Blog
Transforming AI from "Memorizing Questions" to "Solving Problems": Insights from Anti-Money Laundering (AML) Practice on Fine-Tuning
Through an Anti-Money Laundering (AML) case study, this article demonstrates how Taiwan AI Cloud has evolved AI from rote memorization of answers into experts with logical reasoning abilities. The article documents the process from LoRA validation and model distillation to optimization using the CoT (Copy of Thought) and GRPO relative reward modeling, emphasizing that a good model is not formed through a single training iteration, but rather relies on a rigorous pipeline for continuous "management" and evolution. This engineering philosophy proves that only by establishing visible and automated processes can AI be transformed into stable and reliable digital assets in demanding regulatory environments.
From understanding language to thinking: Deconstructing the three major milestones in the evolution of AI models
The evolution of AI can be broken down into three milestones: LLM, which excels at pure text translation; MLLM, which possesses visual perception capabilities; and Reasoning Model, which can deconstruct complex logic and self-correct. The latest technology is not necessarily the best solution. Enterprises should precisely allocate their resources according to the "nature of the task," moving from "understanding language" to "being able to think," in order to truly enable AI transformation to take root and flourish.
Transforming LLM from "Broad Knowledge" to "Specialized Expertise": A Practical Guide to AI Transformation on the Enterprise Map
Taiwan AI Cloud transforms general-purpose LLM models into "sovereign AI" with professional productivity through a rigorous "fine-tuning pipeline." The article likens fine-tuning to specialist training: first, CPs supplement domain knowledge; then, SFTs teach task behaviors; and finally, CoT thinking chains ensure transparent and auditable decision-making. This engineering philosophy emphasizes that "good models are managed," and only by mastering standardized processes can enterprises achieve a balance between technology, security, and efficiency, building truly robust and trustworthy digital assets.
From "Sovereign AI" to "AI City": Building a New Paradigm for Digital Governance by 2026
2026 marks the beginning of Taiwan AI Cloud's "Computing Power Connection Governance" era. As enterprises shift from model training to inference applications and agentic AI, Taiwan AI Cloud is building a digital ecosystem through a "3+1 strategy," translating sovereign AI into a practical force for solving urban pain points. Building on its return to profitability in 2025, Taiwan AI Cloud has initiated its IPO and plans to list on the Emerging Stock Market in May. Its goal is to achieve over 50% overseas revenue by 2028, exporting Taiwan's AI governance experience to Vietnam, Japan, and the Middle East markets.
Peter Wu : Why AI and Trust Must Evolve Together
Artificial intelligence (AI) is reshaping the world order at an astonishing pace. From medical diagnostics and urban transportation to fintech and industrial digital transformation, AI is gradually becoming the "productivity engine" of the new generation. However, as people pursue efficiency and intelligence, a more critical question is becoming increasingly prominent—when we entrust decision-making to AI, can we still "trust" it?
Peter Wu: AI Basic Law to Serve as the Trust Foundation for Smart Healthcare
The AI Fundamentals Law does not overlook innovation itself. It distinguishes between the research and development and application stages, and designs innovative experimental environments to allow technology to be verified and evolved under controlled conditions.
Classification
- Storage (8)
- Digital resilience (2)
- Trend View (7)
- Compute (12)