TAIPEI (Taiwan News) — Taiwan is accelerating development of its own generative AI model, aiming to preserve local language and culture while reducing reliance on foreign AI systems, Nikkei Asia reported Thursday.
The effort is being led by the homegrown Trustworthy AI Dialogue Engine (TAIDE), a large language model developed under the direction of the National Science and Technology Council.
Many Chinese-language datasets used to train major AI models, including ChatGPT, rely heavily on simplified characters content from China, while traditional character materials used in Taiwan are less widely available, per Nikkei Asia. Researchers say this can cause AI systems to produce responses that reflect Chinese terminology, political views, or censored information on historical issues.
Academia Sinica Institute of Information Science associate research fellow Huang Han-hsuan (黃瀚萱) said failing to develop local AI capabilities could mean handing control of society’s “brain” to outside parties.
Huang described TAIDE as a digital “bulwark” that can protect Taiwan’s linguistic, historical and social autonomy. He said the model aims to ensure that future AI systems better understand Taiwan’s language and cultural context.
TAIDE is built using Taiwan-based data, including government documents and ethnic minority languages, combined with foundation models such as Google’s open-source Gemma. Taiwan is using overseas technology as a foundation while adding local data to create a model tailored to its own needs, an NSTC official explained.
The first version of TAIDE was released in 2024, with major performance upgrades planned this year through the use of advanced AI chips from Nvidia. The model is being used on local servers by government agencies, hospitals, and companies.
