by Alimat Aliyeva
At the entrance exams for the University of Tokyo and Kyoto
University, an AI system based on ChatGPT demonstrated results that
in some cases exceeded those of the top human applicants,
AzerNEWS reports.
During testing with the OpenAI ChatGPT 5.2 “Thinking” model, the
system was given real exam papers from these universities. The
questions were presented in image format, simulating the conditions
of an actual entrance examination.
The most striking result was recorded at the University of
Tokyo’s medical faculty, where ChatGPT scored about 50 points
higher than the top-ranked applicant. The AI also achieved the
highest score in mathematics, showing particularly strong
performance in problem-solving and logical reasoning tasks.
Overall, the system performed especially well in the natural
sciences, scoring 503 out of 550 points, surpassing the best human
result of 453. In the humanities section, it also performed
strongly, achieving 452 out of 550 points, which was above the
average passing level.
However, the results were not uniform across all subjects. In
English, the model reached around 90%, while in more interpretive
tasks—such as essay writing or questions on world
history—performance dropped significantly, to roughly 25%. This
suggests that while the AI is highly effective in structured,
fact-based reasoning, it still struggles with open-ended analysis
and nuanced argumentation.
In the Kyoto University exams, ChatGPT also outperformed top
candidates, scoring 771 points in the Faculty of Law (compared to a
top human score of 734) and 1,176 points in the medical faculty
(against 1,098 for the leading applicant).
Researchers note that this progress is especially remarkable
given the short timeframe of development. Back in 2024, earlier
versions of ChatGPT could not even meet the minimum entrance
requirements for the University of Tokyo. By the following year,
however, newer models were already capable of passing—and now, in
some cases, surpassing—top human performance.
An interesting detail is that despite these impressive results,
experts emphasize that exam success does not necessarily mean true
understanding or professional competence. AI can excel at pattern
recognition and structured problem-solving, but it still lacks
real-world experience, ethical judgment, and creative
intuition—qualities that remain essential in fields like medicine
and law.
Some analysts also point out that if such systems continue
improving at this pace, universities may eventually need to
redesign exams themselves, focusing less on memorization and more
on skills that are harder for AI to replicate, such as original
research, practical decision-making, and interpersonal
reasoning.
AloJapan.com