Not sure I enjoy how much people are sharing ICLR submissions without proper reviews, but apparently the conference process is not fast enough? We should have another way to publish in ML/AI that's continuous and fast, as journals are the former but not the latter

Dunno when they planned to officially release it, but it's already on arxiv ¯\_(ツ)_/¯: arxiv.org/abs/2510.12403

In Versailles

Despite all the flashy dancing and kung-fu humanoid videos, real ones know that general manipulation is *the* bottleneck that no one has cracked. Check out some of the progress from our team on humanoid VLAs. Same pretrained checkpoint that controls Alohas and Bi-arm Frankas!

Precisely my issue with many papers using cos sim as a "metric"

Agree with the sentiment, but that's n9t what democratization means lol

Our international internship program, CaCTüS, offers paid research internships at the Max Planck Institute for Biological Cybernetics, the Max Planck Institute for Intelligent Systems, and the AI Center in Tübingen (Germany) to students who face significant constraints in their

the right image is far more interesting than the right one, sadly

my brother in christ, we are doing this now

As a former physicist, this is the most survivor bias post I've ever seen

Thinking of doing a phd? Come work with us in Barcelona advancing autonomous driving. Possible directions are end-to-end driving, their training efficiency, explainability and new architectures, as well as federated learning and V2X, among others.

Sorry guys, I know it’s a bit of a coping mechanism, but everyone sharing their review scores (good or bad) is tacky and cringe af

1 hour to go for cvpr, godspeed

My typical test with new models: prompting FLUX.2 with empty prompts

Thrilled to be at NeurIPS 2025 to present 🍎 RLRF, Reinforcement Learning from Rendering Feedback, our approach that uses Rendering-Aware Rewards to push Vector Graphics Generation with LLMs to the next level. Just touched down in San Diego and ready to dive in 🌴 Project page:

I'll be at NeurIPS from Dec 2 to 6! Come visit our spotlight poster from 11a to 2p this Thursday (morning session, 2nd day)! tl;dr: Just scaling up data and compute is *not* enough to enable RL to solve complex, long-horizon tasks. The culprit is the horizon.

This week I'll be at #NeurIPS presenting our latest work "Elastic ViTs from Pretrained Models without Retraining"! TL;DR: You can make pretrained ViTs elastic in under 5 minutes on a single A100 via structured pruning, without re-training! Photo: in-context semantic

Treating time as an additional spatial dimension (e.g., 4D models) is a wrong imo; things converge thanks to the mercy of (large) NNs

📣 Hiring Research Interns for Meta Superintelligence Labs in Zurich! Work on large-scale generative models (image/video gen, multimodal, world models) with real impact on products used by billions. 📍 Zurich | 🕒 6 months | 🎓 PhD students metacareers.com/profile/job_de…

claude gotta chill with all these README files it creates for the simplest of requests, goddamn