Ok so I am not the only one unbothered about Jev hype 😅
Replying to @moinnadeem
Definitely just a transformer encoder with highly structured inputs and outputs...
I am also baffled by the hype. Like do people not understand that auto regressive LLMs are just classifiers where the number of classes is the size of the vocabulary...
230
The only thing we expect of you is to slash taxes by 80%, not one more agency for you to sip champagne with your schoolmates paid by people who actually ship.
🔴Roland Lescure, Serge Papin, David Amiel et Anne Le Hénanff annoncent la création de la Direction de l'intelligence artificielle et du numérique (DIAN) des ministères économiques et financiers.
Pour + d'infos : presse.economie.gouv.fr/berc…
192
Vincent Lordier retweeted
We are back again :) After three weeks of quiet building.
Introducing Genesis World 1.0, our latest simulation platform, the second release in our full-stack suite. Open-sourced.
Robotics is still bottlenecked by the 1× speed of the physical world. Every model, checkpoint, and data recipe eventually needs to be tested on physical hardware, slowly, expensively, and with limited coverage.
One hour in reality can become 100 days in simulation. That is how robotics model iteration moves from a wall-clock bottleneck to a compute problem.
To make this work, simulation has to be both fast and trustworthy.
Over the past year, we rebuilt the entire stack: a GPU-accelerated cross-platform compiler, penetration-free multi-physics contact solvers, unified rigid and deformable physics, and a photo-realistic renderer purpose-built for physical AI applications.
We built Nyx, a high-performance path-traced rendering engine for robotics application.
Genesis World 1.0 achieves near realtime performance with our latest development for penetration-free IPC solver, supporting various types of deformables beyond rigid bodies. It supports contact-rich, dexterous manipulation simulation across different embodiments: unitree, sharpa, wuji, genesis hand and various types of grippers.
Under the hood is Quadrants, our effort in pushing forward cross-platform GPU-accelerated computation. Quadrants started as a fork of Taichi, and we rebuilt most of the critical parts for optimizing simulation workloads, giving 10x faster launch time and up to 4.6x runtime performance compared to the initial Genesis release.
Together, they bring us to an unprecedentedly low sim-to-real gap, enabling zero-shot real-to-sim model evaluation and much faster iteration of GENE.
All available today.
Genesis World 1.0: github.com/Genesis-Embodied-…
Quadrants: github.com/Genesis-Embodied-…
Nyx: github.com/Genesis-Embodied-…
Every paper should look like this: “how can we infer 10000x faster ?”.
Video diffusion models learn motion indirectly through pixels.
But motion itself is much lower-dimensional.
We introduce 64× temporally compressed motion embeddings that directly capture scene dynamics.
This enables efficient planning -> 10,000× faster than video models.
🧵👇
7
1,326
Every paper should look like this: “how can we infer 3000x faster ?”.
You don't imagine the future by mentally rendering a movie. You trace how things move -- abstractly, sparsely, step by step.
We built a model that does exactly this. It predicts motion, not pixels -- and it's 3,000× faster than video world models.
Myriad, accepted at @CVPR 2026
Every paper should look like this: “how can we learn 800x more efficiently”.
Stereo depth is important in robotics, and relies heavily on synthetic data. But what actually makes for good synthetic data?
In WMGStereo, we study dataset design and discover a powerful data recipe - just 500 samples of our data can match 40k Sceneflow samples! 🧵[1/7]
4
331
This is what those crypto mining rigs from 2018 were actually meant to do.
For whoever had to either do this manually or pay the GPU bills: this should actually compound.
i open-sourced autokernel -- autoresearch for GPU kernels
you give it any pytorch model. it profiles the model, finds the bottleneck kernels, writes triton replacements, and runs experiments overnight. edit one file, benchmark, keep or revert, repeat forever.
same loop as @karpathy autoresearch, applied to kernel optimization
95 experiments. 18 TFLOPS → 187 TFLOPS. 1.31x vs cuBLAS. all autonomous
9 kernel types (matmul, flash attention, fused mlp, layernorm, rmsnorm, softmax, rope, cross entropy, reduce). amdahl's law decides what to optimize next. 5-stage correctness checks before any speedup counts
the agent reads program.md (the "research org code"), edits kernel.py, runs bench.py, and either keeps or reverts. ~40 experiments/hour. ~320 overnight
ships with self-contained GPT-2, LLaMA, and BERT definitions so you don't need the transformers library to get started
github.com/RightNow-AI/autok…
Vincent Lordier retweeted
🚀 Our second Defense Tech Hackathon in Paris is a wrap!
Over 120 participants formed 19 teams tackling challenges from swarm orchestration and stealth communication to real-time threat detection and autonomous mission execution.
Five teams stood out:
1st Place – SandFox Hacking Team: Developing a mesh communication platform for resilient battlefield networking. By Nassim BOUICHE, Christophe BOCHER, Louis Matthews, Sarah Ounes
2nd Place – Rhino Dynamics: Creating a multi-modal detection, modular automatic shotgun turret to counter FPV drones. By Kristóf Domán, Hugo Valade, Lukas Razinskas, Victor B., Alex Tarasov
3rd Place – Excubitor: Building fast target detection and interception powered by a neuromorphic vision feed. By Shannah Santucci, Tobias Rothlin, Jules Lecomte, Charles Schleich
💡 Honorable Mentions
The Swarmers – Drone swarms that switch to vision-based stealth coordination when RF-jammed. By Paul Adam, Jean-Max Rodriguez, Felix Vogeli, Hugo Vaillaud
Arcy – System for precise threat detection and classification in orthonormal spatial frames. By Elliot Le Rolland, Pierre Henaut, Younès Dollé, Yanis Hemdane
Huge thanks to our jury for their support: Moshe Lawlor (Stealth Startup) · David Rogberg (Bifrost) · Antoine Bassoul (Project Q) · Oleksandr Korzh (Dignitas) · Bastien Allanot (ex-Special Forces) · Serhii Lubkin (Synteza AI) · Damien Engemann (Avenor Industries)
And to all our partners and sponsors who made this possible — thank you for pushing the frontier of European defense innovation forward:
Our host @iXcampusSGL, and @Comand_AI, @Parrot, Project Q, Synteza AI, Alta Ares, @NADindustries, Bifrost Defence, Pilotix, Proteus Group, @Dignitas_Fund, Neurobus, Stronghold AI, the Ukrainian Ministry of Defence @DefenceU, @inflection_fund, and DroneAid Collective.
And to our co-organizers who make this possible: Guillaume AUBARD, PhD, Antoine Bassoul, Fawzi ELGHAZOUI, Thomas Bernaudon, Vincent Lordier, Shannah Santucci, and Stéphane Razzon
Keep building — and see many of you this week in Berlin again 🦾
Should be big.
First days in the new @imec_int Paris Office where we are building a new frontier AI research team together with @karl_tuyls . A little bit empty right now 😄 but researchers and engineers are joining in the next days and weeks. Super exciting time 🙃
1
207
The BEST thing we can wish to Europeans is a slight ban from the US, so they wake up and ramp up.
Monday morning read: a fascinating deep dive in recent Chinese chips developments as well as the coming co-evolution with LLM builders
fresh analysis from the HF team
=> huggingface.co/blog/huggingf…
148
Keep an eye on Rémi and al. work
My 15mn talk at the @Google Gemma event in Paris.
Thanks for hosting me! 😘
youtube.com/watch?v=L0uxfZMl…
Job openings 2026: vibe coding code cleaner 🧹
New @a16z podcast: a deep dive on vibe coding
@stuffyokodraws and I walk through the landscape of tools in this space, who is using them, and why they're growing so quickly.
We'll start with why these products are so valuable to non-technical people (like me!) 👇
1
327































