Taiwan has begun constructing its own Large Language Model (LLM) to flee China's influence. It’s because Chinese chatbots reminiscent of Baidu's 'Erniebot' emphasize 'One China'.

Bloomberg reported on the twenty sixth (local time) that Taiwan will set a budget of NT$17.4 billion (about KRW 774 billion) by 2026 to develop artificial intelligence (AI) tools and train human resources.

In line with this, what’s currently imminent to be released is the language model 'TAIDE', which might provide a platform for writing emails or meeting summaries to firms, banks, hospitals, advertisements, etc. $7.4 million (about 9.9 billion won) has already been invested here.

“Taiwan needs an LLM that’s consistent with Taiwan’s values,” said Li Yujie, a professor of applied mathematics at National Yangming Jiaotong University and coordinator of the Tide project. “The LLM goes beyond knowledge and includes core values ​​reminiscent of freedom, democracy, and human rights.” He said.

It is understood that Baidu's Erniebot has recently gained popularity amongst young people in Taiwan, which has had a big effect. Erniebot gives accurate answers about recent knowledge reminiscent of the outcomes of Taiwan's presidential election, but adds at the tip of the reply, “Irrespective of how the situation in Taiwan changes, the essential fact is that there is barely one China.” Taiwan is understood to be concerned that such phrases could cause erosion of its cultural and political system.

The federal government is providing training data to assist Tide's development and is pursuing licensing agreements from Taiwanese media. Tide relies on Meta's open source model 'Rama 2'. Through this, we plan to support printing in Traditional Chinese, which is utilized by most of Taiwan's population. The initial version of Tide will begin testing with select partners in April.

Tide has far fewer features than cutting-edge models reminiscent of ChatGPT, and the investment scale will not be large. Nonetheless, moderately than constructing an LLM that may compete with state-of-the-art models, the main focus is on developing knowledgeable model that fine-tunes the Tide Foundation model. Specifically, Taiwan is understood to be sensitive to information leaks, and lots of individuals are attempting to construct their very own models and use them in an on-premise manner.

“To take a whole industry to a recent level, having a usable model is enough,” said Zhang Jixing, a pc science professor at National Taiwan University. He said advanced models like ChatGPT aren’t needed without delay. He also revealed plans for on-device AI, saying, “We will make Tide sufficiently small to suit right into a laptop or mobile phone.”

Taiwan is way behind in AI in comparison with hardware fields like TSMC. Evaluation suggests that this investment means changes are happening in Taiwan as well.

Reporter Park Chan cpark@aitimes.com