台灣自己的 AI：雅婷逐字稿、TAIDE 與繁中模型的現況與機會

Every time AI is mentioned, someone always asks: "Does Taiwan have its own AI, or can it only use foreign ones?" This is a good question, and the answer is more optimistic than you think - Taiwan indeed has AI that is rooted in Traditional Chinese and local contexts, with practical applications in various fields.

The most down-to-earth application: Yating Transcription

Yating Transcription is probably one of the most widely used local AI tools in Taiwan. It focuses on converting Chinese speech to text and uniquely includes Taiwanese Hokkien, making it very useful for journalists, students, meeting records, and field interviews. When international tools often struggle with Taiwanese accents, Taiwanese Hokkien, and local proper nouns, Yating's "Taiwan-tuned" tool proves its value.

The national team's Traditional Chinese model: TAIDE

TAIDE is a large Traditional Chinese model project supported by the Taiwanese government and promoted by academic and research teams. Its goal is to create a foundational model that truly understands Taiwanese language, culture, and values. Its significance is not just "another chatbot," but to ensure that Taiwan has its own Traditional Chinese model capabilities in the AI era, without relying entirely on foreign companies.

Why does Taiwan need its own AI?

This question deserves a serious answer, with three reasons.

First, Traditional Chinese and Taiwanese contexts are the weaknesses of international models. Most international large models are based on English and Simplified Chinese data, often lacking in their grasp of Traditional Chinese terminology, Taiwanese regulations, history, and local knowledge. Only with its own models can Taiwan accurately capture these "most understanding of Taiwan" details.

Second, data autonomy and security. Sensitive fields like government, healthcare, and law cannot hand over their data to foreign cloud services. Having a deployable Traditional Chinese model locally ensures that AI can be used without surrendering data.

Third, industrial discourse power. Taiwan has strong semiconductor and hardware capabilities, and if it can also establish a foothold in models and applications, it can have more comprehensive discourse power in the AI supply chain, rather than just being a contract manufacturer.

Current situation and opportunities: a more pragmatic view

To be honest, Taiwan's local AI cannot compete with international giants in terms of scale and resources, which is the reality. However, this does not mean there are no opportunities - the experiences of localizing AI in Korea and Japan tell us that the winning strategy for small markets is not about being the "largest," but about being the "most understanding of local needs" and "best at integrating local demands."

For Taiwan, the most pragmatic approach is to leverage international open-source models for general capabilities and create its own depth in areas where international models are weak, such as Traditional Chinese, Taiwanese Hokkien, local knowledge, and specific industries. Yating's success in transcribing Taiwanese Hokkien and TAIDE's establishment of a Traditional Chinese foundational model are both solid steps in this direction.

Does Taiwan have its own AI? Yes, and it is growing. It may not be the largest in terms of parameters, but as long as it excels in "understanding Taiwan," it will have an irreplaceable position.