The builders of GPT-4 claimed that this system can exhibit “human-level efficiency” on sure exams. Whereas the chart of assorted educational check scores appears spectacular at first, the software program failed each AP English exams with a 2 out of 5. It scored extraordinarily effectively on checks the place all that a pc program wanted was the precise info, which isn’t very spectacular to anybody who has used Google previously few many years.
Once more, I don’t suppose that we’re immediately going to be bleeding legal professionals. As somebody who has really carried out moderately effectively on the LSAT, I can even let you know that taking the check and practising legislation are very various things. Legislation exams are about sample recognition, which is precisely what AI is constructed to do. AP exams work equally, and you’ll reply a big bulk of questions by phrase affiliation.
I’m really glad that GPT-4 can rating effectively on the extra rote checks that solely measure how lengthy you’ve spent drilling flashcards. I’ve by no means believed in exams that require college students to memorize tons of knowledge. Its failure to go English exams is proof that essential considering is an important talent that you could get out of training.