Part 4/10:
A critical part of the discussion centers around the limitations of current AI, particularly concerning its understanding of the physical world. Leedan emphasizes the necessity of integrating a common sense understanding of our environment into AI systems. He draws attention to prior discussions with experts like Jan LeCun, who assert that when AI models are trained solely on text, they cannot develop a contextual understanding necessary for real-world applications.
To address this gap, Nvidia is pioneering the development of world foundational models that teach AI about physics and real-world interactions through multimodal inputs beyond text. This includes using simulation to allow AI to process real physical experiences, enriching their understanding of the world around them.