Vietnamese expert spearheads Google's AI advances, from world's best chatbot Meena to Bard and Gemini

Luong Minh Thang, 38, originally from Dong Nai Province in southern Vietnam, currently serves as a senior expert at Google DeepMind, Google's artificial intelligence (AI) research division.

Over nearly 10 years with Google, Thang has significantly contributed to developing several notable AI chatbots, including Meena - recognized as the world's best chatbot in 2020 - and its successors, Bard and Gemini.

Google expert Luong Minh Thang in a potrait photo he provides.

Thang's career as an AI scientist marked a significant shift from his original aspiration to pursue mathematics.

As a math prodigy from the High School for Gifted Students under Vietnam National University in Ho Chi Minh City, he won multiple city-level and national awards. He dreamed of following legendary Vietnamese mathematicians like Le Ba Khanh Trinh, renowned internationally for achieving a perfect score at the 1979 International Mathematical Olympiad (IMO), and Tran Nam Dung, a silver medalist at the 1983 IMO.

However, his journey changed direction when he did not pass the selection round for Vietnam's IMO team. "I didn't make it past the selection round, and that's when I pivoted to information technology," Thang recalled. "I see that as the turning point that changed my life."

In 2006, Thang enrolled at the National University of Singapore (NUS). At 19, he became fascinated with AI, especially its potential in machine translation. During his second year, Thang joined a specialized program, studying machine learning and natural language processing, fields he continues to work in today.

After graduating in 2010, he remained in Singapore to conduct research, publishing four papers within a year. This achievement helped him gain admission to Stanford University's prestigious Computer Science PhD program. At Stanford, Thang pioneered applying neural network-based machine learning methods to develop self-learning software for machine translation.

Thang's career with Google started in 2014 as a research intern at Google Brain (predecessor to Google DeepMind), working on enhancing translation quality by using neural networks. These advancements significantly improved Google Translate, which now serves over 500 million daily users.

In 2018, Thang co-founded the Meena project at Google, developing an AI chatbot capable of engaging in diverse topics. Starting from a conceptual proposal, the team built Meena into a sophisticated chatbot with 2.6 billion parameters trained on 340 GB of text data. Although Meena was recognized as the best chatbot globally in 2020, Google did not publicly release it due to concerns over potential risks related to inaccuracies and controversial responses.

According to Thang, Google's hesitation became critical when OpenAI's ChatGPT was launched in late 2022. "Public excitement around ChatGPT and what it could do was a shock to Google," Thang said. He subsequently joined a rapid 100-day development initiative to create Bard, a chatbot based on Meena's architecture but redesigned for accuracy and helpfulness.

"This part had to be rebuilt almost entirely compared to Meena. Meena was designed to be witty and humorous, while Bard was created to give accurate and useful information," Thang said.

During Bard's development, all Google employees participated in interacting with the chatbot to refine its responses. Thang described this period as highly stressful but incredibly rewarding: "It was one of the most memorable periods—cramming one year's worth of work into 100 days." Bard was officially launched in February 2023, marking the beginning of ongoing refinements.

Luong Minh Thang (5th, L) and the 2024 Vietnam IMO team in the U.K. Photo courtesy of Thang

Outside his Google responsibilities, Thang engages in independent research projects. In 2022, he collaborated with Trinh Hoang Trieu, a PhD student at New York University, on AlphaGeometry, an AI designed to solve complex geometry problems. The project aimed to solve the famously difficult geometry problem from the 1979 IMO, originally solved by Le Ba Khanh Trinh.

In early 2023, AlphaGeometry successfully solved 25 out of 30 previous IMO geometry problems. After additional research and development, the upgraded AlphaGeometry 2, integrated with Gemini (a chatbot evolved from Bard), solved the challenging 1979 IMO problem. The team also developed AlphaProof, capable of solving algebra and geometry problems efficiently, achieving a perfect score on four problems from the 2024 IMO.

"I hope that by 2026, we'll see an AI win a Fields Medal like Professor Ngo Bao Chau," Thang said, referring to the Vietnamese-French mathematician who earned the prestigious mathematics award in 2010.

Currently, Thang leads Google's artificial general intelligence project, focusing on developing AI that can reason coherently and make interconnected judgments, akin to human cognition. He believes AI will eventually surpass human intelligence but views this positively.

"AI is a new source of energy and a tool to help us accelerate scientific and cosmic discoveries," Thang emphasized. "If you want to pursue AI, you need a healthy curiosity, and no fear."