ChatGPT barely passed the law and business postgraduate exam.
CTOnews.com, January 27 (Xinhua) researchers have had to run more machine tests because many people are worried that OpenAI's ChatGPT will help students cheat in exams.
In a new study, law professors at the University of Minnesota asked ChatGPT to try graduate exams in four courses at their school, and AI passed all the exams with an average score of C +.
In addition, in another paper, Wharton professor Christian Terwiesch found that ChatGPT passed the school's business management exam with a B-to-B.
However, as a result, the performance of artificial intelligence is not consistent. The University of Minnesota team points out that ChatGPT is good at solving "basic legal rules" and summing up theories, but is poor at trying to identify issues relevant to the case.
The Wharton Terwiesch points out that it is "surprising" in dealing with simple operational management and process analysis problems, but cannot deal with advanced process problems and even makes mistakes on sixth-grade math problems.
In other words, although AI can barely pass the postgraduate exam, there is still considerable room for improvement.
University of Minnesota professors say they do not adapt to specific courses or questions, believing that students can customize to get better results, while Wharton professors say AI is good at changing answers according to human guidance, and while the ChatGPT itself may not do well in exams or papers, cheaters can make the system produce more standard answers with a little attention.
Some opponents of AI say schools should restrict the use of this technology to prevent ChatGPT-based cheating. They also suggest changing the test bank to prevent the use of artificial intelligence or to raise the challenge to AI. The University of Minnesota says students still need to learn "basic skills" instead of relying on robots for help.
Terwiesch believes that this technology can eventually save time that can be spent on students. What do you think of CTOnews.com?
