Any grounding that comes from humans is scarce. For training we really want automatic grounding from rules that can be applied trillions of times per second rat

Any grounding that comes from humans is scarce. For training we really want automatic grounding from rules that can be applied trillions of times per second rather than labor intensive human grounding at much lower rates. Of course, grounding from robots is also scarce unless you build a whole lot of robots. Ultimately we need systems that generalize rules from a small amount of grounding data. I guess that describes world models, which large language models currently lack (explicitly, at least).