CVPR going to Seattle 3 times within less than 10 years

Lenore Blum (Carnegie Mellon University) presents a really nice, slightly different definition of world models.

🪜 What if humanoids could climb ladders and work on them straight out of simulation? Meet LadderMan: a perceptive system for zero-shot sim-to-real ladder climbing and on-ladder manipulation. Watch the humanoid climb, stabilize, and manipulate—all in one system. 🤖👇

Jack Gallant (UC Berkeley), in his talk “Representation of the World in the Human Brain,” made a really interesting observation about world models: evidence that the mammalian brain implements a Generative World Model is "shockingly poor".

#KostasThoughts: I’ve been thinking about the criteria for award-winning papers 🏆 They should embody what our community values most: originality, potential impact, and reproducibility. Yet (full) reproducibility is often missing, even among award-winning papers. Making code and model weights publicly available should be a criterion for award consideration. This would send a strong signal to decision-makers in both academia and industry about the importance of open and reproducible research. Papers would receive the recognition they deserve, while the community would gain a deeper understanding of the engineering innovations that drive performance. Thoughts?

Our @SimonsInstitute workshop on intelligence starts tomorrow (June 8)! We're bringing together researchers from AI, TOC, Psych, and Neuro to discuss two emerging themes: world models and social cognition. Join us in Berkeley or via livestream: simons.berkeley.edu/workshop…

The SharpaWave hand is really good but the main thing it has is tactile resolution -- thousands of pixels. Huge if you are an AI lab, which is presumably why nvidia picked it for their reference hardware. The price will come down but you want that data

What if a self-driving car could learn its position just by watching millions of hours of driving video? Researchers at Wayve and Simon Fraser University introduce LA-Pose: they train a model to learn "latent actions"—hidden motion patterns—from unlabeled driving footage, then repurpose those features for camera pose estimation using a tiny amount of 3D-labeled data. Result: LA-Pose beats top feed-forward methods by over 10% on Waymo and PandaSet, while needing orders of magnitude fewer labeled examples. First proof that self-supervised inverse-dynamics can power pose estimation. LA-Pose: Latent Action Pretraining Meets Pose Estimation Project: la-pose.github.io/ Paper: arxiv.org/abs/2604.27448 Blog: wayve.ai/thinking/la-pose/ Our report: mp.weixin.qq.com/s/XoUvfQQoj… 📬 #PapersAccepted by Jiqizhixin

AlexNet, Seq2seq, Transformer released neither code nor weights. ResNet, GPT2, BERT, CLIP did not release source code, just weights.