16 financial holding companies collaborate to build an AI large language model that understands Taiwan's finance

📄Full Article· Automatically extracted by trafilaturaGemini 翻譯1047 words

Guided by the FSC and convened by CTBC Financial Holding, 16 iconic financial holdings and banks in Taiwan have jointly launched the "FinLLM" project, aiming to build localized financial AI based on the sovereign AI corpus from the MODA. (Previous coverage: The Taiwan government is actively deploying "Sovereign AI"; NSTC Minister Wu Cheng-wen: Establishing a National AI Strategy Committee) (Background: 11 leading insurance companies in Taiwan launch the "Claims Alliance Blockchain"; single-point filing for multi-company claims makes the process easier) 6 financial holdings, 2 state-owned banks, along with the Taiwan Academy of Banking and Finance (TABF), the Institute for Information Industry (III), and the NCCU FinTech Research Center, are working together to achieve one goal: Taiwan must have its own financial large language model. This project, named FinLLM, was initiated by the FinTech industry alliance under the guidance of the FSC, with CTBC Financial Holding serving as the convener. It was officially launched today (23rd) at FinTechSpace. From "Individual R&D" to "Collective Collaboration" In the past, every financial holding company wanting to develop AI had to collect its own corpus, train its own models, and conduct its own evaluations. High computing costs, fragmented data licensing, and individual legal compliance risks resulted in mediocre outcomes—a lot of money was burned, but few practical application scenarios were realized. The structure is different this time. The 16 participating institutions include CTBC Financial Holding, Cathay Financial Holding, Fubon Financial Holding, Taishin Financial Holding, SinoPac Financial Holding, Taiwan Cooperative Financial Holding, Mega Financial Holding, First Financial Holding, Hua Nan Financial Holding, KGI Financial Holding, Chang Hwa Bank, Bank of Taiwan, Land Bank of Taiwan, Taiwan Business Bank, Chunghwa Post, and Next Bank, bringing almost all major players in Taiwan's financial industry into the same project. Together with research and technical units such as the TABF, III, NCCU FinTech Research Center, and Asia Pacific Intelligent Machines, FinLLM intends to integrate the entire process from data licensing and model training to evaluation and business model maintenance. More importantly, the legal risks of data licensing can be shared, which is often more challenging for the financial industry than the cost of computing power. The Foundation is the MODA's Sovereign AI Corpus FinLLM is not a model trained from scratch. Its foundation is the sovereign AI corpus currently being built by the MODA. Sovereign AI corpus. Translated, this means a state-led initiative to collect local Taiwanese language, regulations, and archival data to serve as the training foundation for local AI models. This project has been included as one of the action plans for the national-level "New Ten Major AI Constructions," holding a significant status. The role of the financial alliance is to take this general foundation and further feed it with finance-specific corpus: FSC regulations, corporate governance codes, bank operational manuals, financial product prospectuses, and risk disclosure documents. No matter how powerful US-based large models are, they may not be able to fully answer questions within the context of Taiwan's financial supervision. For example, if you ask GPT about a clause in Taiwan's Financial Consumer Protection Act, it might provide a general explanation, but when it encounters concepts unique to Taiwan's supervision, such as "offshore structured products" or "suitability reviews for elderly clients," the accuracy of its answers may drop. The alliance conducts training using legally authorized data and, at the same time, establishes standardized evaluation benchmarks through joint deliberation by the member banks.

Data Status✓ Full text extractedRead Original (動區 BlockTempo)

🔍Historical Similar Events· Keyword + Asset Matching0 items

No similar events found (requires more data samples or embedding search; currently MVP keyword matching)

Raw Information

ID:cda7d9b83b

Source:動區 BlockTempo

Published:2026-04-23 08:04:27

Category:zh_news · Export Category zh

Symbols:Unspecified

Community Votes:+0 / −0 · ⭐ 0 Important · 💬 0 Comments