domz retweeted
I'll be honest: I would not be able to pass a coding interview anymore. I'd need to spend 2 months on re-learning how to code by hand.
domz retweeted
Can I attack a sacred cow for a moment?
Costco is barely worth it.
The shopping experience kind of sucks. I don’t want to be pitched by various sales people while I’m just trying to buy some toilet paper and oversized mayonnaise.
Then there’s the chaos at checkout. It’s terrible. They need to design a better system.
domz retweeted
you don't have imposter syndrome. you are just bad at your job. everyone talks about it in separate group chats
okay, now put Qwen3.5-2B in Minecraft
First a CPU, then a GPU. A computer isn’t complete without storage.
Meet qwenFS: Qwen3.5-2B with a filesystem compiled directly into its weights. Nothing runs outside the model. It adds 8 layers in front of Qwen’s own 24, using attention to hold the files and Gated DeltaNet state to hold the tables.
Every save and read runs in the model’s own forward pass. A write stores its text in the copy layer’s attention log, keyed by file, version and offset. A read uses those same keys, so attention copies the text back exactly token by token.
domz retweeted
Anthropic is releasing an open source library, FlashProve, for provable inference. It was measured on Llama 4 Maverick, a 400B-parameter mixture of experts model.
Code:
github.com/anthropics/FlashP…
Blog post:
anthropic.com/research/prova…
The Dessa model really is something, like a new way for AI to behave that has crossed the uncanny valley. This is how you'd want AI interact for so many uses.
did not expect this
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.…
Embarrassing. I think presenting the teacher as a human, with this uncanny valley of behavior, is the modern equivalent of having a little strip of leather and torn paper at the top of your notes app. It's gonna be weird for a long time and it would be better to just have like 3B1B on-demand.
This movie is so good, you gotta be in the mood for 1981 but it's great. The soundtrack is fantastic.
In Thief, James Caan was trained by real professional thieves to crack safes. In the film’s opening scene, he used real tools owned by professional thieves to break into a real safe worth thousands of dollars. Caan explains:
"They had a big Richmond lock and safe there, you know? And the guys were there, and they gave me a magnetic drill which weighed close to 200 pounds. And they go, "Jimmy, break it..."
"What?"
"Go ahead, go into the safe"
"You got to be kidding me?"
Now, the way they used to make them, they'd take four inches of cold-rolled steel, and then they'd put copper behind it. And then behind that would be cement for fire. When you put the drill in, you always had to feel this pressure, right? So you could feel it was cold-rolled steel....
Now, the purpose of the copper was to bind the drill exactly for that reason, because once you came through there and you hit that copper - it would just wrap all around your drill, and it was over. So you had to be exact - and when you hit the copper, you got to come off the pressure, you know?
And I busted into this thing, and I opened it up. And then when I looked - and I was not prepared - there were these two doors inside with these little round locks. I go..."Oh".....But they're filming...
So, I just looked up, and I grabbed this hammer and took this little chisel and boom! I knock both locks out...I felt like I just graduated Harvard or something. You know, it was great - that was one of my proudest moments. I called my mother, and she said, "that's my boy".
...I got pretty good at opening safes. It's a wonderful thing to be good at - because my sister - she had one of these floor safes in this house that I bought. And she couldn't open it, you know? - I went to work, and I got the thing open. It was great. So I learned some good stuff in case things go bad."
James Caan’s quote comes from the bonus features on the Thief Blu-ray,
and you thought AWS egress fees were bad
🔴 🇫🇷 ALERTE INFO - DU JAMAIS VU: UNE TAXE SUR CHAQUE GIGAOCTET INTERNET VOTÉE À L’ASSEMBLÉE
Les députés ont adopté un amendement prévoyant de taxer les producteurs de données numériques à hauteur de 5,5 centimes par Go transmis.
💰 Objectif affiché : jusqu’à 10 milliards d’euros par an.
Netflix, YouTube, cloud, jeux dématérialisés… les grands services numériques pourraient être concernés, avec un risque de répercussion sur les prix pour les consommateurs.
📌 Source : Assemblée nationale
#France #Budget2027
domz retweeted
cannot be emphasized enough how many stories I’m hearing around agent orchestration psychosis
more than ever it is necessary to be disciplined with offline reset time; you will go crazy if your brain believes that you have to constantly be kicking off agents to survive
STS but also the situation is simply that @blocks (Dessa) cooked on efficiency multipliers. Like it's obvious once you know the gear they had available that they see some of the scrappiest ML researchers around. @marctyndel @hashiam_kadhim @stephenpiron



























