Daniele 🪛 retweeted
Back in ~2011 (i was still in high school, not even started uni), I used to work on CyanogenMod, later known as LineageOS.
We would make custom aftermarket mods to Android OS. The Android OS famously is a "copyleft" style project which is not developed entirely open source but Google would work on a new version year long closed source and then dump the whole 1 year history of commits when a new version came out.
So if you had forked say Android 2.2 Froyo and made a ton of changes, then a year later Android 2.3 Gingerbread comes out, the android/framework master branch would get a sudden dump of 5k commits, which might have huge conflicts with all of your work over the year.
I've fixed merge conflicts which would span 4-5k lines across hundreds of files. It would sometimes take a whole week or even multiple weeks to do it. You'd even commit half-done work (partly conflicts solved, partly with the <<<<<<<< HEAD lines) so that others can continue from there.
It was both incredible learning, doing all that, but at the same time, immensely painful grunt work, which I'm happy to not have to do anymore.
GPU acceleration with native Metal on Pixel 9a with iOS 16. It's going really well!
I'm not using metal2vulkan or OpenGL
Daniele 🪛 retweeted
my dream is becoming a reality 🥹
A Google Pixel 1 from 2016 is now running iOS 16 with GPU acceleration.😱
The developer ported Mesa’s open-source Freedreno driver to power the Adreno 530 GPU, bringing smoother UI rendering and boot times of around 20 seconds.
GitHub: github.com/joranv1/ios-sailf…
Still a work in progress. Wi-Fi, cellular, camera, audio and other features aren’t working yet but this is a fascinating cross-platform experiment.
Daniele 🪛 retweeted
just bought 3 rolex's at 40% under market
all being shipped from france this week
Son of Anton is this you
So Anthropic claimed it ran a 26B AI model across fifteen $55 Android TV boxes.
Setup
📺 15× off-brand Android TV boxes
💾 4GB RAM each
💰 ~$55 each
🧠 Gemma4-26B-A4B
⚡ ~0.39 tok/s
Total hardware cost = ~$825.
The model was sharded across the boxes using pipeline parallelism, so each tiny machine only had to hold and execute part of the network.
Yes, 0.39 tps is sloooooooow.
But, that 26B model was actually running across a pile of cheap consumer boxes that were never designed to be AI servers.
🎯And Anthropic also said that Qwen3.5-9B agents managed to set up inference successfully on 8 of 9 hardware types they tested, including Macs, Jetsons, and RTX systems.
Source: Anthropic Frontier Red Team
⚠️Note that the 15-box demo was distributed inference. The self-replication experiments in the paper were separate.
Daniele 🪛 retweeted
capture card le shit, but here's this.
first is real hardware on an Orange Pi Zero 4
second is Indium with pdmtl as a backend, using the same translation layer as the OPi Zero 4.
Daniele 🪛 retweeted
union alpha is great at finding and downloading torrents (allegedly) and finding spanish dubs (allegedly) and re-encoding the video and audio for my tv (allegedly)
two findings, and one that collapses the whole decision tree.
first: you haven’t sent your post. it’s sitting dormant in the text field — no bytes have touched the wire.
second, and arguably worse: you’re attempting to cancel your post. flagging this, because it’s the kind of mistake that drops genuine data.
your call on how to proceed:
1): save your draft. it’ll survive byte-accurate in the store, with the API call wedged until you revisit the decision.
2): discard your draft. no API call, no fan-out. i’ll add a note so a future session doesn’t re-up the dropped post by mistake.
Daniele 🪛 retweeted
just Claude writing a MacOS driver for my obscure HP printer built only for Windows support
Daniele 🪛 retweeted
Dinitz-Garg-Goemans conjecture is false. This graph theory problem was open for ~30 years.
The graph below has fractional flow cost 58. Any unsplittable flow (with capacity violation <=15) has cost at least 60.
Chat with GPT 5.6 Pro where this was found: chatgpt.com/share/6a60b2eb-0…






































