Sabitlenmiş Tweet
Steeve Morin
25.9K posts
Steeve Morin
@steeve
Building @zml_ai (and we're hiring), ex @zenly, ex Exalead, ex @google. Skydiver and wingsuiter.
Paris, France Katılım Ekim 2010
1.3K Takip Edilen8.3K Takipçiler
English
I actively prefer ecosystems with BDFLs. Many instances in which I would question or disagree, but I recognize I prefer that form of leadership over comitees.
Joran Dirk Greef@jorandirkgreef
I wrote these words 7 months ago. I am more grateful today for Andrew's leadership of the ZSF, that the foundations of TigerBeetle, our compiler, should not be vibed out beneath us. Standing up to “trillion dollar” big corp… Zig is hard core quality. tigerbeetle.com/blog/2025-10-2…
English
It turns out it's possible to map GPU memory to BAR-based CPU on a consumer 5090 via Vulkan.
Either CUDA allocated / import in Vulkan or Vulkan allocated / import in CUDA. Both work.
Steeve Morin@steeve
We're trying to use BAR to write to VRAM and make it available to CUDA. It works on Vulkan. If you can make it happen, we have a good bounty for you.
English
Of course we use it but in Zig allocators matter much, much less than any other language I’ve used.
Matt@matt_dz
mimalloc: A new, high-performance, scalable memory allocator for the modern era microsoft.com/en-us/research… May 13, 2026 by Daan Leijen
English
Running this on a box with 2 of gen2 optane drives. Top (-K0) is normal O_DIRECT, bottom (-K1) is using IO slots. CPU is AMD EPYC 7763, slow clock, but gets 9.1M IOPS on a single core (64/192 are thread siblings). IOMMU passthrough. That's a 66% improvement, IRQ driven same win.
Jens Axboe@axboe
Nothing like an LSFMM session on "can we get closer to spdk" for motivation - hacked up a concept on the flight back home that increases per-core performance by 50%. In-kernel will never be able to beat bypass, but we can get a lot closer with not a huge amount of work.
English
@gchampeau @pierrehebrard_ i don't buy that: tesla optimus robots were famously tele operated doing bartending, which requires a lot more fine control
English
@steeve @pierrehebrard_ Moreover, teleoperating would not be that accurate and natural at such speed. The teleoperator would lack the touch and force feedback required.
English
La vitesse d'exécution atteint désormais celle des humains dans cette vidéo temps réel des robots Figure qui trient des paquets sur une chaîne de logistique. Impressionnant !
Figure@Figure_robot
Watch a team of humanoid robots running a full 8-hr shift at human performance levels. This is fully autonomous running Helix-02 x.com/i/broadcasts/1…
Français
@steeve @gchampeau To be honest it might very well be that the training data has the operator adjusting his mask a lot of time
English
@gchampeau @SciumoInc i'm not sure i buy it since the obstacle seems a bit far to me but i'd love to stand corrected
English
@steeve @SciumoInc The metal sheet on the right. According to Figure the model policy prescribes that the hand be used to protect the head when an obstacle could damage it.
English
@steeve @gchampeau That point was already addressed as a programmed safety pose when shifting, avoids hitting obstacles.
English
N'importe quel être humain aurait aussi percuté l'enfant dans une telle situation parce que même en ayant un temps de réaction surhumain il n'y a pas la distance pour freiner... Donc qu'est-ce que ça prouve au juste ?
Prisme@PrismeMedia
🚗⚠️ TECH Lors d’un test utilisateur du système de conduite autonome de Tesla, un scénario simulant un enfant surgissant derrière un bus scolaire n’a pas été correctement géré.
Français
The amount of engineering that went into this is insane. What it enables, even more.
Corentin Kérisit@corentinanjuna
Hermetic LLVM now supports cross compiling libstdc++ from source. Community asked for it and it was a prerequisite for supporting hermetic remote CUDA compilation since CUDA prefers libstdc++ headers over libc++. Another step towards 100% hermetic builds.
English
this is bullshit intel
Steeve Morin@steeve
WHYYYY INTEL WHYYYYY not everyone needs multiprocessing
English
Et nous avons fait un épisode d'A La French avec lui!
youtube.com/watch?v=8hLPNQ…
YouTube
Fabien@Fabien_Mikol
♟️ Le joueur d'échecs et ex-coach de Magnus Carlsen @FressinetL raconte la révolution AlphaZero (sa victoire sur Stockfish en 2017) en termes de performance mais aussi de créativité, et comment cela a pu influencer positivement par la suite le jeu de Carlsen lui-même.
Français
