The world's two leading technology companies - Alphabet (Google's parent company) and OpenAI - have just announced that their artificial intelligence (AI) models have outstanding math solving capabilities, achieving scores equivalent to a gold medal at the 2025 International Mathematical Olympiad.
This information marks an important turning point in the mathematical thinking ability of AI. Both Alphabet and OpenAI AI systems correctly solved 5 out of 6 problems in this year's exam, surpassing the score threshold to win the gold medal at the competition.
AI is making leaps and bounds in mathematical thinking ability (Photo: iStock).
The International Mathematical Olympiad is considered the pinnacle mathematics competition for high school students. This year, the competition gathered 630 contestants from more than 100 countries and territories.
The fact that AI is rapidly improving its problem-solving abilities has experts expecting that AI is getting closer to being able to solve unsolved problems.
Mathematics professor Junehyuk Jung, a lecturer at Brown University (USA) and one of the researchers at Google DeepMind (Google's AI research unit), said: "When AI can solve complex reasoning problems using natural language, instead of relying on programming languages as before, this opens up the possibility of real cooperation between AI and mathematicians."
Meanwhile, OpenAI’s model is optimized by increasing the computational power during the experimental thinking process. This allows the AI to think longer, deeper, and think in parallel with many lines of reasoning at the same time. OpenAI researcher Noam Brown said the cost of building this type of AI model is “very expensive,” but he did not disclose a specific number.
According to Mr. Noam Brown, the progress in solving the problem given by AI at this year's International Mathematical Olympiad is clear evidence that AI can handle complex chains of reasoning, with the potential to expand into other fields beyond mathematics.
To solve this year’s problem, Google used a general-purpose reasoning model called Gemini Deep Think. Unlike previous AI experiments that relied on programming languages, the latest Gemini model operates entirely in natural language and solves problems within the competition’s allotted time of 4.5 hours.
DeepMind - Google's AI research unit - said that last year they only achieved a silver medal score, last year's AI model was specifically designed to process mathematical content, running in programming languages. This year, DeepMind switched to a general model, using natural language, the results showed that their AI model has come a long way.
Thus, both OpenAI and Google DeepMind have built their own test versions to serve the purpose of solving the questions of the International Mathematical Olympiad. The results are very positive, but neither side intends to release this model publicly in the near future.
This year is also the first year that the Organizing Committee of the International Mathematical Olympiad has collaborated with a number of AI laboratories to test the mathematical thinking ability of current AI models. The Organizing Committee of the International Mathematical Olympiad will verify and certify the results of AI research companies.
“We adhere to the requirements of the competition organizers that AI labs will only publish results after they have been verified by independent experts, and after the students participating in the competition have received the recognition they deserve,” said Demis Hassabis, director of Google DeepMind.
Source: https://dantri.com.vn/giao-duc/ai-cua-google-va-openai-doat-huy-chuong-vang-olympic-toan-quoc-te-20250726093259564.htm
Comment (0)