Taipei, May 3 (CNA) Friday's presentation on Taiwan's indigenously built language model TAIDE, which was released commercially on April 15, showed how it can be used in many applications, from language learning and agricultural knowledge search to banking customer service. It was shown that it can be applied to various fields.
The TAIDE development project, which stands for Trustworthy AI Dialogue Engine, was approved by the Cabinet-level National Science and Technology Council (NSTC) in April 2023 to create a basic model of a traditional Chinese generative AI dialogue engine specifically for Taiwan. It was launched in May.
NSTC has been working with multiple institutions on this project over the past year, many of whom appeared at Friday's presentation in Taipei to promote the system and demonstrate some of its applications.
For example, a team at Tainan University led by computer science and information engineering professor Li Jianxing developed an AI chatbot for elementary and middle school students to learn Taiwanese and English based on TAIDE. .
Meanwhile, National Chung Hsing University has developed an agricultural knowledge search engine called “Divine Farmer TAIDE” that can answer specialized agricultural questions with citations, said Huang Yaochun, an associate professor of computer science and engineering at the university. (Fan Yaozhong) said.
He said that because the new model's database includes reports from Taiwan's Ministry of Agriculture, the answers generated through Divine Farmer TAIDE are more “contextual” than the engine the team previously developed based on ChatGPT. Based on details.
Taiwan Commercial Bank collaborates with an AI company to apply TAIDE to help bank employees access internal financial product information (which can be complex and continuously updated) and provide better customer service. We helped provide the following.
The TAIDE model is said to be “rooted in Taiwanese culture and incorporates unique elements such as Taiwanese language, values, and customs, allowing generative AI to understand and respond to local user needs.” .
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 15th, and another version for research only (TAIDE-LX-13B) was also released.
Wu Tsung-tsong, head of NSTC, said that more than 6,000 downloads have been made in the six months since the release of TAIDE LX-7B, and there is a strong demand for a traditional Chinese-based basic model that understands the Taiwanese context. He said it shows that there is.
Announcing that the project will be extended for another year, Wu likened TAIDE to a car engine and said it was up to various sectors to use the model to “build their own cars.”
However, Lee Yuh-jye, a researcher at the Academia Sinica Information Technology Innovation Research Center and the convener of the TAIDE project, said that TAIDE is not intended to compete with other major engines. Stated.
Llama 3 is trained on 24,000 Nvidia's H100 GPUs (the graphics processing units that are the backbone of AI training), while Mixtral, another large language model, has 1,500 There are only 72 of them.”
“What we have to do is play smart and not play big,” Lee said.
