Another issue that my daughter noticed in her senior HS classes was that students who paid for the latest and best AI models naturally got better answers (and better teaching / coaching if they used it that way).
2 Likes
Follow up interview of the Brown Econ prof on the cheating in his class. Many new details:
Thank you for posting this link.
I’m honestly surprised at the administration’s lack of response in lieu of so much evidence.
1 Like
Same! I wonder if the publicity will change things?
I also appreciated the prof’s details that make it sound like plenty of cheating was going on, especially this exchange:
Evan Goldstein: You have for many years taught an advanced undergraduate course in mathematical economics. This year, for the first time, you decided to offer the midterm and final as take-home exams. Here was the grade distribution for the midterm: Among the 86 students in the class, the average score was 96. Forty students — nearly half — earned a perfect 100. What was your initial reaction?
Roberto Serrano: I immediately knew that something was fishy. In previous editions of the course, the average grade in the midterm ranged from 65 to 80. Historically this course has attracted very strong students, and the enrollment was never too high — up to 31 and as low as eight. So my first surprise was that there were so many students registered. I assume that was because they looked at the syllabus and saw that the two exams were going to be take home, which I had never done in the course.
Goldstein: When you and your graders look more closely at the exams, you also detected evidence of AI use.
Serrano: I asked my assistants to run the exam through ChatGPT, and ChatGPT did a decent job solving the exam. However, some of its answers were really odd. For example, there was one question in which the student was asked to prove something, and if you understand the material and know what you’re doing, it’s a direct argument. ChatGPT, for some reason, constructed this very convoluted argument that eventually took it to the right conclusion but in an odd way.
Lo and behold, that convoluted argument appeared in quite a few of the exams. So I told the class — imagine how uncomfortable this was — about my strong suspicion that there had been massive cheating in the midterm. I said, “I’m not going to declare it void for now. I’m going to give the class a chance to prove me wrong. If the distribution of grades in the final exam looks roughly similar to the distribution of grades on the midterm, then I’ll count the midterm. If not, I will declare the midterm null and void. Also, the final will be in person.”
Goldstein: You administered the final in mid-May. The average score dropped from 96 to 48. Whereas 86 students took the midterm, only 59 even showed up for the final. Of the students who did not show up, 22 had scored 100 on the midterm.
Serrano: Between the midterm in March and the final in May, there was a consistent flow of students dropping the class — many of them had scored 100 on the midterm. Of the 59 students that took the final, 19 of them failed, so they also failed the course. Quite a few showed up, signed the exam, and turned it in blank. The average grade for this final was by far the lowest in the history of this course.
3 Likes
More on that class at Brown from Brown Professor Suspects Most of His Class Used AI to Cheat
2 Likes