The deployment of Artificial Intelligence (AI) in drafting theses, articles, or crafting congratulatory messages no longer sparks wonder. Yet, the question of whether a neural network could successfully navigate the complexities of the Unified State Exam (EGE) and secure a university place remains a topic of keen interest.
Specialists at the Moscow Institute of Physics and Technology (MIPT) embarked on an experiment to evaluate the prowess of the generative neural network, Chat GPT-4 Turbo, against the EGE’s formidable physics section.
It was determined that the AI fell short of the 85-point threshold requisite for MIPT admission.
MIPT lecturer, Leonid Koldunov, highlighted the neural network’s deficiencies, attributing them to inaccuracies in final calculations, terminology usage, and the comprehension of physical processes. Furthermore, the AI’s logic was found wanting, as it struggled with interpreting the conditions set by various problems.
Igor Pivovarov, the principal analyst at MIPT’s Centre for Artificial Intelligence, noted that generative neural networks, akin to Chat GPT, grapple with mathematical and logical challenges. However, he posited that with tailored context and training, AI’s precision could witness improvement.
The experiment’s outcomes underscore that artificial intelligence, while promising, has yet to achieve the human-like level of logic required for tasks of this nature.
Photo: freepik.com