DeepSeek R1 is rising, but Taiwan's AI is still stagnant? Legislator Ke Jinyi questioned the Executive Yuan's 'ban' policy.

Chinese AI company DeepSeek recently launched the DeepSeek R1 large language model (LLM), which has shocked the industry with its open weights, MIT authorization mode, and strong performance, causing high attention from global tech giants. However, AI development in Taiwan is still stuck, with limited resources, vague policies, and even the Executive Yuan's "comprehensive ban" policy, leading to questioning. Kuomintang legislator Ke Junchun recently raised four major issues regarding the current development of sovereign AI, criticizing the government for long-term "empty talk about sovereign AI" but lacking concrete actions, causing Taiwan's AI technology to stagnate.

DeepSeek R1 debuts strongly, is Taiwan's AI resource sufficient?

DeepSeek R1, which recently came out, has 671 billion parameters (671B), which is 50 times the size of Taiwan's self-developed TAIDE 13B model and is comparable to OpenAI's o1 model in terms of performance. In addition, DeepSeek R1 adopts the MIT License, allowing developers to optimize and apply it freely. However, Ge Rujun is concerned that the Taiwan AI field has long faced three major challenges: 'insufficient traditional Chinese language corpus, lack of research and development talent, and shortage of computing resources', which leads to a lack of sufficient conditions and international competitiveness even with the technology.

The Executive Yuan calls for AI sovereignty, but prioritizes 'banning' DeepSeek?

Regarding the development of AI, the Executive Yuan recently held a press conference to express its follow on the sovereignty of AI and computing resources, and emphasized that the National Science Council will fully promote relevant research and development. However, the government subsequently changed its stance and announced a comprehensive ban on the use of DeepSeek R1 by government agencies. If public universities and research institutions want to use it, they must obtain approval from the Executive Yuan and take necessary control measures.

Ke Jun Ge questioned that such a decision is no different from 'building a car behind closed doors'. While world-class companies are actively investing in DeepSeek R1, Taiwan chooses to block technology instead of thinking about how to make up for its shortcomings and accelerate the development of independent AI.

Four key issues: Taiwan's AI development is still stuck.

Without the "Taiwanese Element Database", there would be no sovereign AI.

Ge Rujun emphasized that the core of sovereign AI lies in training data, but the current Taiwan authorities have not established enough local corpus in Taiwan. Four months ago, he suggested that government departments open up materials to TAIDE as training texts, but progress has been slow.

Taking the digitization of 180,000 cultural relics in the National Palace Museum as an example, the current progress is only 35%, and both the National Security Council and the National Palace Museum have not yet joined the TAIDE database. In addition, the government's draft of the "Promotion of Data Utilization Act" has not yet been passed, resulting in the inability of public sector data to be effectively opened, leaving Taiwan's AI lacking local data support.

With a vague budget, how to support AI development?

Ge Rujun pointed out that AI development requires clear financial planning, but the current government's AI budget is not only 'opaque', but also simply passed over with 'business expenses', making it impossible to confirm whether it is really invested in AI construction. He once questioned the Chairman of the National Science Council, who not only admitted that he had not seen the complete budget, but also said that 'this plan by the Executive Yuan is for reference only', which raises doubts about the government's determination to develop sovereign AI.

How long will it take to relax the copyright law?

In terms of AI training data acquisition, Japan has amended the law to allow AI to use copyrighted content within a reasonable range, while Taiwan has not made relevant legal progress. So far, the Executive Yuan has not clearly stated whether to amend the Copyright Law, only that it "will consider it in the future." However, AI development is highly competitive, can Taiwan afford to wait?

International competition for AI talent, Taiwan can't even coordinate across departments

AI competition is not only a technical competition, but also a talent competition. Countries have successively introduced 'AI green cards' to attract top talents from around the world, but Taiwan cannot even achieve basic cross-departmental integration.

Currently, Taiwan's AI budget is allocated by the National Science Council, but the businesses involved are governed separately by the Ministry of Education, the Ministry of Digital Development, and the National Science Council, lacking a unified leading organization. When questioned by Ke Ru-jun, the officials responded, 'The responsibility is shared by each department.' He questioned whether Taiwan can truly develop independent AI with such an evasive attitude.

The warning four months ago has now come true.

Ge Rujun emphasized that during the general interpellation four months ago, he warned the government that if it did not promptly propose a 'sovereign AI strategy', Taiwan's AI would only stagnate and ultimately become a 'castle in the air'. Today, global AI technology is advancing rapidly, and DeepSeek R1 has already changed the industry ecosystem, while Taiwan's AI development is still mired in political and administrative inefficiency.

He sighed at last: 'How many more "four months" can Taiwan waste now?' If the government does not take action, Taiwan's AI competitiveness may be completely left behind by the world.

Is the government's budget and data openness really as Ge Ru Jun said?

In response to Ge Rujun's criticism, some opinions believe that his accusations about AI training data and budget issues are debatable. First of all, he questioned whether the Palace Museum cultural relics and nuclear safety committee materials were not included in the TAIDE training, but the digitization of the Palace Museum cultural relics is mainly a cultural preservation project, and is it directly related to the development of LLM? Large language models require a large number of language and knowledge materials, can classical Chinese training really bring AI technology breakthroughs? As for the NSC, it is also worth carefully considering whether information involving state secrets is suitable for inclusion in AI training. On the other hand, Ge Rujun criticized the Taiwan authorities for not explicitly listing the budget for AI training data, and only brought it in general terms of "business expenses", but some people also pointed out that the KMT had drastically cut the AI budget in the Legislative Yuan in the past, but now it has questioned whether the use of government funds is reasonable, and its position seems to be contradictory. When the AI R&D budget is cut, should it be better to follow how to supplement resources rather than simply criticize the way the Executive Yuan budgets are presented?

This article DeepSeek R1 rises, Taiwan AI still stands still? Legislator Ge Rujun questioned the Executive Yuan's "banning" policy, which first appeared in Chain News ABMedia.