@AdinaYakup: BAAI just released the Orca paper on @HuggingPapers A Multimodal Latent World Model: it learns the world itself first, …

X AI KOLs Following Papers

Summary

BAAI released the Orca paper describing a multimodal latent world model that learns a unified world representation first, then decodes into text, images, or actions using frozen backbones and tiny decoders, with weights coming soon.

BAAI just released the Orca paper on @HuggingPapers A Multimodal Latent World Model: it learns the world itself first, and text/images/actions are just different ways to read it out How? They froze the whole backbone and only trained tiny decoders on top. So any win is the representation, not the decoder. And it scales across all 3 at once. Weights coming soon!
Original Article
View Cached Full Text

Cached at: 07/01/26, 10:05 AM

BAAI just released the Orca paper on @HuggingPapers

A Multimodal Latent World Model: it learns the world itself first, and text/images/actions are just different ways to read it out

How? They froze the whole backbone and only trained tiny decoders on top. So any win is the representation, not the decoder. And it scales across all 3 at once.

Weights coming soon!

Similar Articles

Orca: The World is in Your Mind

Hugging Face Daily Papers

This paper introduces Orca, a world foundation model that learns a unified latent space from multimodal data using next-state-prediction, outperforming specialized baselines on downstream tasks like text generation, image prediction, and embodied action generation.

@artemZholus: thanks! in the second paper (https://arxiv.org/abs/2605.06388) we used your (and RAE's) recipe and it worked.

X AI KOLs Following

This paper systematically compares reconstruction-based and semantic latent spaces for action-conditioned latent diffusion world models in robotics. It finds that semantic encoders like V-JEPA 2.1 generally outperform reconstruction encoders on policy-relevant metrics, advocating for semantic latent spaces as a stronger foundation for robotics world models.