Link: Nvidia shares Cosmos World Foundation Models, a family of world models that can generate "physics-aware" videos, available from Hugging Face and Nvidia's API (Kyle Wiggers/TechCrunch)

Nvidia has launched a new series of AI world models named Cosmos World Foundation Models (Cosmos WFMs). These models, designed to simulate and generate "physics-aware" videos, were unveiled at CES 2025.

The Cosmos WFMs are available in three categories: Nano, Super, and Ultra, tailored to various performance needs. They adjust from low-latency applications to high-quality outputs and range from 4 to 14 billion parameters.

Nvidia's new models are openly accessible via their API, NGC catalogs, GitHub, and Hugging Face, promoting broad usability. The models are versatile, fine-tuned for diverse tasks including physics simulations and synthetic data creation.

Despite controversy over their training data sources, Nvidia maintains its models are developed responsibly without infringing on copyrights. The company asserts the data used was gathered from lawful public and private sectors.

Nvidia stresses that although the Cosmos WFMs are labeled 'open,' they aren't fully open source. The firm has yet to release comprehensive data or tools essential for model replication from scratch.

These advancements are poised to enhance applications in robotics and autonomous vehicles, with companies like Uber already integrating these models. Nvidia aims for Cosmos WFMs to significantly advance industrial and robotic AI capabilities. #

--

Yoooo, this is a quick note on a link that made me go, WTF? Find all past links here.