According to Foresight News, Nvidia CEO Jensen Huang has announced the launch of a comprehensive series of models based on Llama, utilizing the NVIDIA Llama Nemo Tron language foundation model, named Nemotron. Additionally, Nvidia introduced NVIDIA Cosmos, a foundational model designed to comprehend real-world scenarios, also employing the NVIDIA Llama Nemo Tron language foundation model.
The Cosmos model is capable of processing prompts in the form of text, images, or videos to generate virtual world states, which serve as video outputs tailored to the unique requirements of autonomous driving and robotics applications. Developers can leverage Cosmos to produce AI feedback for reinforcement learning, thereby enhancing strategy models and testing performance across various scenarios. Nvidia is collaborating with the global developer ecosystem to advance the next wave of physical AI.
Cosmos is the world's first foundational model trained on 20 million hours of video, with a focus on physical dynamics and human interaction.