No doubt. I don't think anyone's arguing that LLMs have richer, deeper understanding of anything just yet. On the other hand I also don't think it would prove much to vaguely connect a language model to a robot arm and then ask it to do non-language tasks.