Open books exams are bounded by time, something an AI has plenty of, ridiculously more than a person, to say it’s giving them an advantage is missing the absurdity of the situation as it also had much more time to train and “study”, and yet it produces the results it does.
Open books exams are bounded by time, something an AI has plenty of, ridiculously more than a person, to say it’s giving them an advantage is missing the absurdity of the situation as it also had much more time to train and “study”, and yet it produces the results it does.
What results? We are assuming it passes med school, so the result must be good.
The results people get from prompting current models, which hallucinate, a lot, and that’s not really a fixable issue
So then ChatGPT wouldn’t pass the exam, nothing to worry about.