Building Taiwan’s Next AI Frontier—A Pragmatic Approach to DeepSeek R1 and Beyond

If there’s one thing I learned from growing up in Taiwan and later founding iCHEF—a restaurant technology business—it’s that our ability to rise to global challenges often hinges on getting our hands dirty and actually building rather than sitting around debating the “what ifs.” We did it with semiconductors; now it’s AI’s turn.

In this post, I want to share why I support the Project: FreedomGunpla R1 (also known as the plan to adapt and retrain DeepSeek R1 for Taiwan), how we can balance idealism with a much-needed dose of economic pragmatism, and why now is a critical window for Taiwan to put its stake in the ground for AI leadership.

Why Support This Project?

1. We’ve Done It Before: From Semiconductors to Software

Taiwan’s mastery in semiconductors, championed by TSMC and others, didn’t happen overnight. It was a combination of vision, capital, and relentless experimentation. Fast-forward to today, and Taiwanese semiconductor technology powers smartphones and data centers worldwide.

Now, the next wave is AI—and the question is, will Taiwan stand by as a consumer, or lean in as an innovator? We have the technical chops and the community to do it. Yet we often get stuck in endless debate or wait for “someone else” to take the lead. If we learned anything from semiconductors, it’s that early, aggressive investment in both technology and talent can generate massive leverage.

2. Moving Beyond the Status Quo

My friend, Liang-Bin Hsueh (aka Pudding), and his team have launched FreedomGunpla R1 as an open-weight adaptation of DeepSeek R1—one of the most advanced and fully open-weight AI models (MIT License) on the planet right now. It has performance levels that rival OpenAI’s o1 series in math, code, and certain types of reasoning.

DeepSeek R1 might have some “China-centered biases” baked into its training, but as far as raw performance goes, it’s impressive. The plan is to fine-tune and retrain it for our local contexts, particularly in Traditional Chinese and English. In other words, to reorient the model’s perspective so it aligns with Taiwanese and broader global values—without sacrificing the advanced reasoning engine.

3. Economic Pragmatism Over Unending Ideological Battles

Sure, there are concerns: “Is this model too ‘China-focused’? Can we really align it to Taiwan’s values?” Yes, the pretraining data might lean a certain way. Yet from an economic and innovation standpoint, the real question is: Do we use the momentum of this open-weight model to leapfrog, or do we wait for the perfect solution while losing valuable time?

Some critics say it’s impossible to “completely rebuild the model’s consciousness.” They argue that any fine-tuning is superficial if the core pretraining data has biases. My response is simple: OK, we wait. Then the timing passes and nobody moves—just like we’ve seen too many times before. That leaves Taiwan out of the global AI race. We need to be pragmatic here, focusing on:

• Real-world economic benefits
• Rapid market deployment
• Tech innovation at scale
• Empirical results

This Economic Pragmatism is about making sure Taiwan doesn’t get left behind while we debate intangible issues forever. We can always refine and ethically align the model further, but the best way to make an impact is to build something tangible and deploy it now.

Addressing the Critics: “We Need Some ‘F*cking Business’”

I don’t mean to be crass, but at some point, we need to ship. Prolonged debate with no tangible output leads to zero impact, zero market share, and zero leadership. Yes, it’s important to consider ethical and cultural implications, especially regarding any potential “China bias” in the model. But you can’t adjust or improve a model you never deploy in the first place.

Our job is to harness an open-weight solution like DeepSeek R1, put it into the real world, and see what works—and what doesn’t. We’ll refine as we go, guided by local knowledge and global best practices. In simpler terms: We need some “f*cking business” to move the needle.

Technical and Cultural Vision

The FreedomGunpla R1 plan aims to:

1. Retrain and Optimize
   • Use the foundational DeepSeek R1 models to create a variant that performs better in Traditional Chinese and English than in Simplified Chinese.
   • Ensure the model is “open weight,” meaning anyone can run and adapt it. This fosters a robust ecosystem of tools, plug-ins, and specialized AI services rooted right here in Taiwan.
2. Cultural Reframing
   • Adjust the model’s existing biases or “value alignment” to respect a broader set of cultural, democratic, and humanistic values—instead of what might be baked into the original training data.
3. Boost AI Ecosystem
   • Encourage local companies to experiment, fine-tune, and deploy. We can’t rely solely on global providers. Having a homegrown (or at least home-tailored) large language model means more control over data, censorship, and language nuances.
4. Capitalize on Semiconductor Strength
   • Taiwan’s GPU clusters and chip know-how can reduce the cost and increase the speed of training and inference. This synergy can translate into a powerful competitive edge in the AI marketplace.

The Business Rationale

• Global AI Spending is soaring, with multinational corporations racing to adopt large language models for everything from coding assistants to medical research.
• Domestic AI Talent: We have engineers who already know how to train big models. Some have contributed to previous large-scale projects like Taiwan LLM, TAIDE, and Breeze.
• Semiconductor Advantages: Our position in the global supply chain means we can source advanced GPUs and custom chips more reliably.
• Immediate Market Impact: Early movers will define best practices and capture initial market segments in local language AI services—from enterprise-level solutions to creative and cultural industries.

If Taiwan acts now, we can build on the open-weight foundation that DeepSeek has (flaws and all) to launch something uniquely ours—a model that not only speaks Traditional Chinese but also thinks in line with our needs and global values.

Call to Action

We need community support and a shared vision to move this proposal forward. Here’s how you can help:

1. If You Agree:
   • Like this post, share it with your network, and offer constructive comments on how to refine and expand the plan.
2. If You Disagree:
   • Feel free to share your perspective to broaden the conversation, but please refrain from leaving purely negative remarks; we’ll simply block those. We want productive dialogue, not mudslinging.
3. Funding & Collaboration:
   • We’re seeking both private and public investment to acquire legally licensed data and scale our existing GPU capacity.
   • If your company wants to be an early adopter or partner for fine-tuning the model, now is the best time.
4. Spread the Word:
   • Taiwan’s AI community needs more visibility. If you can amplify this message, do it. The more voices, the more momentum we’ll have.

Don’t let the moment slip away. We’ve been through this cycle of waiting too many times. Let’s focus on real-world outcomes and build something that advances Taiwan’s standing in the global AI arena. That’s how we make sure history remembers us—not just as the island of semiconductors, but also as a powerhouse for next-generation AI.

Together, we can transform open-weight models from a theoretical opportunity into a practical driver of economic growth, cultural identity, and global tech influence.

Written by:
Ming-Cheng Ho
Founder of iCHEF, passionate builder in Taiwan’s tech scene, and firm believer in forging Taiwan’s next AI frontier through economic pragmatism and relentless innovation.

Let’s get to work.