Photo credit: Shutterstock
Over 800 participants from 17 countries and regions reached the last stage of this year’s Alibaba Global Math Competition, the organizing committee announced on Thursday along with the preliminary round results.
Contestants from top global academic institutions including Peking University, the University of Oxford and the University of Bonn will battle for a grand prize of $30,000 during an eight-hour online final round on June 22.
Dinh Vu Tung Lam, a Vietnamese second-year mathematics student at the Hong Kong University of Science and Technology said he leveraged materials from a wide range of sources to help him solve the problems in the preliminary round, including internet sources, physical books, research papers and articles.
Chan Tin Ping, a computer science major from Malaysia, has been participating in Alibaba Global Math Competition since 2022. He’s made it to the final for the last two years.
“I participated in this competition because it’s interesting to do math, especially for people who like math but their research area is quite different to math,” Chan told Alizila.
The top five scorers in the final round will be crowned with gold awards and a grand prize of $30,000 each. Ten silver winners will each claim $15,000 in prize money, while 20 bronze award winners will each receive $8,000.
AI Models Fall Short
This year, the competition widened the playing field to welcome AI models from individual developers, startups and academic institutes to take a crack at the math problems.
Over 500 teams took up the challenge and tapped a wide range of techniques including prompt engineering, AI agent and model finetuning to build the AI models’ math capabilities.
“The introduction of AI brings a new perspective to the competition, helping the public become more aware of the capabilities of AI in logical thinking and problem-solving,” the Alibaba Global Mathematics Competition organizing committee told Alizila earlier this year.
Despite the best efforts of teams, no AI models ultimately made it into the final round.
AI models were able to secure points by using knowledge retrieval when it came to math solutions and proof questions, according to the organizing committee.
However, they began to experience hallucinations, generating false or illogical answers when the questions required more in-depth analysis, especially in multiple-choice questions.
Role-play prompts significantly boosted AI performance.
One contestant gave the prompt to imagine AI as a math master, competing for a reward of $300,000. It has improved the score by 20%, according to the testing results.
Discover more about Alibaba’s commitment to research and development here
