Fields Medal-winning UC Berkeley professor ends graded homework over AI fraud – says in-person exams are the only solution to widespread AI use
Follow us
Add us as a preferred source on Google
Subscribe to our newsletter
A University of California, Berkeley (UC Berkeley) professor has stopped giving his students graded homework, saying rampant AI use has made it pointless. According to the New York Postrenowned mathematician Richard Borcherds said it is now almost impossible to tell the difference between a student-written submission and one created with artificial intelligence. Therefore, he now mainly uses personal exams without electronic devices to assess his students and see how well they are doing in his class.
“Online exams are now a waste of time because some students just use AI to answer the questions. I can’t say how students use AI, but I notice that the average score increases a lot in online exams,” Borcherds told the publication. He also added: “Grading for correctness just adds a huge incentive for them to use AI for homework instead of doing it themselves and punishes students for honesty. And paying to grade AI generated answers is a waste of time and money.”
This problem appeared when OpenAI released the first popular LLM, GPT-3.5, in November 2022. But while LLMs were prone to making mistakes at the time, which made their irregular use easy to see, the technology has improved to the point that it is now very difficult to tell if it was used for written work.
Latest Videos FromTom’s hardware
Several researchers have proven that students regularly get better scores when they take exams online. Although they did not definitively prove that the students used AI to answer the questions, the difference between the scores was significant enough that the researchers concluded that uncontrolled online exams no longer reflect what the students learned in the course.
A Brown University professor determined this earlier this year when his students scored an average of 96% on his take-home midterms—a huge jump from the historical average of 65% to 80%. Because of this, he gave his students their final exams in person, which resulted in only two students receiving a score that was within 10% of their midterm grades, with only one scoring better.
However, not all AI tools are so keen on helping students cheat, with Cursor AI telling a user to learn coding rather than asking it to generate code. Still, this is just one tool in a sea of many, making it nearly impossible to stop students from using AI.
As for UC Berkeley, it had already looked at AI detection software to try to deal with the situation, but found that it was unreliable. Therefore, the university has decided to give professors free rein on their own AI policy, as long as they write their AI rules into the course syllabus. As for Borcherds, he says that if the worst thing AI can do is make it harder for teachers to grade students’ homework, then he’ll be “very relieved.”
Get Tom’s Hardware’s best news and in-depth reviews, straight to your inbox.
follow Tom’s Hardware on Google Newsor add us as a preferred sourceto get our latest news, analysis, & reviews in your feeds.
