In terms of bits seen during training, LLM are more akin to a 3 year old. Robots roaming around and learning to interact with the in environment and sharing knowledge might be a game changer, assuming that the current methodologies are sufficient (LLM + RL)