Google AI pioneer Le Viet Quoc says Vietnam must innovate, not imitate

Quoc, 43, holds a PhD in computer science from the U.S.’s Stanford University and was recognized by MIT Technology Review as one of the world's top innovators under 35. His contributions to AI at Google Brain, now part of Google DeepMind, have earned him global recognition.

During his visit to Vietnam for the 2025 International Conference on AI and Semiconductors from March 12 to 16, he shared insights on AI development, challenges and opportunities for Vietnam.

- As an AI researcher, how do you assess the competition between the U.S. and China?

This has been a widely discussed topic, especially after DeepSeek’s emergence, which highlights China’s rapid AI progress. Before DeepSeek, Google had the Gemini Thinking model, and OpenAI had its o1 model.

Earlier some had estimated China to be two years behind the U.S. in chip resources and talent. However, with the launch of DeepSeek’s I estimate the gap has shrunk to about six months.

There are indications that DeepSeek distilled models from OpenAI or Google, but that is difficult to confirm since no details have been disclosed. If true, the gap may still be substantial.

- DeepSeek gained attention for developing an AI model with a US$5 million investment. What can Vietnam learn from this?

While the $5 million figure applies to one model, the total investment likely exceeds $1 billion. I know some Vietnamese companies have also committed similar amounts to AI development.

Vietnam should focus on open-source AI. Most organizations and companies should use open-source AI, as it enables faster development and makes use of existing resources. For example, when developing AI to address challenges in healthcare, academia or education, models like DeepSeek, Llama and Gemma can be leveraged.

However, Vietnam should not only adopt but also contribute to open-source AI, which would help build global recognition and attract top talent. The idea of contributing to open-source AI has not been widely discussed in Vietnam, but I believe it is essential. Instead of merely using open-source models, Vietnam should actively develop them.

Universities and companies with AI models like PhoGPT, or innovative datasets, should consider making them open-source to benefit the global AI community.

Google AI pioneer Le Viet Quoc. Photo by VnExpress/Luu Quy

- Vietnam is far behind other countries in AI. What advice do you have for local AI projects?

Being a latecomer has advantages. Just like in a marathon, trailing runners avoid headwinds.

In AI, late entrants can learn from the mistakes of early adopters. Gemini and ChatGPT have large user bases, but maintaining these services requires vast computing resources, which diverts focus from improving models.

Newer companies do not face this challenge and can concentrate on building better AI models. Additionally, AI development costs are decreasing. Early models required hundreds of millions of U.S. dollars, but newer iterations are significantly more affordable.

As this trend continues, Vietnam will have a realistic opportunity to develop its own AI models. Vietnam should not rush to catch up. If we aim to match DeepSeek today, by the time we do they will have already moved ahead. Success is not just about building AI models but ensuring they are widely adopted.

Even DeepSeek is still trying to catch up with major U.S. firms due to resource limitations. To succeed, Vietnam must prioritize innovation. DeepSeek made an impact because of its unique approach, not by simply replicating others.

If Vietnam wants to develop foundational AI models like those powering today’s chatbots, it must focus on areas where it has strengths and room for innovation. Every AI model has weaknesses. If Vietnam can identify and improve upon these, it could surpass competitors.

At AISC 2025, AlphaChip, a project using AI for semiconductor design, was introduced. This is an exciting field, as it combines AI with chip design, representing a promising new direction.

- AI development relies on young talent. How can Vietnam cultivate AI experts?

Talent is fundamental to any field—whether building pyramids, space exploration, or AI research. Vietnam has a solid STEM education foundation, but more investment is needed.

Higher education and postgraduate programs must improve. Research institutions should provide an environment where top talent can thrive within Vietnam. Currently Vietnam’s university education lags behind world-class institutions.

Many talented individuals study abroad and settle there, making it difficult to attract them back. Those who remain in Vietnam often lack access to top-tier education and research.

Bridging this gap requires investment. One approach is to invite globally recognized professors to Vietnam or equip local universities for groundbreaking research.

However, education alone is insufficient. Vietnam must also nurture a startup ecosystem and attract major tech firms like Google and Facebook to open R&D offices. This would give Vietnamese talent access to global expertise and opportunities. This is a long-term journey, but one worth pursuing.

In the region, Vietnam’s closest competitor in this field is India. If Vietnam leverages its strengths and invests in talent, it has a strong chance of becoming a major AI player.

- Vietnam has recently introduced policies to support science, innovation and digital transformation. What are your thoughts on this?

Attracting AI talent depends on individual motivation. Personally, aside from family, I am passionate about research and cutting-edge science. Google DeepMind provides the necessary resources for top-tier research—something Vietnam currently lacks due to resource constraints.

Vietnam’s leadership is committed to AI development, which is encouraging, and I am personally impressed by their dedication. However, achieving major progress takes time.

Developing AI talent through education and training will require at least four to five years before tangible results appear. Countries like China, Singapore and India have heavily invested in education, science and technology, giving them an advantage. However, Vietnam’s current momentum suggests it has the potential to make significant AI advancements.

- AI is advancing rapidly, making older models obsolete. What are your predictions for AI’s future?

AI agents will be the next major breakthrough. Between 2022 and 2025, chatbots dominated AI development. Over the next two to three years, AI agents will take center stage.

The biggest challenge in developing AI agents is data availability. Early chatbots were trained using large internet datasets to predict the next word in a sentence. More recently, reasoning-based AI models have emerged, relying on vast amounts of data to verify accuracy.

However, AI agents require real-world interaction data, which is scarce. These systems will need to conduct experiments and interact with the physical world to generate meaningful responses—this is a major challenge.

Unlike chatbots, AI agents will actively assist users by researching information online and engaging with individuals. This shift will lead to more practical applications. Many companies are already investing in AI agents, and I believe we will see major breakthroughs this year.