Part 4/13:
He explains that Tesla’s AI chips, such as the upcoming AI5 and AI6, are versatile for both inference (what runs in the vehicle) and training (the heavy lifting behind model development) — key for their synthetic data generation approach, where the car trains itself on imagined scenarios. Elon suggests that rather than developing separate chips for training and inference, Tesla will consolidate efforts into a single, scalable solution—what he refers to as "Dojo 3," which involves assembling multiple AI6 chips for both inference and training.