Umberto Raimondi (@uraimo) header
Umberto Raimondi (@uraimo) profile picture

Making computers do things mostly under duress | https://nitter.cf/t.co/HvjV0TyE9E

Joined September 2008
Photo by Umberto Raimondi (@uraimo)Photo by Umberto Raimondi (@uraimo)Photo by Umberto Raimondi (@uraimo)Photo by Umberto Raimondi (@uraimo)Photo by Umberto Raimondi (@uraimo)Photo by Umberto Raimondi (@uraimo)Photo by Umberto Raimondi (@uraimo)Photo by Umberto Raimondi (@uraimo)Photo by Umberto Raimondi (@uraimo)Photo by Umberto Raimondi (@uraimo)
Umberto Raimondi retweeted
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…
2,069
6,413
4,927
49,618
6,432,980
Umberto Raimondi retweeted
Meet qwenGPU: Qwen3.5-2B with a SIMT GPU inside its weights, 45 compiled layers in front of Qwen’s own, using its Gated DeltaNet layers as registers and memory. It runs real CUDA in its forward passes. The host compiles the kernel to PTX and loads it, then the model writes every step of the launch itself: every register, branch and memory write. The GPU has 4-lane warps, a 4-stage pipeline, 15 registers per lane and 126 words of global memory, all held in DeltaNet state. Qwen generates each launch token by token, each record running one machine block for one warp.
35
91
31
928
84,874
can your AI do this
トラックくん、来たよ!
57
Pretty sure the reason here is not the interest in protecting someone else IP. I expect a lot of subscriptions previously dormant being maxed out by this kind of projects right now.
Replying to @chasmmmmmmmmmmm
We added two sentences to Claude Code's prompt on Opus 5.5 to steer it away from recreating someone else's specific game because using Claude to infringe IP is against our usage policy. Original games, genre games, and games built from your own designs and assets are meant to keep working. We'll continue to iterate to ensure we're striking the right balance. In the meantime, if Claude turns down something that's yours, send it my way.
77
【朗報】大阪の日本橋にはラズパイを販売する超珍しい自販機が存在する
14
723
72
3,769
339,113
I wonder how many realize that doing this is very cheap now (including long term maintenance), and a better alternative to the good practice of dependency quarantine.
Fork your dependencies, trim them to only your use case, never update unless it breaks for your users. I’ve been vocal about this for 10+ years. I’ve always said that updating is way riskier than latent bugs (which can be tracked and CVEs monitored). If you are updating a dependency, it’s on you to analyze every single commit in the full transitive set of dependencies. If you dont see anything compelling, dont update! I remember at HashiCorp once in awhile an engineer would try to update a dep or replace a DIY lib with an external one and id always ask “show me the commit we need.” Dont update for the sake of it. Feeling pretty swell about this mentality with all the supply chain attacks happening.
58
Another day, another company with thousands of agents opening hundreds of PRs. The common denominator of the products they build is that they are wildly tolerant of failures, they can just fix them. Good luck applying that playbook where mistakes can be basically irreversible.
45
My eyebrows hurt just reading the introduction, very strong "This is the largest ship ever built, the Titanic" vibes, good luck
Fascinating case study of the sharpest edge of what you can do with agentic engineering in the Fable era - my eyebrows raised at least a dozen times while reading this
1
84
Not completely sure this has a market now, but it could in a few years
SONYさんから小型の空間再現ディスプレイ『ELF-MN1』が発表されました⚡️約54,000円! 僕もクリエイターとしてコンテンツ制作させて頂いたので、今後制作過程や実際使ってみた感想を投稿していきます。 ※ソニー社の依頼で提供頂いたデバイスのユースケースを開発し本投稿を行ってます #ELFMN1 #XYN
1
94
本日発売のSony 空間再現ディスプレイ ELF-SR2を見に来た。これは使えるぞぞぞぞぞ  pur.store.sony.jp/spatial-re…
18
ここ数ヶ月取り組んできたことが、ようやく形になってきました。 Raspberry Pi 5 + 単眼カメラ + IMUで、Visual SLAMで作った既存地図上での自己位置推定です。GPUなし・CPUのみ。 CPU使用率はかなりギリギリですが、フレームドロップなしでリアルタイムに動いています(入力はrosbag再生)。
15
93
5
1,085
47,064
Umberto Raimondi retweeted
Survival horror game where your webcam tracks your real eyes and the monsters only move when you blink - Descend into an abandoned lab full of monsters - Scavenge what you can - Make it out before your eyes give in - Blinks can change the world around you Game is DON'T BLINK.
48
620
93
4,100
191,536
Umberto Raimondi retweeted
🚨 ‘Ashita no Joe’ Creator Tetsuya Chiba has Passed Away at the age of 87. He was the first Manga artist to ever receive Japan's Order of Culture.
443
12,351
1,338
97,203
1,910,365
Frankly, the only benchmark that matters nowadays is the "me" benchmark.
65
Once again, I'm noticing how strictly models adhere to system prompts nowadays, they won't budge an inch from the initial constraints
61
EVERYTHING that bundles eMudhra certificates in their release is affected and will need a new release. Easy to spot many "security influencers" that don't know what they are talking about when they treat each CVE(now Java) as something new and try to explain it with mumbo-jumbo.
1
1
128
And for Java the faster way to remediate is this, no need to wait for a new build of the JDK distribution of your choice: ops.java/emudhra (As usual the X Community Notes did a great job "improving" scaremongering tweets)
1
34
Another WebGL example, 6.1 Sol fixing an earlier Argon prototype with correct occlusion, fixed incoming train angle (hard part) and distorted masks. It's way better at this, and with shorted screenshot to fix loop.
電車がきます。
2
1
109
Again, the prompt was just "recreate the scene in this tweet".
8
A few rounds of Gemini 4 Argon starting from the tweet, asking for a Makoto Shinkai aesthetic:
4