Vietnamese at Google creates AI to solve Math problems that 'defeat' international Olympic gold medal

AI 'defeats' international Olympic gold medal

3 weeks ago, the team of Dr. Minh Thang - a senior researcher at Google (USA) successfully developed the AlphaGeometry 2 version. The most outstanding feature of this version is that the level of solving Math problems surpasses the gold medalists of the International Mathematical Olympiad (IMO) in solving Mathematical Geometry. "If IMO gold medalists can solve 40/50 problems, AlphaGeometry 2 will solve 42/50 problems," Dr. Thang happily shared.

20 years ago, Dr. Thang studied in the specialized math class at the Gifted High School in Ho Chi Minh City, a student of Mr. Le Ba Khanh Trinh - the first Vietnamese to win an IMO gold medal (in 1979). Mr. Trinh is also the only Vietnamese to date to have won the special prize of this competition thanks to his short, impressive solution to geometry problem number 3.

Inspired by his talented teacher, Mr. Thang always pondered the question, “Can our AI solve the problem of Mr. Le Ba Khanh Trinh in the past?”. This question was the source of Mr. Thang’s determination to build AI software to solve math problems.

Born in 2022, Alpha Geometry went unnoticed by everyone, as the early software could only solve a few small math problems.

Dr. Luong Minh Thang (right) shares with experts at the GenAI Summit 2024 Artificial Intelligence Conference.

After 2 years of development, in January 2024, AlphaGeometry officially “debuted” in the AI ​​world with a research paper published in the world-renowned journal Nature. The article featured 3 Vietnamese PhDs: Luong Minh Thang, Trinh Hoang Trieu and Le Viet Quoc, along with two foreigners, Dr. Yuhuai Wu and Dr. He He.

In this version, AlphaGeometry surprised many people by using neural language models and logical reasoning combined with large-scale synthetic data to solve IMO (international mathematics competition) problems.

“However, at this time, AlphaGeometry still cannot solve the legendary problem of Mr. Trinh. His problem has moving points connected to each other but AlphaGeometry 1 cannot describe that movement,” Dr. Luong Minh Thang pondered, continuing to find ways to perfect and upgrade the software.

20 years ago, Thang missed out on the IMO when he finished 8th in the national team, while only 6 candidates were selected to participate in the international arena. Despite his unfinished dream of Mathematics, he turned to artificial intelligence (AI) when he entered university. This turning point brought him to the IMO 2024 in a special way - not as a candidate, but with the AI ​​AlphaGeometry software developed by his team, competing with mathematical talents from more than 100 countries in Bath, UK.

“In July 2024, for the first time, our team brought AI to participate in the International Mathematical Olympiad IMO and won a Silver Medal,” Dr. Thang shared.

As a special guest at the competition, Thang had a great memory introducing AlphaGeometry to contestants from all over the world, and also met teacher Le Ba Khanh Trinh right at the competition.

9cbb024943b3f2edaba2.jpg

Our goal in the future is to develop a program, an AI version that can solve 6/7 millennium problems that the world has not solved.

Dr. Luong Minh Thang

Thang said that AlphaGeometry is strong in solving geometry problems, but the IMO 2024 exam matrix only has one geometry problem, the rest are algebra, arithmetic and combinatorics. AlphaGeometry is just 1 point short of winning the international Olympic gold medal, which is what motivates him to continue leading the team to develop AI with superhuman reasoning capabilities.

In particular, the desire to create software that can conquer all geometry problems at IMO helped Dr. Thang's team successfully develop AlphaGeometry 2 - released 3 weeks ago, successfully solving Mr. Trinh's problem.

Dr. Luong Minh Thang believes that AlphaGeometry marks an important milestone towards human-like intelligence and self-learning software. This is the prerequisite for achieving AGI (Artificial General Intelligence) - systems that can learn all knowledge and perception, surpassing human intelligence.

"Our goal in the future is to develop a program, an AI version that can solve 6/7 millennium problems that the world has not solved. Even if AI can win the Fields Medal like Professor Ngo Bao Chau, it would be great," Thang said.

According to Thang, people previously thought that it would take a few more years for AI to solve international math problems, but his team's software has now done so. Thang and his colleagues' goal is not just to stop at AlphaGeometry or solving math problems.

“We want AI to reach new heights, not just imitating humans but having new reasoning, searching and creating practical solutions for the world in many different fields such as Physics, Chemistry, for example drug discovery,” Thang shared.

Dr. Luong Minh Thang has more than 50 research articles, 40,000 citations and 20 patents.

At Google DeepMind, Dr. Luong Minh Thang built the most advanced models in both language (QANet, ELECTRA) and vision (UDA, NoisyStudent). He co-founded the Meena project - the world's best chatbot in 2020, which later became Google LaMDA, Bard and now Gemini - Google's main AI product. The Meena chatbot platform is still the world leader and competes with GPT chat.

Officially working at Google Brain since September 2016, Dr. Thang started with research expertise in machine learning and natural language processing. Thang is the only Vietnamese in the key research team on the Parti model (Pathways Autoregressive Text-to-Image), automatically converting text into images at Google Brain.

To date, he has more than 50 research papers, 40,000 citations, and 20 patents. According to Thang, at Google, there are researchers who just want to write papers, but his team thinks more about creating papers with great impact and at the same time can make applications that impact billions of people.

For example, in 2021, Thang researched an application to change Google's search algorithm and was one of the top 10 best improvements Google made to help billions of users more conveniently collect information.

“That requires careful thinking because at my level there are many problems that can be solved. The key is to find the problem that most people need, even if they haven't thought of it yet, but it must be applicable in practice,” Thang emphasized.

Starting out as a student majoring in Mathematics at Nguyen Binh Khiem High School for the Gifted in Bien Hoa Province, Thang has been passionate about solving math problems since he was a child. While studying at the Gifted High School of Ho Chi Minh City National University, Thang met teachers Le Ba Khanh Trinh and Tran Nam Dung, who brought their university teaching style down to high school level, helping Thang become more open-minded about thinking about solving math problems. With a solid foundation in mathematics, the Vietnamese male student went to Singapore to pursue Computer Science at the National University of Singapore with a full scholarship.

Here, Thang began to study natural language and translate it into many languages, so he clearly felt the power of artificial intelligence. Thang was retained as a research assistant at the school until 2011 when he received a PhD scholarship from Stanford University, USA.

At the world's leading school, Thang was mentored by Professor Christopher Manning, a renowned figure in the field of applying deep machine learning to natural language processing.

Dr. Thang with mathematician Terrence-Tao (middle) and his wife, Wendy Uyen Nguyen (right).

During his five years at Stanford, there were times when Thang's level of optimism was at its lowest. During the first three years, he worked in many different fields but his research results did not achieve great success.

At the end of his third year, Thang interned at Google and found his path in artificial intelligence translation. At that time, Thang was fortunate to work with three prominent people in the field of AI: Dr. Le Viet Quoc, one of the most cited people in the world about AI; Ilya Sutskever - Co-founder and former chief engineer at OpenAI; and Oriol Vinyal - Vice President of Google DeepMind.

Finding a direction for himself, Thang immediately wrote his thesis in this field and focused on developing it. When he became an official employee at Google, Thang realized that many groups were competing with each other, so he thought of switching to the field of super intelligence to expand his search for new knowledge.

Working on things that are at the forefront of artificial intelligence globally, Thang considers this research group to be the “pioneering brain” of AI in the world. That is both a pressure and a challenge, but also an inspiration for the whole group to conquer new goals in their careers.

Putting Vietnamese AI on the world map

Assessing that Vietnam has great potential in the field of AI, Dr. Luong Minh Thang aspires to contribute to putting Vietnamese AI on the world map.

Over the past five years, he and his colleagues, who are Vietnamese experts, have co-founded the non-profit organization VietAI, training a total of more than 4,000 high-quality AI engineers. Among them, 4 young engineers became the first Google machine learning development experts in Vietnam.

Every morning, Thang wakes up early to take his son to school, then goes to Google headquarters. He starts his morning with tasks that require the most creativity and alertness, such as reading recently published scientific research, thinking of new ideas, and programming and writing code.

Occasionally, Thang organizes lunch with the team, discussing ideas or reading scientific articles with colleagues. During lunch breaks, he meets with senior managers and groups to give direction.

Dr. Thang and his colleagues in the AlphaGeometry founding team work at the office.

Screen Shot 2025-03-17 at 12.01.55 PM.png

Aspiration to train hundreds of thousands of Vietnamese AI talents with the motto "zero to hero".

Dr. Luong Minh Thang

In the late afternoon, he usually goes jogging and comes home for dinner with his family. His wife is Master Wendy Uyen Nguyen, Director of Global External Relations, founder of the Stanford Institute for Microbiology and Epidemiology (Stanford University). The 8X scientist confided that the couple is very compatible. His wife always encourages and accompanies him to Vietnam to do projects to connect the bridge of AI and medicine between Vietnam and the US.

In August 2024, Thang invited Mr. Jeff Dean - "the wizard" at Google to organize an AI conference in Ho Chi Minh City, Vietnam, attracting thousands of attendees to spread new knowledge about the field and make recommendations for AI development in Vietnam in the future. Thang's wish is to train hundreds of thousands of AI engineers by 2030, and in 2025 alone, he aims to train 1,000 high-quality Vietnamese engineers in the field of generative AI.

His AI Institute has built an AI curriculum framework, inspired by Stanford University and Google, to train human resources in Vietnam since 2018. Thang realized that universities in Vietnam teach AI but the knowledge is not updated, still quite old.

His biggest wish is to be able to train AI in a connected way, between universities, between high schools and universities, and between his institute and universities, and especially between universities and businesses. That creates horizontal and vertical connections in training, students can learn AI in different places but all are recognized.

In the near future, Thang plans to seek foreign scholarships to train a generation of Vietnamese AI talents with the motto "zero to hero", from someone who knows nothing to becoming an expert.

Source: https://vtcnews.vn/ts-viet-tai-google-tao-ai-giai-toan-danh-guc-huy-chuong-vang-olympic-quoc-te-ar931477.html