Some elite AI researchers say language is limiting. Here's the new kind of model they are building instead. (www.businessinsider.com)

Elite AI researchers, including Fei-Fei Li and Yann LeCun, are shifting the focus of artificial intelligence from large language models (LLMs) to innovative "world models" that mimic human mental constructs. At World Labs, co-founded by Li, the goal is to elevate AI's spatial intelligence, allowing systems to predict outcomes and reason about 3D environments similarly to human intuition. This contrast to LLMs highlights a movement towards creating AI that not only processes language but understands and interacts with the physical world, enhancing applications in robotics, creative industries, and more. The development of these world models faces significant challenges, primarily due to the limited availability of high-quality spatial data compared to the abundant textual data used in LLM training. LeCun's startup, Advanced Machine Intelligence, aims to bring this vision to fruition by creating systems with common sense, reasoning abilities, and persistent memory, which he believes are essential for AI to rival human intelligence. Meanwhile, other companies like Moonvalley are exploring generative AI video models as foundational elements for world modeling, highlighting a collaborative yet competitive landscape in advancing AI capabilities beyond current language-centric systems.