
Part-time thinker AI Enthusiast Principal Software Engineer @RedHat
Brno
Joined December 2009
- Tweets566
- Following1.7K
- Followers241
- Likes3.3K
Pinned Tweet
Quidquid latine dictum sit, altum sonatur.
- Whatever is said in Latin sounds profound.
15
Friendly reminder on why open models matter.
With @JensenHuang out, I hope that commitment doesn't leave with him. 🙏
teapaw.org/JensenHuang/status/208…
For my first post, I’m sharing a letter @NVIDIA signed on why open models matter.
AI will transform every industry, power every company, and be built by every country.
Open models strengthen safety and cybersecurity, accelerate innovation and diffusion, and enable sovereignty.
The world needs both frontier closed models and frontier open models.
images.nvidia.com/pdf/Open-W…
14
10M tokens of context at full-attention accuracy is wild. 🤯
This might be a new game for serving economics. 💸
teapaw.org/jcrwhittington/status/…
Today we are happy to share one of our first open research releases: Latent Sparse Attention (LSA), a step towards subquadratic attention that matches accuracy of full attention at ultra-long context.
Sparse attention attends to only a few tokens per query, but it has to find the right ones. At 10M tokens, that search takes 98% of the attention layer's time. LSA makes sparse attention up to 22.7× faster and full attention 72.4x faster.
With @kubicek_ales @caciolai @PengLiangzu @SukritSumant @Ben_Hoov @FraserGreenlee @RMBattleday @radim_urban @thoefler @jcrwhittington
1
133
QUADS 3.0.4 is out 🍂
It's a small bugfix and feature release, and our 25th since 1.0 shipped in June 2016. Ten years of automating bare-metal lab scheduling 🎉
What's new 🧵👇
1
1
43
📊 --json output for every table-producing CLI action, which makes scripting much easier
📦 instack files are now copied to the first host on release
✂️ Self-schedule descriptions are truncated to a configurable length
1
3
🔧 Fixes for inactive assignments, Juniper verify, parent watchers, and bios_id in host exports
Heads-up: it needs Fedora 43+ and runs a DB migration automatically.
Release notes: github.com/quadsproject/quad…
1
15
QIIP 0.2.0 has Arrived! quads.dev/2026/10/28/qiip-v0…
1
3
57
Provisioning got an overhaul:
• Power-on new nodes via Redfish
• Reuse compatible GPU drivers and verified llama.cpp builds
• Size vLLM launches from available memory
• Resume verified stages after transient failures
• Fedora RPMs for every release
1
12
Source and docs: github.com/quadsproject/qiip
Huge thanks to @sadsfae for driving this release. Try it, and if something breaks, file an issue with the timestamp and what you were doing. That feedback is what shaped this release.
1
13
Me crucé con esto, Messi está casi 6 desviaciones estándar por encima de la media de delanteros de grandes ligas en cuanto a goles y asistencias en 90 minutos. Estadísticamente es prácticamente imposible que vivas para ver a alguien así
Gonzalo Rafuls retweeted
Mexican net artist Canek Zapata presents a glitch-art video synchronized to experimental audio titled ‘Música Música,’ January 2026.
Gonzalo Rafuls retweeted
This is what we're up against if we allow AI vibe coding to go unsupervised.
Strong proof that LLM-generated code can cause more harm than good in unskilled hands.
I stand by 100% of the points in this post.
Hiring software developers is going to become a much bigger pain than it used to be, since recruiters cannot make a justified decision just by glancing at their GitHub profile.
Open-source projects will be harder to manage due to the higher costs of reviewing PRs to prune AI slop.
This is probably the best quote from the post that sums it all up:
"AI multiplies what you already know.
- 10 years of experience × AI = 10x output
- 0 years of experience × AI = 10x slop"
The skill is not yet dead.
webmatrices.com/post/vibe-co…

















