Fei-Fei Li argues spatial intelligence is the next AI frontier, a much harder problem than language due to the physical world's complexity and the lack of readily available training data. A Hacker News discussion explores this challenge. One commenter highlights the difficulty in representing arbitrary spatial relationships scalably on computers, citing unsolved computer science problems and "dark state-of-the-art" research not in the public domain. Other commenters debate whether LLMs can truly grasp spatial dynamics, with some arguing they are inherently limited while others point to promising multimodal models integrating vision and action. The need for interaction with the real world and potentially leveraging simulations for data generation is discussed. Some also suggest that less complex, heuristic-based approaches might be sufficient for many real-world applications. Others share projects related to AI for satellite imagery and efficient 3D modeling, highlighting the current research landscape.