Saturday, November 30

Is ChatGPT smart enough to get an engineering degree?

Deutsche Welle avatar

By Deutsche Welle

Nov 29, 2024, 9:31 PM EST

Researchers from the Federal Polytechnic School of Lausanne (EPFL), one of the most prestigious universities in Europe, put the ChatGPT artificial intelligence (AI) systems to the test of various studies of that institution, observing that they answer most of the questions correctly.

AI approves various courses at Swiss university

In the study, published in the scientific journal Proceedings of the National Academy of Sciences (PNAS), researchers from the EPFL Faculty of Computer and Communication Sciences analyzed 50 EPFL courses on a large scale to evaluate the current performance of extensive language models (LLM) in the evaluation of higher education courses.

The selected courses are part of nine online bachelor’s and master’s programs and cover a wide range of disciplines, including computer science, mathematics, biology, chemistry, physics and materials science.

ChatGPT 4 and 3.5 language models correctly answer about 65% of questions using immediate strategies and basic databut with increased knowledge they can achieve up to 85% correctness, a surprising result according to scientist Anna Sotnikova, from the EPFL Language Processing Laboratory, who led the study.

“If the study were repeated today the results would be even better,” the researchers warned, taking into account the continuous advances of this technology.

Dangerous “shortcuts” for students

“Could ChatGPT get an engineering degree? Assessing the vulnerability of higher education to AI assistants.” That is the title of the study, which sought above all to examine the effects that the use of AI can have on higher education students.

The researchers’ concern is that the use of these language models affects the educational process of university students.necessary to increase their academic abilities.

“There are fears that if the models are as capable as they seem, students will use them to take ‘shortcuts’ in this process necessary to learn new concepts, which could lead to weaker knowledge bases that will hinder the future absorption of more complex knowledge,” those responsible for the study warned.

“We conceptualize these challenges from the perspective of vulnerability, the potential for university assessments and learning outcomes to be affected by students’ use of generative AI,” they explained.

“Our results call for reviewing the design of program-level assessment in higher education in light of advances in generative AI.”they concluded.

rml (efe, PNAS)