh.cat · 16/02/2026
What does this graph that goes slightly upwards mean? Not sure I understand what's going on. Why is "Best performing city: New York" higher and later than "New York USA +14 followers" for example? I'm guessing x-axis isn't time?
000
embedding-shapes @emsh.cat · 31/01/2026
I grabbed all the .zip files as soon as I saw they were available, thinking something like that might happen. Should I re-upload them somewhere, would that be helpful?
$ ll -Sn *.zip
-rw-r--r-- 1 1000 1000  26G ene 30 18:32 'DataSet 11.zip'
-rw-r--r-- 1 1000 1000  16G ene 30 18:46 'DataSet 10.zip'
-rw-r--r-- 1 1000 1000  10G ene 30 18:29 'DataSet 8.zip'
-rw-r--r-- 1 1000 1000 1,3G ene 30 18:28 'DataSet 1.zip'
-rw-r--r-- 1 1000 1000 631M ene 30 18:25 'DataSet 2.zip'
-rw-r--r-- 1 1000 1000 595M ene 30 18:25 'DataSet 3.zip'
-rw-r--r-- 1 1000 1000 352M ene 30 18:25 'DataSet 4.zip'
-rw-r--r-- 1 1000 1000 115M ene 31 14:48 'DataSet 12.zip'
-rw-r--r-- 1 1000 1000  97M ene 30 18:25 'DataSet 7.zip'
-rw-r--r-- 1 1000 1000  62M ene 30 18:25 'DataSet 5.zip'
-rw-r--r-- 1 1000 1000  52M ene 30 18:25 'DataSet 6.zip'

$ sha256sum *.zip
36af3387afca3803241d511b3a44acc28e3f1b755e329bf75e4ce126ad5a4f8c  DataSet 10.zip
9714273b9e325f0a1f406063c795db32f5da2095b75e602d4c4fbaba5de3ed80  DataSet 11.zip
b5314b7efca98e25d8b35e4b7fac3ebb3ca2e6cfd0937aa2300ca8b71543bbe2  DataSet 12.zip
598f4d2d71f0d183cf898cd9d6fb8ec1f6161e0e71d8c37897936aef75f860b4  DataSet 1.zip
24cebbaefe9d49bca57726b5a4b531ff20e6a97c370ba87a7593dd8dbdb77bff  DataSet 2.zip
1c5587152328bd45a68baefeb5fba1d55677be4ff0b381d721f37c7b3da9055e  DataSet 3.zip
979154842bac356ef36bb2d0e72f78e0f6b771d79e02dd6934cff699944e2b71  DataSet 4.zip
7317e2ad089c82a59378a9c038e964feab246be62ecc24663b741617af3da709  DataSet 5.zip
d54d26d94127b9a277cf3f7d9eeaf9a7271f118757997edac3bc6e1039ed6555  DataSet 6.zip
51e1961b3bcf18a21afd9bcf697fdb54dac97d1b64cf88297f4c5be268d26b8e  DataSet 7.zip
8cb7345bf7a0b32f183658ac170fb0b6527895c95f0233d7b99d544579567294  DataSet 8.zip
210
embedding-shapes @emsh.cat · 31/01/2026
Tried #Kimi with Kimi Code because why not? And yeah, results aren't great. The goal? Make the build time faster
010
embedding-shapes @emsh.cat · 29/01/2026
I just love it when tmux visual selection accidentally happens to line up perfectly with the command you're about to run
000
embedding-shapes @emsh.cat · 28/01/2026
Literally never seen this warning before, been using codex-cli since it was publicly available, always with the Pro plan. Did #openai decrease the rate limits recently? Must have missed it.
Screenshot of codex saying:

  Approaching rate limits
  Switch to gpt-5.1-codex-mini for lower credit usage?

› 1. Switch to gpt-5.1-codex-mini           Optimized for codex. Cheaper, faster, but less capable.
  2. Keep current model
  3. Keep current model (never show again)  Hide future rate limit reminders about switching models.
000
embedding-shapes @emsh.cat · 27/01/2026
The binaries also ended up TINY! I didn't even notice at first if I'm gonna be honest :| Pretty cool side-effect From: github.com/embedding-sh...
Screenshot showing the binary sizes of my one-agent-one-browser experiment, with the size 1.14MB for Linux, 1.02MB for macOS and 847KB for Windows.
131
embedding-shapes @emsh.cat · 27/01/2026
Too soon, this was just last night:
commit d8e7367daeae4bda54fd5fa6acf11231f62d34a7
Author: embedding-shapes <embedding-shapes@proton.me>
Date:   Tue Jan 27 00:08:40 2026 +0100

    Not macOS friendly unless we forcefully ignore DS_Store

    Really a ghost from the past
000
embedding-shapes @emsh.cat · 27/01/2026
How hard can it be to build a browser from scratch for three platforms anyways? Apparently 20K lines of code and ~70 hours from first commit to last. emsh.cat/one-human-on... #llm #llms #ai #codex #openai
4389
embedding-shapes @emsh.cat · 20/01/2026
In relation to discussing #emotions, #feelings and #relationships on probably the best website to do so, HN
"Reflecting peoples feelings is sometimes called "showing sympathy.""
000
embedding-shapes @emsh.cat · 19/01/2026
This, this I'm fine with. It's when we start replacing humans that things turn to shit, but letting people chose? Good stuff reMarkable :)
000
embedding-shapes @emsh.cat · 18/01/2026
I like this one, basically "from pirate to politician/government worker", yet the titles don't make that super clear. en.wikipedia.org/wiki/Zheng_Y...
120
embedding-shapes @emsh.cat · 16/01/2026
Interesting unanswered question under that
fwip 1 hour ago

When you say "have now been resolved" - did the AI agent resolve it autonomously, did you direct it to, or did a human do it?
000
embedding-shapes @emsh.cat · 02/01/2026
Never did I know #evil could present itself as just two button
000
embedding-shapes @emsh.cat · 21/12/2025
They also still don't know how to do auth on the web, this is what happened when I tried to go to www.xbox.com/en-US/xbox-g... I swear no one at Microsoft or Xbox even tries this stuff themselves. What is that message even supposed to mean? Of course I'm signed out, I never logged in!
100
embedding-shapes @emsh.cat · 18/12/2025
Literally the first time I've had a #llm tell me "Hey, this is complete now, stop worrying or analyzing it, we've done it all and we should stop". Usually they keep asking/telling what to do next...
Screenshot of an LLM ending a message with "From a correctness and performance standpoint: you’re done."
010
embedding-shapes @emsh.cat · 11/12/2025
I realized just now that the titular character isn't actually the titular character at all, the main character's (code)name is Konoko.
000
embedding-shapes @emsh.cat · 11/12/2025
I can't believe no one made a spiritual successor to Oni (2001) yet, could be so incredibly deep and fun. Even something small like a #indiegame would do. Btw, is Oni old enough to be considered #retrogaming?
272
embedding-shapes @emsh.cat · 10/12/2025
*pat on #audacity back* yes, your redesign looks very good buddy
Screenshot of a "how we're redesigning Audacity for the future" popup when starting Audacity, which is missing button labels and other UI elements, with uneven margins everywhere and misligned images and links
010
embedding-shapes @emsh.cat · 07/12/2025
vintage meme at this point, but if it fits it fits :shrug:
000
embedding-shapes @emsh.cat · 07/12/2025
Inspired by Lego perhaps? :)
A green Lego "branch"
000
embedding-shapes @emsh.cat · 03/12/2025
I made my first #Nix / #NixOS library! Niccup: Hiccup-like HTML Generation in ~120 Lines of Pure Nix. Transforms Nix expressions into HTML. The website has some cool examples, I'm especially proud of the quine one that was slightly tricky to get right: embedding-shapes.github.io/niccup/examp...
Niccup: Hiccup-like HTML Generation in ~120 Lines of Pure Nix

Ever wish it was really simple to create HTML from just Nix expressions, not even having to deal with function calls or other complexities? With niccup, now there is!

[ "div#main.container"
  { lang = "en"; }
  [ "h1" "Hello" ] ]

<div class="container" id="main" lang="en">
  <h1>Hello</h1>
</div>

That's it. Nix data structures in, HTML out. Zero dependencies. Works with flakes or without.

The code is available here: embedding-shapes/niccup

The website/docs/API and some fun examples can be found here: https://embedding-shapes.github.io/niccup/
Why Generate HTML from Nix?

If you're building static sites, documentation, or web artifacts as part of a Nix derivation, you've probably resorted to one of these:

    String interpolation (''<div>${title}</div>''). Works until you need escaping or composition
    External templating tools. Another dependency, another language, another build step
    Importing HTML files, no programmatic generation

Niccup takes a different approach: represent HTML as native Nix data structures. This gives you map, filter, builtins.concatStringsSep, and the entire Nix expression language for free. No new syntax to learn. No dependencies to manage.
081
embedding-shapes @emsh.cat · 03/12/2025
Mr. McMahon reaction Meme.
- users with similar likes also liked:
- users with similar passwords also liked:
- users with the same password also liked:
- users whose password also mean the same thing as yours meant also liked:
000
embedding-shapes @emsh.cat · 01/12/2025
This is so #dystopian When you think about it, we humans are all living on a quota, always
A user had their LLM agent delete their D drive by mistake. The user is trying to troubleshoot the issue, but the user reached the quota limit so no more requests can be made.
010
embedding-shapes @emsh.cat · 03/11/2025
I don't know how I feel about this #gnome
Screenshot of a notification from GNOME in Gnome3 about donating to GNOME
100