You can never have enough compute.
192.80.53.0
Joined May 2026
- Tweets1.8K
- Following386
- Followers792
- Likes6K
Charles retweeted
#they ousted the only major AI figure who isn't in the rationalist safety doom cult... it's gonna get spooky
Open source will win. They can't stop us.
NVIDIA is pushing local (on-device) AI hard so ordinary users with RTX hardware can run capable models and agents privately, quickly, and without ongoing cloud costs or data leaving their machines. Wow, I wonder why the establishment would want their CEO gone…
1
50
Deepseek Harness has a lot of telemetry you might want to remove before you run it on your computer.
"Sends logged session history through that feedback event, including inherited fork history, to DeepSeek’s OpenTelemetry collector. Works independently of your model provider, so a local-model conversation can qualify. There are no shipped default redaction rules."
Ever wonder what claude code or codex collects?
4
2
238
Charles retweeted
I wouldn't be surprised if some of the extreme anti AI sentiment in game dev is actually partially being funded and astroturfed by AAA companies like EA to crush competition in the Indie game sector.
The big studios all use it now but for some reason it's only your average indie devs getting dragged by the mob for daring to utilize the new technology.
Idk, it's what I would do if I was an evil company trying to protect the status quo and maximize shareholder value
Charles retweeted
I wanted to see if this exploit could be used to boot the Note 4 Exynos back into ARM64, nobody has ever gotten the Note 4 Exynos to run ARM64 code, as the S-Boot forces it into ARM32 mode... until now!
It restarts the CPU back into ARM64 mode, it shockingly fucking works.
Writeup github.com/UnsignedChad/CVE-…
I'm proud my work makes it possible to run custom images on locked bootloaders. I'll mention this work in my write-up and link the repo. Amazing work.
github.com/Gabriel2392/SVE-2…
For Exynos enthusiasts:
You can now fuse your own secure boot keys in your device, allowing to permanently run custom images on locked bootloader.
Repo: github.com/Gabriel2392/SVE-2…
Up to 2026-10-05 security patch for fusing (once it's fused you can flash whatever you want, it never goes away)
Only s5e8825 is supported in the main branch. You can always port it to your Exynos device and PR it.
As always big up Gabriel
Writeup github.com/UnsignedChad/CVE-…
Charles retweeted
Qualcomm Snapdragon 6 Gen 1 dieshot annotated
kurnal-insights.com/dieshot/
The cheaper you go, the smaller the GPU is
brb
We asked GPT-6 Astra to edit Qwen3.5-0.8B weights to improve math.
It directly changed only 0.016% of weights, fixing 0.8B's reasoning loops & improving math up to 6x and even instruction following 3x, without gradients or training.
We call this Weight-Space Programming: frontier LLMs can now natively read/write neural network weights! 🧵
Blog: rikiyat.github.io/blog/weigh…
Readers added context they thought people might want to know
LLM was given weights explicitly as text in prompt and used code/forward passes (per authors' logs) to ablate Qwen looping direction, a known thinking-mode issue per model card. Primarily fixed repetition in that mode, not general 'natively' improving weights/math.
rikiyat.github.io/blog/weight-sp…
github.com/RikiyaT/lm_und…
huggingface.co/Qwen/Qwen3.5-0…
huggingface.co/RikiyaT/Qwen3.…
