Besides the AI being able to interpret the questions, which is the entire point of a language model, isn't this just the same as someone googling the answers to the exam?