What can we learn if AIs are able to pass exams?

The debate over artificial intelligence and the traditional methods of testing academic skills, such as exams, quickly became a reality. Researchers at Illinois Institute of Technology and Stanford University submitted GPT-4, an AI model developed by OpenAI in spring 2023.

The robot answered nearly 76% questions with multiple choices. This is 7% higher than the average candidate. You have performed both well on MCQs as well as in writing argumentative responses.

This is a special test.

All university specialties are concerned

Researchers intuitively believed that algorithmic models would be better suited to assimilate knowledge from scientific fields.

The academics of the Ecole Polytechnique Federale de Lausanne published a report (New Window) at the end November 2024 on the application of generative AI to the exams of the 50 EPFL courses, mainly those in computer science, math, biology, physics, or materials science. It is a test to get a Bachelor degree (Bac+3) or a Masters degree (Bac+4).

Learning like a student

As a first step, AI needed to be trained with information from university classes, past exams, and assessment models. It was necessary to gather the information that a serious student would seek in order to study for an exam. It is time to acquire knowledge.

After the authors had completed their study, they began to think about how best to interrogate the AI. In order to get the best answers, the “prompts”, or written instructions are used.

Researchers have found that “prompts” are most effective in interacting with AI.

Scientists found that the most simple formulations (i.e. Scientists have found that the most basic formulations, i.e. As soon as prompt writers are more knowledgeable about the topic, the score increases to 85%.

An opportunity to reconsider teaching

These results aren’t simply a demonstration that machines can pass exams. This is a great symbol of how technology has evolved, but it’s not very useful to us.

Above all, it’s about how to make the best use of technology.

It’s a chance to reconsider the ways we teach. Students who rely too much on AI-generated tools for their assignments will not acquire the basic knowledge they need.

It is only with this kind of expertise that you can learn from the very best.

The kind of reasoning that will later be able to support a more complex one. It may seem like a time-saving move, but it can actually make the student less able to understand more complex concepts. This results in a loss of skills. AI models will only improve and become more specialized. And so become more expert assistants.

Valuing the specificity of the human mind

In a world of AI, this is a great opportunity to highlight what makes human minds special. This is the ability to use different types of information to solve a problem. This will likely mean that the way skills are measured in education needs to be rethought.

Exam subjects will no longer be a simple recitation of information from the theoretical courses. It is the same as working on projects, where you need to draw upon a variety of ideas and skills.

It means we will be in project mode.

The universities will need to rethink their own teaching techniques. Questioning the subjects and areas that should be included in an initial curriculum. On revised forms of assessing knowledge and on ways to cross-utilize different know-how.

Calculators can be a great tool for students if they are used correctly. The same should be done with AI. We should do the same with AI.

Streamline Tasks Using OpenAI and Python