Artificial Intelligence & Machine Learning Articles
Vietnamese Expert At Google Says Vietnam Should Prioritize AI, Chip Design
Quoc, a Vietnamese AI scientist, is renowned for his pioneering role in the field of deep learning at Google. His groundbreaking research has contributed to integrating AI into practical applications such as natural language processing and computer vision.
At the 2025 International Conference on AI and Semiconductors (AISC), Quoc shared his perspective on investment strategies for the development of the semiconductor and AI sectors.
What do you think about the view that now is a historic opportunity for Vietnam to develop semiconductor and AI technology?
I agree that this is a significant opportunity. The timing is ripe—neither too early nor too late. For example, it's now too late to start making phones because other countries have been doing it for years. Meanwhile, technologies like AI and chips are still new and full of potential.
Vietnam's advantage lies in its strong human resources. Anything worthwhile requires talented people. Building pyramids, going to the Moon, and now creating AI—all of these rely on people. Vietnam has a robust workforce, and STEM is a particular strength of Vietnamese people.
In Asia, the countries competing with Vietnam in terms of human resources are Singapore, India, and China. Singapore doesn't have as many human resources as Vietnam. China faces its own challenges as it directly confronts the US. Vietnam's biggest competitor is India. If we seize this opportunity to leverage our human potential, we can join the playing field of AI and semiconductors.
Vietnam needs to reinforce its existing strong foundation by improving the quality of higher education and postgraduate studies, as well as developing research institutes to create a "playground" for domestic talents.
Given the current challenges, what policies should Vietnam adopt to participate in this "game"?
In my opinion, the semiconductor industry requires vast resources—potentially billions of dollars—if we pursue chip manufacturing. If pursuing the strategy on manufacturing low-cost chips, Vietnam will have to compete with many countries.
If Vietnam has to choose between AI and chips, we should invest more in AI because it demands fewer resources. Moreover, AI still has a long road ahead, giving Vietnam a chance to catch up and surpass others.
For semiconductors, Vietnam should focus on chip design rather than chip manufacturing. Chip design offers higher profits and is better suited to Vietnam's available resources and workforce.
Where should we focus if we want to accelerate the development of AI?
My advice, based on years of experience, is not to hurry. If we build a company mimicking DeepSeek, by the time we catch up, they'll already be far ahead on a different path.
Vietnam should focus on new directions that emphasize innovation. For instance, using AI to design chips is a new subfield of AI that ties into semiconductors.
If we pursue chatbots, we should target the fields where there are opportunities for innovation. Specifically, foundational models have many weaknesses. If we identify these weaknesses and focus on new ideas to address them, Vietnamese companies could outpace others.
Do you think Vietnam has the advantage of being a latecomer, standing on the shoulders of giants?
Being a pioneer has advantages, and being a latecomer does too. For example, many marathon runners stay behind to avoid the wind. Latecomers can also learn from the mistakes of those who went first.
ChatGPT of OpenAI and Gemini of Google have massive users. The problem these companies face is that with so many users, most of their resources have to be reserved for serving users rather than improving their AI models.
Meanwhile, latecomers don't have to expend resources on user support and can focus solely on creating better models. That's the advantage of coming later—avoiding the headwinds.
Can Vietnam create products with low investment but high efficiency, like DeepSeek?
I believe it's feasible. Globally renowned models in computer vision, speech processing, and natural language processing (NLP) have seen their costs drop rapidly. The first model might cost hundreds of millions of dollars, but the next one costs $20 million, and then $5 million, and it keeps getting cheaper. Therefore, it's entirely feasible for Vietnam to develop such models.
Trong Dat
Comments
Post a Comment