SerpApi.Org
Nvidia is setting its sights on a world in which humanoid robots work alongside people, and it’s probably coming sooner than you think.
They may not have been delivered with the sonorous tones of Star Trek’s William Shatner or Patrick Stewart speaking over the show’s iconic intro, but when Nvidia’s CEO Jensen Huang talked about the “frontier” of AI this week, his words still carried significant weight.
That’s because AI is seemingly everywhere and in everything at the moment, and Nvidia‘s chips are on a (much longer than five year) mission to power as much of the revolution as possible. Huang was technically unveiling a new AI model—which you may think is not really news in a time when new AI models are popping up like weeds through cracked pavement. But the new Cosmos 3 Edge model Nvidia revealed is something different. It’s a “world model.”
Speaking in Japan, Huang’s comments were aimed partly at the nation where Nvidia is creating a coalition with some of the biggest tech manufacturing names, but they’re also a significant sign about where AI is going in the near future: into robots.
“The next frontier of AI is in the physical world,” Huang said in a statement, adding that the coalition being created to tap physical AI is a “once-in-a-generation opportunity for Japan.” He noted that, in many ways, “Japan invented modern manufacturing,” and as he sees it, the country now has an “opportunity to reinvent it for the age of intelligent industries.” But Huang’s words will echo far beyond Japan’s borders.
While OpenAI, Anthropic, and other players in the game release large language models to power their linguistic-centered AI tools, Cosmos 3 Edge is designed to help systems perceive and then navigate and manipulate physical environments in real time, CNBC reported.
This kind of “world model” takes in very different types of data to the often text-centric material used in AI chatbot training, technically teaching the AI physical details about things in our human world. In a press release, Nvidia claims Cosmos “combines vision reasoning, world generation, and action prediction.” It’s all about understanding and generating “text, images, video, ambient sound and actions with leading physics accuracy.”
The goal is to allow robots to maneuver in the real world and manipulate items they encounter. Think of a world model as a kind of enhancement of the AI that gives Teslas their real-time self-driving powers. Robots—be they humanoid or more purpose-designed, like welding bots in car factories—will need this kind of understanding if they’re to move into our workplaces and homes in a safe way, and if they’re to have the ability to take on different jobs they’re assigned in real time.
This is where the Cosmos 3 Edge gets its name, because it’s designed to do “real-time inference at the edge.” This means that significant AI processing will happen inside robots, at the “edge” of the digital network they’re connected to, allowing the robot to, for example, predict which way something that it throws will go.
Why should you care about this?
Robotics company 1X recently revealed a spectacularly dexterous robot manipulator it had created for its humanoid bots. The new hand is capable of unsettlingly human-like movements, and is both delicate and precise enough to pick grapes from a bunch or to speedily manipulate a gaming controller. In its press release, 1X suggested that human-like hands were one of the last barriers to allowing robots to work in human workspaces, and that the only limit left was the sophistication of the AI code.
Huang proclaiming physical AI as the “next frontier” his tech giant company is going to conquer is another giant sign that the robotic revolution may be arriving much sooner than you think. Remember, ChatGPT only really arrived in late 2022… and now sophisticated AI tools are everywhere online.
If your company’s plan for the next couple of years doesn’t involve robots, that’s probably fine, but you may want to get your planning team thinking about how robots will be working for you inside five to 10 years.
This post originally appeared at inc.com.
“Click here to subscribe to the Inc. newsletter: inc.com/newsletters"
