Nation’s AI dialogue engine highlighted
A presentation yesterday on Taiwan’s self-built language model TAIDE, released commercially on April 15, showed the many fields it can be applied to, from language learning and agricultural knowledge searches to banking customer service.
The project to develop TAIDE, which stands for Trustworthy AI Dialogue Engine, was initiated by the National Science and Technology Council (NSTC) in April last year to create a foundational model for a traditional Chinese generative AI dialogue engine specifically for Taiwan.
The NSTC has collaborated with several institutions on the project over the past year, and a number of them appeared at yesterday’s presentation in Taipei to promote the system and demonstrate some of its applications.
For example, a team from the University of Tainan, led by computer science and information engineering professor Lee Changshing (李建興), developed a Taiwanese Hokkien-English AI chatbot for elementary and junior-high students based on TAIDE to learn the languages.
National Chung Hsing University has created an agricultural knowledge search engine called “Divine Farmer TAIDE” that can answer professional agricultural questions with citations, said Fan Yao-chung (范耀中), an associate professor of computer science and engineering at the university.
He said the answers generated through “Divine Farmer TAIDE” are more “context-based and detailed” than an engine the team previously developed based on ChatGPT because the new model includes reports from the Ministry of Agriculture in its database.
Taiwan Business Bank, in collaboration with an AI company, has applied TAIDE to help the bank’s employees access internal financial product information — which can be complicated and is continuously being updated — to help them provide better customer service.
The TAIDE model is “grounded in Taiwanese culture, incorporating unique elements such as Taiwanese
language, values, and customs, enabling generative AI to understand and respond to the needs of local users,” the council said.
A TAIDE model based on Meta’s Llama 2 (Large Language Model Meta AI) model (TAIDE-LX-7B) was released for commercial use on April 15, and another version for research only (TAIDE-LX-13B) has also been released.
NSTC minister Wu Tsung-tsong (吳政忠) said the TAIDE LX-7B has been downloaded more than 6,000 times in the half month since it was released, showing that there is demand for a traditional Chinese-based foundational model with a contextual understanding of Taiwan.
Announcing that the project would be extended for another year, Wu compared TAIDE to a car engine, adding that it would be up to different fields to use the model to “make their own cars.”
However, Lee Yuh-jye (李育杰), a research fellow at the Research Center for Information Technology Innovation and a TAIDE project convener, said that TAIDE is not aimed at competing with other major engines.
Llama 3 is trained by 24,000 Nvidia’s H100 GPU (graphics processing unit, which has become the backbone of artificial training training), while Mixtral, another large language model, has 1,500, and “we only have 72.”
“What we have to do is to play smart,” not big, Lee said.