Kimihiko commented on a stock · Jan 9 09:19

NVIDIA is introducing GenAI to the physical world with Cosmos.

During the CES 2025 keynote, Nvidia's CEO Jensen Huang unveiled the company's latest GeForce RTX50 series graphics cards, the new Nemotron AI platform model family, and AI blueprints for AI-enabled agents.
One of the most intriguing but perhaps least understood announcements was the series of global foundation models and platform features that the company refers to as Cosmos.
Cosmos functions as an extension of Nvidia's Omniverse digital simulation environment, incorporating the digital physical properties of models and systems created in Omniverse and converting them into real-world physical behavior.

Undoubtedly the most eagerly anticipated, most watched, and most attended CES keynote speech in history, Nvidia (NASDAQ: NVDA Mr. Jensen Huang, the CEO, once again revealed a remarkably wide range of announcements across many of the hottest technology Topics, including AI, Robotics, Self-Driving Cars, etc.
Dressed in his trademark black leather jacket with a Las Vegas twist, this Technology industry leader went on to explain the company's latest Geforce RTX50 series graphics cards, the new Nemotron AI base model family, and the AI Blueprint of AI-enabled agents. He also promoted the company's expanded features of the Omniverse Digital Twin/Simulation platform that extends AI into the physical world, new safety certifications for the autonomous driving platform, and a new mini desktop-sized AI supercomputer named Project Digits powered by the Grace Blackwell GPU. Needless to say, it was quite difficult to understand everything.
One of the most intriguing (albeit probably the least understood) aspects of the announcements was a series of base models and platform capabilities named Cosmos. Defined as a set comprising a World Base Model, advanced Talknizer, Safety Guardrails, and an advanced Video Processing Pipeline, Cosmos is designed to bring forth advanced outputs in the physical world from the AI training functions generated from the digital world. In other words, while GenAI typically creates new digital outputs built from training over billions of documents, images, and other digital content, Cosmos helps generate new physical actions (let's call it analog outputs) utilizing data trained in digitally simulated environments.
The concept may be complex, but the real-world implications are simple yet profound. In applications like Robotics, Self-Driving Cars, and other Machinery systems, this means Cosmos can assist these systems in reacting in more accurate, safe, and helpful ways to physical stimuli. For instance, humanoid-style robots can be taught to physically emulate the most effective or safe ways to perform specific tasks such as flipping omelets or picking up and tidying parts on a production line. Similarly, self-driving cars can be taught to dynamically react to various types of situations and environments.
While much of this type of training is currently happening, most of it is manual, involving humans filming the same actions hundreds of times or running millions of miles for autonomous vehicles. Furthermore, even after that, thousands of people spend vast amounts of time manually labeling or tagging those videos. Using Cosmos automates these types of training methods, resulting in significant cost savings, time efficiency, and a broader range of data used in the training process.
At its core, Cosmos functions as an extension of Nvidia's Omniverse digital simulation environment, converting the digital physical properties of models and systems created in Omniverse into real-world physical behaviors. This may seem like subtle differences, but it is an extremely crucial distinction for Cosmos to be able to generate physical outputs utilizing GenAI, allowing the creation of physical outputs based on millions of hours of video content depicting an understanding of the physical world, so-called World Base Models. Cosmos essentially places the digital models of physical objects and environments that can be created in Omniverse into these World Base Models, generating photorealistic video outputs that show how these models are anticipated to react in the world. These videos serve as synthetic data sources for training models in robotic systems, self-driving cars, and other GPU-powered machinery systems, leading to systems that can react more effectively in various environments.
Another crucial point is that Nvidia is offering the Cosmos World Base Model for free to further stimulate development and experimentation in the fields of Robotics and Self-Driving Cars.
In the short term, the direct impact of Cosmos is likely to be limited. This is because Cosmos primarily targets a small number of individuals developing advanced robotics and autonomous driving car applications. However, in the long term, the impact could be significant as it is expected to dramatically accelerate the development of these product categories and improve the accuracy and safety of these applications. What's even more important is that Nvidia is showing signs of anticipating and continuing to plan for larger technology trends like robotics. It also highlights an ongoing but less recognized trend of Nvidia transforming into a software company building platforms for these new applications. For those wondering where the company is headed and how it can sustain remarkable growth, these are interesting and crucial indicators.

Disclaimer: Community is offered by Moomoo Technologies Inc. and is for educational purposes only. Read more