It’s doubtful that training on static representation of dynamic physical systems would make the text model be able to reason about changing physical environment

It’s doubtful that training on static representation of dynamic physical systems would make the text model be able to reason about changing physical environments described in words/question. It would likely continue producing word salad output, but prove me wrong.