Franck Marchis

Here we are, alone in space on a spaceship we call Earth. I am disheartened; I thought by now we would have found our cosmic brothers and sisters to share our challenges, joys, and defeats with. But no, we are floating in this galaxy, arguing about our tiny lives and wasting the time that could be used to make us a better civilization. #todaythought

@smoos_tweets Sauf qu’il ne dit pas ça… mais bon le mensonge a l’air de fonctionner sur cette plateforme …

On y est. Les populations immigrées vont désormais fixer des conditions culturelles aux Français désireux de vivre où bon leur semble...sur leur propre terre. Tout va beaucoup plus vite que prévu ⌛️

Sophisticated visits by multiple jamming-resistant drones in swarm disrupted an AFB in Louisiana. Their origin is unknown and no counter measures could be activated. If only we had an alert system to detect and destroy those in advance… @usairforce should check what our team recently demonstrated.

Republicans could literally fund TSA, FEMA, and the Coast Guard right now. But once again, they’re choosing Donald Trump over the American people. Absolutely shameful.

LeWorldModel: Yann LeCuns Radical Simplification of World Models Just Made Physics-Aware AI Practical In the race for artificial general intelligence, two paths have emerged. One is the familiar scale everything route: bigger LLMs trained on ever-larger text corpora. The other, championed for years by Yann LeCun, is building world models: compact systems that learn the underlying physics of reality directly from raw sensory data (pixels) so AI can plan, predict, and act in the physical world like a robot or self-driving car actually would. Until now, the second path has been frustratingly difficult. Joint-Embedding Predictive Architectures (JEPAs) - LeCuns elegant framework for learning predictive representations without reconstructing every pixel - kept collapsing during training. Researchers had to resort to a laundry list of hacks: multi-term loss functions (up to six hyperparameters), frozen pre-trained encoders, stop-gradients, exponential moving averages, and other duct-tape tricks just to keep the model from mapping every input to the same useless output. LeCuns team (Mila, NYU, Samsung SAIL, and Brown University) dropped a bombshell: LeWorldModel (LeWM) - the first JEPA that trains stably end-to-end from raw pixels using only two loss terms. No more house-of-cards engineering. Just a clean, simple recipe that works on a single GPU in a few hours with only 15 million parameters. The Core Breakthrough: SIGReg Saves the Day LeWorldModels secret weapon is a new regularizer called SIGReg (for spherical isotropic Gaussian regularizer). It enforces a simple Gaussian distribution on the latent embeddings. This single term prevents representation collapse without any of the previous heuristics. The training objective now has just two parts: 1. Next-embedding prediction loss - the model predicts what the next latent state should be. 2. SIGReg - keeps the latent space well-behaved and diverse. Thats it. Hyperparameters drop from six to one. Training becomes stable, reproducible, and dramatically cheaper. The model learns directly from raw video frames (no pre-trained vision encoders needed) and produces a compact latent world model that can be used for fast planning. Impressive Results on Real Benchmarks Despite its tiny size, LeWorldModel punches way above its weight: - Trains on a single GPU in a few hours. - Plans actions up to 48 times faster than foundation-model-based world models. - Uses roughly 200 times fewer tokens than alternatives. - Matches or beats far larger models on diverse 2D and 3D control tasks (e.g., manipulation, navigation). - Its latent space encodes meaningful physical quantities (position, velocity, etc.) - proven by direct probing. - It reliably detects physically implausible surprise events, showing genuine causal understanding. Crucially, adding a decoder and reconstruction loss hurts performance on downstream control tasks. The pure JEPA objective already captures everything needed for planning - extra visual details just get in the way. Project website: le-wm.github.io Official code: github.com/lucas-maes/le-… Why This Matters for the Future of AI LeCun has been saying since 2022 that world models (not next-token predictors) are the key to real intelligence. Critics always pointed to the training instability. LeWorldModel removes that objection with elegant simplicity. This is a philosophical reset: AI can learn physics the way babies do - by watching the world unfold - without needing supercomputers or endless text. The implications for robotics, autonomous vehicles, and embodied agents are enormous. Suddenly, building a physically grounded planner is something a researcher (or even a hobbyist) can do on consumer hardware. 1 of 2

They are outside the launch window, so the Spectrum launch is aborted for today. Let's try again @isaraerospace . The view is spectacular and I hope yo be invited one day to see a launch.

T-0 was crossed but the launch was aborted... More soon

T-3min Let's do this @isaraerospace

They are about to try to launch a rocket from Europe. @isaraerospace second flight of the ‘Spectrum’ launch vehicle from Andøya Space might happen in 30 min. Live here: youtube.com/watch?v=MsbZj8…

SkyMapper CEO @AllPlanets had a productive visit to Hansamatrix in Ventspils to discuss manufacturing our #SkyBridge in Latvia. We look forward to growing our presence in the Baltic and Nordic region and bringing the SkyMapper network closer to more observers.

People don’t look up because of science. They look up because something pulls them there. The difference is what happens next. Astronomy turns that curiosity into observation, measurement, and understanding. That’s where the real story begins. Do you want to help write it?

Aren't you tired on winning?

A warm welcome to Vincent Vrignaud from @ISAE_officiel in Toulouse, France 🇫🇷 who has just joined the @SETIInstitute and @Skymapperspace to work on complex maneuvers of satellites.

Will this be in the front page of major news papers in the western world? I am still hoping it will. #IranRevolution2026‌

Thrilled to be partnering with @Skymapperspace to open up the space market to the world! There's so much work to be done but we believe space should be global, inclusive, and accessible to everyone. Finding a partner who shares those values makes the journey that much better. 🛰️🔭

We’re excited to partner with @decenspace to advance space situational awareness. 📡Decen Space connects satellites to a global ground station network. 🔭 SkyMapper adds a decentralized layer of optical tracking from ground-based telescopes. Together, we can bring: better detection, tracking, and visibility of objects in orbit.

☀️@SpaceTechExpo comes to sunny Anaheim this June. Join us as @Skymapperspace CEO and @SETIInstitute Director of Citizen Science @AllPlanets takes the stage to explore how observation itself is changing - from isolated instruments to global, connected networks. The way we understand space is evolving. So is the infrastructure behind it.

@McCormack_Tara @AsteroidEnergy You should better read the news before saying that there is not a whisper from the EU. reuters.com/world/europe/f…

Quite astonishing that Israel can simply invade, expel 100,000s, say they can’t come back and not a whisper from the EU or America.

The SkyMapper team spent a few days in Riga connecting with builders at Start Up House. Before leaving, we installed our first #SkySphere in Europe in Mārupes Novads and got first light immediately. Thank you for hosting the first node and helping kick off the European SkyMapper network.

L’IA peut désormais générer des images et des signaux indiscernables de vraies observations astronomiques. Bonne nouvelle pour la recherche… mais aussi un risque pour la confiance dans les données scientifiques. Un défi que la science va devoir affronter. futura-sciences.com/sciences/actua…

Félicitations à une partie du paysage médiatique français qui a réussi à polariser un pays qui a pourtant tout pour réussir, en donnant aux Français l’impression qu’ils doivent choisir entre deux extrêmes : l’un qualifié d’islamo-gauchiste, l’autre soutenu par la Russie et ouvertement raciste. La France mérite mieux que ce faux choix. Français, réveillez-vous.