True, and likely lots of other data that should not have been used to train a model. It is at this point hard to imagine a model that is good at reasoning that

True, and likely lots of other data that should not have been used to train a model. It is at this point hard to imagine a model that is good at reasoning that doesn't also have vast implicit "knowledge".