All Sessions：

大型語言模型（LLM）的快速發展為各行各業帶來了重大機遇，但其成功部署需奠基於任務設計、資料集創建與評測這三大核心工作。基於我在 TAIDE 第一、二期計畫中的經驗，本次演講將深入探討在推動 LLM 落地過程中所面臨的挑戰，如如何確保資料品質和提升模型的可擴展性，並分享我們如何克服這些挑戰的具體方法。這些工作對於確保 LLM 能夠在不同領域實現擴展和應用至關重要。透過健全的資料集建設、有效的任務設計與嚴謹的評測，台灣正逐步邁向成為全球 AI Foundry 的目標，推動創新與技術領導力。本次討論也將闡明這些關鍵步驟如何為各行業及社會帶來深遠影響。 The rapid development of Large Language Models (LLMs) offers significant opportunities across industries, but their deployment relies on a strong foundation in task design, dataset creation, and evaluation. Based on key insights from the first and second phases of the TAIDE project, this talk will highlight the specific challenges we faced, such as ensuring data quality and scalability, and how we addressed them. These foundational efforts are crucial for scaling LLMs and ensuring practical applications in various sectors. By focusing on building robust datasets, designing impactful tasks, and establishing rigorous evaluation frameworks, Taiwan is well-positioned to accelerate its journey towards becoming a global AI Foundry, driving forward both innovation and leadership. This talk will also explore how these critical steps create far-reaching benefits, not only for industries but also for society as a whole.

This presentation primarily explores how generative AI is revolutionizing knowledge management. It begins by elucidating the necessity of knowledge management and the limitations of traditional approaches. The presentation then introduces how generative AI optimizes knowledge management processes through its comprehension capabilities, retrieval-augmented generation techniques, and integration with external tools. Several real-world case studies are presented, demonstrating the application of generative AI across various domains, including disaster prevention and relief, financial information, and construction. The benefits and risks of implementing generative AI in knowledge management are discussed, including considerations for information security. The presentation concludes by emphasizing the future directions of generative AI in knowledge management, anticipating its emergence as an essential tool for knowledge workers. 這份簡報主要探討生成式 AI 如何革新知識管理。首先說明知識管理的必要性，以及傳統知識管理的局限性，接著介紹生成式 AI 如何透過理解能力、檢索增強生成技術、以及與外部工具整合，來優化知識管理流程。簡報還舉出多個實際案例，展示生成式 AI 在不同領域的應用，例如災害防救、金融資訊、建築營造等，並討論導入生成式 AI 知識管理的效益和風險，包括在資訊安全上的考量。簡報最後強調生成式 AI 知識管理的未來發展方向，預期將成為知識工作者的重要工具。

Artificial intelligence (AI) stands as humanity's most pivotal invention, possessing the potential to reshape society profoundly, akin to the transformative power of fire and electricity. While earlier innovations such as the wheel and the printing press laid the groundwork for societal development, AI introduces unprecedented complexities, particularly concerning trustworthiness and fairness. As AI systems increasingly govern decision-making, it is crucial that these technologies operate transparently to prevent biases that could exacerbate existing inequalities. In this talk, we will present our recent work on addressing these issues, including data collection, privacy, and robotics, while reflecting on our choices regarding AI technology as we navigate the future.

In this talk, we will present E.SUN's experiences in the application and development of generative AI. Besides the commonly seen conversational AI platforms, we will also discuss methods for rapidly embedding generative AI into existing systems, as well as cases of generating news summaries and meeting minutes through prompt engineering. Lastly, we will share some other interesting cognitive computing cases that can be done using generative AI.

Large Language Models (LLMs) have been a major focus in the tech world for nearly two years, sparking debates over whether they are overhyped or genuinely transformative. In this talk, I will explore how CyCraft Technology has harnessed LLMs across various cybersecurity domains, focusing on three critical aspects: - The Good: Discover how LLM-empowered products significantly enhance the efficiency of security experts, reducing incident response times and revolutionizing threat management. - The Bad: Uncover the domain-specific challenges we encounter when integrating LLMs into cybersecurity, including issues of scalability, accuracy, and adaptation to rapidly evolving threats. - The Ugly: Examine the fundamental security vulnerabilities inherent in LLMs, highlighting the risks they pose and the potential for exploitation by malicious actors. Join us for an insightful journey into the practical applications, challenges, and pitfalls of deploying LLMs in the cybersecurity landscape.

By leveraging digital transformation and AI-driven insights, and incorporating international trends and regulations, AI technology drives growth in businesses and investment market value, meeting the demands of niche missions and talent needs. As domestic industries in Taiwan face the advent of the AI era, the need for digital transformation has accelerated. Meanwhile, organisations are tackling the challenges of merging computational power on the production side with AI resources on the application side. Through case studies in smart logistics and retail services within the context of Industry 5.0, Deloitte demonstrates how AI can optimise logistics and enhance consumer experiences, while also interweaving sustainable development and circular power systems to establish a one-stop ecosystem. Furthermore, this ecosystem fosters the integration of hardware and software, process management, and talent empowerment, paving the way for a brand new, win-win business model. 透過數位轉型與 AI 驅動的洞察，整合國際趨勢與規範，帶動技術與投資市值增長，滿足特定任務與人才需求。國內產業面對 AI 的時代的來臨，加速了數位轉型的需求。面對生產端算力與應用端 AI 資源整合的挑戰。Deloitte 透過工業 5.0 的智慧物流和零售服務案例，說明使用 AI 來最佳化物流與增加消費者體驗，同時結合永續發展及循環電力系統，建立一站式生態圈，實現軟硬體整合、流程管理與人才賦能，開創共贏的新商業模式。

Mr. Peter Hu, CIO of FET, will explore the challenges and opportunities of AI adoption in enterprise, such as lack of scenarios, monetization difficulties, ambiguous investment decisions, and insufficient .Peter Hu effective data. He will discuss how FET enhances productivity, fosters innovation, drives data-driven decisions, manages net-zero targets, and ensures digital resilience and security. This session will also cover how these capabilities align with FET's technological, organizational, cultural, and financial transformation strategies.

This research focuses on the challenges faced by Small and Medium Enterprises (SMEs) manufacturers in adopting AI technologies. Despite their successful decades-long operations, SMEs are seeking ways to leverage AI for enhancing product value and manufacturing efficiency. However, the lack of resources and collaborative development talent in AI technology presents a significant barrier. To address these challenges, the study proposes a shared AI software platform where SMEs can utilize AI from design to operation, collaborate with each other, share resources through open-source, and add proprietary features for their businesses. Specifically, the research aims to establish a lathe machining quality and tool damage diagnostic system using AI models. This system is expected to predict machining errors and tool damage, ensuring quality control without extensive manual inspection, reducing inspection costs, and increasing tool lifespan. The goal is to cater to the current machining demands of small batches with quick changeovers, overcoming the lack of versatility in existing technologies.

We will share our view of recent LLM development from different aspects, like open-source vs. proprietary, use vs. training, application and product, and etc. We will also share several applications with scenario in industries based on our experience.

91APP has been dedicated to serving the retail industry for the past decade, assisting over 1,000 medium to large retail brands in their digital transformation. In recent years, the company has further advanced by promoting AI transformation. The aim of introducing new technologies is to solve industry-specific problems. This topic will begin with the fundamental issues of the retail industry, discussing the role of AI in retail, how 91APP develops AI models for retail, and predicting the future landscape of the retail sector. 91APP 專注於服務零售產業，10 年來協助 1,000 多家中大型的零售品牌數位轉型，近年來更近一步推動 AI 轉型。導入新科技，目標都是要解決產業問題。本講題將從從零售業的基本問題出發，討論 AI 在零售業內的角色，91APP 如何發展零售的 AI 模型，並預測未來零售業的面貌。

AI is leading a revolution in music streaming, reshaping user experiences and industry dynamics. This session will examine how AI innovations drive personalized recommendations, influence content distribution, and open new paths for engagement. Discover how AI is setting the stage for the future of music streaming. AI 正在引領音樂串流的革命，重塑用戶體驗和產業動態。本次演講將探討 AI 創新如何推動個性化推薦、影響內容分發並開創新的互動方式。了解 AI 如何為音樂串流的未來鋪路。

In this session, we will explore how an open platform for enterprise ready AI can address the fragmentation challenges of generative AI (GenAI) technology. While open-source projects have accelerated AI innovation, they have also led to the dispersion of technologies and tools, creating barriers for enterprises to adopt GenAI. Developers face numerous options and struggle with integration. To tackle this issue, we need a robust open-source collaboration framework to build and evaluate composable GenAI solutions. This framework will provide a systematic approach to grading and assessing GenAI based on performance, reliability, scalability, and flexibility, ensuring its suitability for enterprise needs. Additionally, this collaborative model will help address current challenges in retrieval-augmented generation (RAG) technology and provide an innovative platform for developers to unlock the full potential of GenAI. 在此次演講中，我們將探討企業級 AI 的開放平台如何應對生成式人工智慧（GenAI）技術的碎片化挑戰。開源專案推動了 AI 的迅速創新，但也導致了技術和工具的分散，對企業採用 GenAI 形成障礙。開發人員面臨眾多選擇，難以整合。為解決這一問題，我們需要一個強大的開源協作框架，以建構和評估可組合的 GenAI 解決方案。這個框架將根據性能、可信度、可擴展性和彈性對 GenAI 進行分級和評估，確保其適合企業需求。同時，這種合作模式將有助於解決目前 RAG（檢索增強生成）技術的挑戰，並為開發人員提供創新平台，發揮 GenAI 的最大潛力。

Starting from the four major cycles of IT development, this outline explains how the era of AI marks the fifth cycle, referred to as the "Age of It," and explores its profound impact. Addressing the specific needs of Taiwanese businesses, Dr. Tsai will introduce a strategic framework of "Five Dimensions of AI Application" to help enterprises find optimal solutions in improving productivity, integrating digital platforms, connecting with the AI ecosystem, implementing AI models, and addressing AI ethics. In the end, Dr. Tsai will share tangible "low-hanging fruits" for businesses applying AI.

The demand for AI has been growing rapidly across various sectors, driven by its potential to revolutionize industries and enhance efficiency, decision-making, and innovation. Computing is a foundational requirement for AI as it provides the hardware and software infrastructure necessary for process data, perform calculations, and run complex algorithms. Arm’s energy efficient foundations and pervasive footprint from cloud to edge, make us the AI platform the world can rely on, both today and in the future. In this presentation, we will share Arm’s vision and technologies building for AI world.

AMD not only has the broadest AI product portfolio but also leads in silicon design and packaging technologies. AMD pioneers chiplet modular design and 3.5D packaging in AI Accelerator. Together with Open and Proven software/hardware stack, AMD aims to advance the AI development with the ecosystem through deep collaboration.

The emergence of generative artificial intelligence (AI) at the end of 2022 marked a significant evolution, as these technologies increasingly integrate into daily human activities. The scale of generative models has been expanding annually at an exponential rate, with architecture advancements evolving from single-model frameworks to complex multi-expert systems. As industries across the board incorporate generative AI into their operational processes, concerns regarding information security and controllability have come to the forefront. However, the high deployment costs associated with generative AI technologies pose a significant barrier to widespread adoption. In response, Phison has introduced the aiDAPTIV+ initiative, a revolutionary approach that leverages NAND Flash to expand High Bandwidth Memory (HBM), thereby substantially reducing costs.

Abstract: This presentation delves into the forefront of medical artificial intelligence, particularly focusing on the role of agentic Large Language Models (LLMs) in transforming healthcare. These advanced models exhibit agency by autonomously generating realistic and contextually appropriate medical data and interventions, marking a shift from theoretical exploration to tangible, real-world applications. This evolution emphasizes the need for a unified, interdisciplinary approach. The session starts with an analysis of how agentic LLMs have been seamlessly integrated into existing healthcare frameworks, notably enhancing the accuracy and efficiency of disease prognosis and diagnosis. Specific case studies, such as the use of these models in diagnosing acute ischemic stroke, highlight how agentic LLMs are instrumental in critical health domains. Acute ischemic stroke, a leading cause of disability and mortality worldwide, requires rapid and accurate diagnosis to initiate timely intervention, which can significantly reduce long-term disabilities and improve outcomes. This illustrates the crucial role of agentic LLMs in improving diagnostic accuracy and refining medical decision-making processes. The objective of this presentation is to provide a rigorous academic perspective on the potential and current applications of agentic LLMs in healthcare. It aims to foster a comprehensive understanding of their impact on the future trajectory and developmental prospects of medical AI.

Taichung Veterans General Hospital (TCVGH) has been recognized as the only hospital in Taiwan to be listed in Newsweek's top 300 World's Best Smart Hospitals in 2023 (Ranking #254) and 2024 (Ranking #246). By the survey of Healthcare Information Management System Society (HIMSS) of United States, Taichung Veterans General Hospital (TCVGH) is the Second Highest Global Ranking of DHI (Digital Healthcare Indicator) Score in 2024. TCVGH was also the first National Medical Center in Taiwan to achieve HIMSS EMRAM Stage 6 and 7 validations in 2023, it takes only 8 months, also created the fastest record in Taiwan. In the National Recognition of Taiwan, Taichung Veterans General Hospital (TCVGH) is the first hospital in Taiwan to Achieve 3 Times of Smart Hospital Award (2015, 2019 and 2023). Taichung Veterans General Hospital (TCVGH) also deployed hospital information system to 12 Regional Branch Veterans Hospitals around Taiwan. In this session, we will introduce Digital Strategy, Hospital Information System, and some innovations of Smart Hospital."

Jubo is a leading SaaS provider for nursing information systems. Its solutions have been successfully implemented in over 1,200 long-term care facilities across Taiwan, Canada, and the United States. Dr. Kang, the founder of Jubo, will introduce four key AI technologies embedded in Jubo’s solution, each designed to support and enhance the work of nursing professionals. The first AI focuses on detecting abnormalities in vital signs, ensuring timely interventions. The second AI is dedicated to wound detection, aiding in accurate and early diagnosis. The third AI automatically summarizes nursing records between shifts, streamlining communication and reducing errors. The fourth AI aims to reduce the time nurses spend on family communication, allowing them to focus more on patient care. In his presentation, Dr. Kang will share insights from both an academic and commercial perspective. He will discuss his experiences in integrating AI into nursing practice and the impact of these technologies on the healthcare industry.