Jiang Ping Photo: CCTV

 
None of the 563 artificial intelligence (AI) teams from the world's leading universities, colleges and enterprises reached the final round in the 2024 Alibaba Global Mathematics Competition, in which a 17-year-old Chinese girl from a vocational high school ranked 12th, sparking heated discussion on Chinese social media platforms for days.

The web-based competition has no diploma requirements and is open to math enthusiasts from all walks of life to participate, and this year's competition was open for the first time to AI powered participants.

Jiang Ping, a fashion major trainee at a vocational high school in East China's Jiangsu Province, is unique as the first vocational high school student to reach the final round, alongside students from top institutions including MIT, Tsinghua, and Peking universities.

The average score of the AI teams participating in the competition was 18 points, which has now caught up with the average of human players, according to the official statistics of the competition, reported by media, but AI's highest score was only 34 points, a far cry from the highest score of 113 points achieved by a human competitor.

The AI marking review given by the competition organizing committee also highlighted the AI's mathematical shortcomings, including the weak logical reasoning ability and the difficulty in providing a complete score for proof questions.

The AI program with the highest score among AI teams has used multiple large models to perform multiple rounds of self-questioning, self-answering and self-verification to find the optimal solution to the problem, and then allow the AI to select the only answer, according to the Economic Observer.

The two teams that came second and third among the AI teams designed their program based on a multi-agent model, where one of them is to let multiple large models play different roles in thought analysis, problem solving, evaluation, and output the answer by using Python, and the other is to dynamically adjust multiple stages including inference, logic verification and interpreter, and select appropriate reference cases for each sub-problem, and finally give the answer after multiple verification and integration.

According to the report, a technical director from an internet enterprise said that although in the calculation-intensive problems with clear rules, the ability of AI model performance can be better than human, but in the need for deep logical reasoning and a high degree of innovative thinking, human players still occupy an absolute advantage.

From the list for the final round, the AI model's ability to solve problems is still struggled to surpass that of humans, but the director said the math competition duel will possibly push human understanding of the nature of AI one step further, and is even expected to inspire the development of new mathematical theories and AI technology.

Jiang had demonstrated strong math skills in middle school, but her performance in the high school entrance exams was not as strong, leading her to enroll in a vocational school, according to previous report.

However, her math proficiency caught the attention of her teacher, Wang Runqiu, who recommended that Jiang undertake a self-study course for college-level math courses. Wang, who holds a master's degree in math, ranked 125th among the 801 finalists.

As vocational high schools and colleges often bear the stigma of being seen as "places for failed students" in China, Jiang has been going viral on Chinese social media for days, with many Chinese citizens shocked at her talented math skills and sending her well wishes. 

The official accounts of Tongji University, and Jiangsu University on China's X-like Weibo also said that they welcomed her to apply to these universities.

Jiang said she used to think that she wasn't qualified to participate in such contests, but maybe it can be an opportunity for her to prove herself. She added that she sees fashion design as her Plan A, with math as her Plan B, but she still wants her Plan B to remain viable.

The 801 competitors, who outsmarted AI in this aspect, will answer questions within any 8-hour period on June 22, and the winners will share a prize of 4 million yuan ($550,000), according to a previous report.

Global Times