Mind Matters Natural and Artificial Intelligence News and Analysis
abstract-numbers-background-mathematical-numbers-concept-3d-347387030-stockpack-adobestock
Abstract Numbers background - Mathematical, Numbers concept. 3d rendering
Image Credit: Sashkin - Adobe Stock

AI scores 100% in math contest?: Robert J. Marks explains

Share
Facebook
Twitter/X
LinkedIn
Flipboard
Print
Email

The recent math contest buzz has been that AI catches up with humans to score 100% at top math contest. From Andrew Zinn at Techxplore:

A perfect score at the world’s most prestigious math contest used to be the preserve of a handful of bright young minds each year, but for the first time, artificial intelligence has also earned a perfect score.

Two Chinese tech companies, Huawei and Xiaohongshu, said their AI models had each achieved 100% on the questions given to human competitors at this month’s International Mathematical Olympiad (IMO). July 23, 2026

Ais had won gold medals in 2025 but could not beat human contestants.

The International Math Olympiad, for contestants younger than 20, was held in Shanghai this year with 666 contestants, of whom 7 got perfect scores.

Our director, computer science professor Robert J. Marks offers some thoughts on on what this means:

Of course AI did well on this, for the same reason it will eventually earn perfect or near-perfect scores on the GRE for graduate school admission, the SAT for undergraduate admission, the LSAT for law school admission, and many other standardized tests.

The questions on these exams are drawn from domains that already exist in the corpus of material used to train modern generative AI systems. How do I know? Because someone already knows the answers — the people who designed and grade the tests. If the AI has been trained either on a particular problem or on the methodology needed to solve that class of problems, it can readily regurgitate that knowledge to new instances.

Large language models are stochastic parrots.

Excelling on standardized tests is therefore not evidence of knowledge creativity. It is evidence that the system has become highly proficient at recognizing patterns and applying methods learned from an enormous body of human-generated material.

Perhaps it is the same sort of situation as using AI to write formula fiction or make automatic short films out of it. If humans have already done all the creative work, AI can repackage it in a variety of ways.


Enjoying our content?
Support the Walter Bradley Center for Natural and Artificial Intelligence and ensure that we can continue to produce high-quality and informative content on the benefits as well as the challenges raised by artificial intelligence (AI) in light of the enduring truth of human exceptionalism.

AI scores 100% in math contest?: Robert J. Marks explains