It suggests that it would. It is suggesting that the quality of the models’ ability to correctly understand a fast food order indicates its fitness to be used in teaching, but I would say that even a model which can flawlessly take a cheeseburger order—even a model which can flawlessly mimic the action of teaching—wouldn’t be appropriate for the task. The quality of its output is unrelated to its fitness to the purpose.