Google wants to develop Vietnamese AI model

This is one of the cooperation proposals mentioned by a representative of Google Vietnam at the Conference on Science, Technology and Innovation Development on the afternoon of February 11 in Hanoi.

According to Google Vietnam General Director Marc Woo, the field of artificial intelligence in the country has made many strides in recent times. "However, to have outstanding development, a skilled intellectual force, labor force as well as businesses that are proficient in AI technology are extremely important," he said.

The first proposed collaboration is on AI models. According to Mr. Woo, data and models can determine the power of this technology. He said that over the years, the company has spent a lot of resources building large language models, and at the same time, he commented that these models "can only be fully effective when they are trained on data sources, especially diverse and high-quality Vietnamese".

"Therefore, we look forward to cooperating with the Government, organizations, and universities to jointly develop a large, modern language model imbued with the quintessence of Vietnamese," said Mr. Woo.

Regarding the construction of an open national dataset and allowing exploitation to build AI applications, Google considers this a positive point of Vietnam, ready to exchange and share lessons learned for development.

Regarding human resources, the General Director of Google Vietnam said that it will continue to provide scholarships and training programs to improve AI skills for those who qualify. In addition, for students and teachers, the company said it will provide learning and teaching support tools using artificial intelligence so that they can be exposed to AI early. "Google is ready to provide students and teachers in Vietnam with free access accounts to the Google WorkSpace and Classroom solution packages," he said.

Most popular AI chatbots support Vietnamese, but the processing capabilities of the large language models behind them vary widely. The report on the development of Vietnamese large language models (LLMs) through the VMLU standard set published in January shows that there will be at least 45 large Vietnamese language models built by the end of 2024 from domestic and foreign businesses and organizations.

Among them, Meta's Llama-3-70B, Zalo AI's KiLM-13b-v24.7.1, and OpenAI's GPT-4 are the leading models in general Vietnamese language ability, higher than Google's Gemma or Gemini.