Associate Professor Luu Anh Tuan's dream of 'AI sovereignty' and his determination to build a large-scale Vietnamese language model.

DNVN - Returning to Vietnam after 17 years working in Singapore, Associate Professor Luu Anh Tuan - Executive Director of the Artificial Intelligence (AI) Research Center at VinUni University - brings with him the aspiration to realize a "Make in Vietnam" large language model (LLM) to maintain "AI sovereignty".

Associate Professor Luu Anh Tuan - Executive Director of the Artificial Intelligence (AI) Research Center at VinUni University.

When foreign AI "gives up" in the face of the complexity of the Vietnamese language.

In the context of Vietnam's strong shift from an "AI-using" nation to an "AI-creating" nation, Associate Professor Luu Anh Tuan argues that relying entirely on foreign models like ChatGPT carries significant risks. Although these tools support the Vietnamese language, their quality still falls short of expectations due to a lack of in-depth understanding of local culture and context.

The biggest challenge that Associate Professor Tuan pointed out lies in the complexity of the Vietnamese language.

"The Vietnamese language has many unique features compared to many languages ​​in the world . Even the way we address each other, like 'aunt, uncle, grandparent,' is very complex, so common AI systems around the world cannot understand it," said Associate Professor Luu Anh Tuan.

In addition, the dialectal diversity between the three regions of North, Central, and South Vietnam, along with specific historical factors, are also "blind spots" where international models—which are trained on general data—often encounter problems when processing information.

"If we keep asking ChatGPT and then paying, we can only use AI, not develop artificial intelligence," Associate Professor Luu Anh Tuan emphasized.

This raises urgent issues regarding digital sovereignty and AI sovereignty, forcing Vietnam to achieve technological self-reliance to avoid dependence on foreign countries.

To build its own large-scale language model, Vietnam's biggest challenge today is data. Associate Professor Tuan frankly acknowledges: "Vietnam's data is both scarce and scattered across different locations."

Currently, data collection relies primarily on research groups independently gathering data from the internet and manually filtering it; there are no specialized companies for large-scale data processing. Furthermore, Vietnam lacks standardized evaluation criteria to measure the accuracy and reliability of language models within the domestic cultural context.

In light of this situation, Associate Professor Tuan suggested that the government should play the role of "conductor," establishing national data centers to consolidate data from various ministries and sectors (health, law, education , etc.), helping to avoid wasting resources and ensuring consistency. He also proposed the need for a unified agency to verify lists of "clean" data, creating a foundation for future developers to use them safely.

The "open source" strategy and the desire to contribute.

Despite the immense challenges, the belief and determination of the Associate Professor from Nanyang Technological University (NTU Singapore) are incredibly strong. After 17 years of studying and working abroad, Associate Professor Tuan decided to accept Professor Duong Nguyen Vu's invitation to return and become the Executive Director of the AI ​​Research Center at VinUniversity.

Sharing his reasons for returning, he said: "Actually, Vietnamese people living abroad like me always want to return to serve our country." He sees this as an opportunity to connect international knowledge sources with the training of talent in Vietnam. Currently, with the flexible cooperation mechanism between NTU Singapore and VinUniversity, he is working to build a bridge to bring Vietnamese students abroad for further education and vice versa.

The strategy that Associate Professor Tuan is pursuing is not to create a closed model, but to make it open source.

"Vietnam should focus on open-source models so that small companies and communities can inherit and develop them, creating a strong AI ecosystem," Mr. Tuan commented.

He also placed emphasis on building "trusted AI" based on five pillars: Safety, Responsibility, Privacy, Fairness, and Transparency.

"Vietnam's advantage is that it's a latecomer, not burdened by outdated data systems containing a lot of fake news, so it can establish an ethical foundation right from the start," said Associate Professor Luu Anh Tuan.

According to Associate Professor Luu Anh Tuan, the journey to build a large-scale "Make in Vietnam" language model is not simply a technological race, but also a story about cultural and intellectual sovereignty. Instead of continuing to depend on expensive and culturally insensitive "black boxes" from foreign countries, Vietnam is facing a golden opportunity to build a reliable, transparent, and open-source AI ecosystem. With the advantage of being a "latecomer," allowing for the establishment of ethical standards from the outset, and with the cooperation of the Government, Associate Professor Tuan believes that the aspiration to transform Vietnam from a nation that "uses AI" to a "creator of AI" and a leader in the region by 2030 is entirely feasible.

Hien Thao

Source: https://doanhnghiepvn.vn/cong-nghe/giac-mo-chu-quyen-ai-va-quyet-tam-xay-dung-mo-hinh-ngon-ngu-lon-tieng-viet-cua-pgs-luu-anh-tuan/20260216100302351