Part 7/11:
Research from ARK Invest illustrates that the cost of training AI models is rapidly declining, following laws akin to Moore’s Law—projected at a 75% annual reduction. This steep decline means that data, the most valuable resource in AI, will become both more abundant and more critical.
Tesla's multimodal models—those capable of interpreting text, images, videos, and audio—are particularly suited for robotics and perception in the physical world. Training their humanoid robots, which are still in developmental stages, is increasingly feasible as data quality and quantity improve. The analogy of a fetus emerging into a productive adult underscores the current infancy and potential acceleration of Tesla’s robotics breakthroughs.