Recording at ai.lu.se

Topic: GPT in examination - experience report and experiment design

When: 6 March at 12.00-13.00

Where: Online 

Speakers: 

• Johan Grönqvist, PhD student, Automatic Control, Lund University
• Julia Adlecreutz , PhD student, Automatic Control, Lund University 
• Felix Agner, PhD student, Automatic Control, Lund University

Spoken language: English

Abstract

Johan Grönqvist presents his experience using ChatGPT to solve written exams in automatic control. In 2023, it failed badly. In 2024, its responses to two exams in different courses were graded by the same people grading students exams. ChatGPTs responses received grades 4 and 5 respectively (5 is the best grade and 3 is required to pass).

Then Julia Adlecreutz and Felix Agner present a plan for verifying if this result holds for a user without the necessary background in automatic control. In particular, can a student with a basic engineering background together with GPT-4 solve the exam in the basic course on automatic control? After that we leave the floor open for questions and discussion.