...but that's not what the models are. You can interrogate them on the rules of chess, and they'll (statistically likely) give you a decent breakdown of the rules. Evidently the rules are in their training material, they just fail to apply them in the manner of an intelligent system for some reason or another.