While GPT scored only 60 out of 100 on the Korean CSAT math exam, it solved all the math practice questions from the College Board perfectly.
An article discussing
topics related to the Korean CSAT exam :
https://sunrisensetkr.blogspot.com/2025/01/ai-tutors-how-do-they-score-on-2024.html
I tested GPT 4o's
ability by having it solve U.S. SAT math problems. The College Board's
Full-Length SAT Practice Test: Bundle 1 includes 54 math problems, all of which
GPT answered flawlessly. Out of the 54 questions, 6 were short-answer, while
the remaining were 4-option multiple-choice questions.
All the calculations
were not only correct but also carried out effectively. While some steps could
have been shortened, the overall process was smooth without any significant
problems.
As previously stated, GPT performed poorly on Korean CSAT
math problems. GPT answered all the low-scoring 2-point questions correctly but
struggled with most of the high-scoring 4-point questions. Compared to the U.S.
SAT, even the 2-point questions on the Korean CSAT often required more complex
calculations.
While the U.S. SAT focused on practical arithmetic skills
for everyday scenarios, the Korean CSAT demanded higher-level problem-solving
without providing real-world context. If math problem-solving were likened to a
fitness test, the U.S. SAT would focus on exercises like push-ups, step tests,
or long jumps to check general strength and stamina, while the Korean CSAT felt
more like a powerlifting competition involving deadlifts and weighted pull-ups.
The College Board's practice question:
https://satsuite.collegeboard.org/practice/practice-tests/paper
Comments
Post a Comment