<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by (KJH) Kuan-Jung, Huang on Medium]]></title>
        <description><![CDATA[Stories by (KJH) Kuan-Jung, Huang on Medium]]></description>
        <link>https://medium.com/@mybaseball52?source=rss-4bcc562a9bb9------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*q93HOXipNFqKbq3xCVR9_A.jpeg</url>
            <title>Stories by (KJH) Kuan-Jung, Huang on Medium</title>
            <link>https://medium.com/@mybaseball52?source=rss-4bcc562a9bb9------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Mon, 16 Nov 2026 21:54:53 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@mybaseball52/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[探討大型前端專案中 Named Export 與 Default Export 的選擇與實務考量]]></title>
            <link>https://mybaseball52.medium.com/named-vs-default-export-frontend-26f4f2c0bac7?source=rss-4bcc562a9bb9------2</link>
            <guid isPermaLink="false">https://medium.com/p/26f4f2c0bac7</guid>
            <category><![CDATA[nextjs]]></category>
            <category><![CDATA[frontend-development]]></category>
            <category><![CDATA[default-export]]></category>
            <category><![CDATA[code-quality]]></category>
            <category><![CDATA[named-export]]></category>
            <dc:creator><![CDATA[(KJH) Kuan-Jung, Huang]]></dc:creator>
            <pubDate>Sun, 14 Sep 2025 04:11:14 GMT</pubDate>
            <atom:updated>2025-09-14T04:11:14.972Z</atom:updated>
            <content:encoded><![CDATA[<h4>深入比較 Named Export 與 Default Export，找出適用於大型團隊的最佳解</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*djbxWPedRZjck3UeXVUAVw.png" /></figure><p>在前端開發實務中，ES Module 的 export default 與 named export（具名匯出）兩者各有應用場景。然而，隨著專案規模增大、團隊協作人數增加，選擇適當的 export 方式，會對於程式碼可維護性、重構效率產生顯著影響。</p><p>本文整理近期團隊的討論與實際經驗，說明為何我們傾向以 named export 作為標準，僅於特定情境（如 Next.js 頁面元件）保留 default export。</p><h3>技術問題說明</h3><h4>1. Default Export 的彈性與其風險</h4><p>export default 提供了彈性：import 時可以自由命名元件：</p><pre>// Tooltip.js<br>export default Tooltip;<br><br>// 可被這樣 import<br>import Tooltip from &#39;./Tooltip&#39;;<br>// 也可被這樣 import<br>import CustomTooltip from &#39;./Tooltip&#39;;</pre><p>這種彈性，雖然在小型專案或個人專案時帶來便利，但在大型專案、多位開發者共同維護下，容易導致以下問題：</p><ul><li><strong>追蹤困難</strong>：預設匯出允許任意命名，在重構或查找元件時，全專案搜尋元件名稱會遺漏引用（例如有的檔案取名 CustomTooltip，全域搜尋 Tooltip 無法找到）。</li><li><strong>重構風險</strong>：若進行元件名稱變更或搬移時，容易遺漏部分 import 未正確更新，產生 runtime error 或難以察覺的 bug。</li></ul><h4>2. Named Export 的一致性與安全性</h4><p>具名匯出要求 import 時名稱必須一致：</p><pre>// Tooltip.js<br>export { Tooltip };<br><br>// 必須這樣 import<br>import { Tooltip } from &#39;./Tooltip&#39;;</pre><p>其帶來的好處包括：</p><ul><li><strong>可維護性提升</strong>：專案內引用名稱統一，有利於全域搜尋與批次替換，重構時風險顯著降低。</li><li><strong>團隊溝通效率提升</strong>：新進成員或跨團隊協作時，不易因為 import 命名不一致產生誤解。</li><li><strong>IDE 支援性更佳</strong>：許多 IDE 與型別工具（如 TypeScript）對 named export 有更明確的自動補全與 refactor 支援。</li></ul><p>以我們近期實際參與的大型 React 專案為例，早期部分元件採用 default export。後續在進行重構（如元件分離、合併或重新命名）時，發現多數 bug 來自於：</p><ul><li>預設匯出的元件，在各檔案被以不同名稱 import，導致 refactor 時無法完整覆蓋。</li><li>部分測試與自動化工具因為 import 名稱不一致而失效。</li><li>pull request code review 時，發現命名無法統一管理，溝通成本上升。</li></ul><p>因此團隊內部達成共識，逐步將所有通用元件、hook、工具函式全面改為 named export，僅針對 Next.js 之 page、layout、middleware 等框架強制要求 default export 的檔案維持原本寫法。</p><h3>結論</h3><p>在大型專案、多人協作情境下，統一採用 named export 能顯著提升程式碼可維護性，減少重構與跨檔案搜尋的風險。預設匯出的彈性雖有其適用場合，但應謹慎評估是否符合團隊實際需求。建議團隊在專案初期即制定明確的 import/export 標準，以降低日後技術債累積的機率。</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=26f4f2c0bac7" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Perception on CFA L2]]></title>
            <link>https://mybaseball52.medium.com/perception-on-cfa-en-l2-ed09d5ce33cf?source=rss-4bcc562a9bb9------2</link>
            <guid isPermaLink="false">https://medium.com/p/ed09d5ce33cf</guid>
            <category><![CDATA[cfa-exam]]></category>
            <category><![CDATA[cfa-preparation]]></category>
            <category><![CDATA[cfa-l2]]></category>
            <category><![CDATA[cfa]]></category>
            <category><![CDATA[tips-to-pass-exam]]></category>
            <dc:creator><![CDATA[(KJH) Kuan-Jung, Huang]]></dc:creator>
            <pubDate>Sat, 05 Jul 2025 04:42:47 GMT</pubDate>
            <atom:updated>2025-07-05T04:42:47.611Z</atom:updated>
            <content:encoded><![CDATA[<h4>After many months of studying, I am very excited to share that I have passed the #CFA L2 examination today.</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/612/1*1-vUrsm6Yk-s_3rRvzoCsA.png" /><figcaption>Passed L2 Exam</figcaption></figure><h3>Preface</h3><p>As I mentioned in my reflection on the CFA Level 1 exam, my background is in engineering, and my daily work primarily involves software development. Despite not coming from a finance-related field, I managed to pass Level 1 after a year of dedicated preparation. With that accomplishment — and having gained some financial knowledge through running my own company — I decided to continue my journey and take on the challenge of Level 2, even as a newcomer to the world of finance.</p><p>After completing the Level 1 exam in August 2023, I immediately transitioned into preparing for Level 2. During this time, I had multiple discussions with my instructors to refine my study strategy and ensure I was in the best possible shape for the exam.</p><p>You might wonder: if I started preparing for Level 2 in August 2023, why did I not sit for the exam until May 2025? The answer lies in an unexpected health issue I encountered in 2024. I had to undergo emergency surgery, which made it impossible for me to attend the originally scheduled exam. By the time I recovered, the November 2024 session was already fully booked. As a result, the earliest available date I could register for was May 2025.</p><p>In this article, I will also share the process of how I requested a deferral from the CFA Institute, what the waiting period and outcome were like, and how I navigated the emotional and practical challenges of adjusting my study plan. I hope my experience can offer some guidance and reassurance to those of you who are preparing for the exam or facing similar setbacks.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*o91Drjrm4tudg2xJhbfPKA.png" /></figure><h3><strong>Study Approach</strong></h3><p>There is one crucial point I urge you to keep in mind: <strong>before diving into CFA Level 2 preparation, it is essential to thoroughly review and truly understand the content from Level 1</strong>. This is especially important if you didn’t devote enough time or follow a structured approach during your Level 1 studies.</p><p>Many Level 2 materials skip over fundamental explanations of certain concepts because the CFA Institute assumes you’re already familiar with them from Level 1. In other words, these concepts are treated as foundational knowledge and are rarely — if ever — reviewed in depth again.</p><p>So, if you realize that much of your Level 1 knowledge has faded, I strongly recommend setting aside <strong>at least one month</strong> for a systematic review of all the core Level 1 topics. During this period, you can also apply what I’ll describe later as <strong>“Preparation Tip #1”</strong>, to ensure you’re not just rereading the material, but truly understanding it and laying a solid foundation for Level 2.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*JwRILqK1U7lPpmDvw6BeoA.png" /><figcaption>MPS around 2650</figcaption></figure><p>Just like I did for Level 1, I chose to continue my studies for Level 2 with <strong>Professor Xiang Wang’s course</strong>. At the time, the 2024 CFA curriculum had just undergone significant changes, and with my workdays often exceeding 12 hours, managing my study time effectively became more critical than ever. The structured materials provided by the course, along with the professor’s clear and methodical explanations, helped me quickly grasp the overarching framework within a limited time. Not only did I understand the details, but I also saw the bigger picture — gaining a clearer sense of the core concepts the CFA Level 2 exam expects us to master. The lecture recordings, in particular, were invaluable for repeated review and deeper absorption.</p><p>Overall, my study resources were intentionally kept minimal and focused. Aside from Professor Wang’s course materials, I relied solely on the <strong>official CFA curriculum</strong> and <strong>the CFA Institute’s question bank</strong>. I didn’t purchase any additional prep books or third-party question banks — I made a deliberate choice to stay focused on the official content and course instruction.</p><p>If Level 1 paints a broad blueprint for newcomers to the finance world, then Level 2 is a deep dive into the world of <strong>valuation</strong> — including equity valuation, bond pricing, corporate value assessment, and even the logic behind real estate appraisal. I used to find analysts’ price targets for stocks difficult to decipher, but it wasn’t until I studied Level 2 that I began to truly understand the frameworks and theoretical foundations behind those valuations.</p><p>In my Level 1 reflection, I shared some foundational study techniques. However, the increased complexity and volume of material in Level 2 pushed me to further refine and optimize my approach. Especially in 2023, with the rapid rise of <strong>generative AI</strong>, I began leveraging these tools to boost my learning efficiency and output under intense time constraints.</p><p>In the next section, I’ll share several practical techniques I used — hoping they’ll provide you with helpful strategies for your own journey.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*B6Uiku9lZIcBDltRvchsyA.png" /><figcaption>Creating mindmap is a good way to enhance the memory</figcaption></figure><p>During my Level 2 preparation, I not only built upon the foundation I established in Level 1 but also adopted <strong>more structured strategies</strong> to deepen my understanding and improve retention. Specifically, I created <strong>mind maps</strong> to synthesize key points across chapters, reorganized my study schedule with clear timelines, and managed my progress more intentionally. I also revisited a <strong>personal SWOT analysis</strong> to assess my strengths and weaknesses, allowing me to focus on areas that required targeted improvement.</p><p>For subject-specific study strategies, I followed the guidance provided by <strong>Professor Xiang Wang</strong> during his lectures. His insights aligned closely with the CFA Institute’s expectations — certain topics consistently receive greater emphasis, and his course helped prioritize these effectively.</p><p>When it came to reviewing mistakes, I took a <strong>differential analysis approach</strong>. Instead of merely recording the correct answers, I focused on understanding the root of each mistake — <strong>why I got it wrong, what the misconception was, and how to avoid repeating it</strong>. This method not only reinforced the correct concepts but also sharpened my critical thinking and exam instincts.</p><p>In the following section, I’ll dive deeper into the actual techniques and tools I used throughout my study journey. I hope these practical insights will offer useful guidance and inspiration for your own preparation.</p><h3>Study Techniques</h3><h4>Mastering Logical Reading in English</h4><p>For many Taiwanese candidates, <strong>reading comprehension</strong> remains one of the toughest challenges when preparing for English-based exams. This is especially true for CFA Level 2, where questions often begin with a <strong>concise yet information-dense vignette</strong>. Candidates are expected to quickly grasp the scenario and select the correct answer — placing a high demand on one’s ability to process <strong>English logic and structure efficiently</strong>.</p><p>I’ve shared more details about how I personally trained this skill in a separate article on <em>English reading comprehension</em>. Feel free to check it out if you’re looking for more guidance.</p><p>Of course, there are many ways to improve your English reading skills. Aside from my own methods, <strong>one of the most fundamental and effective approaches is extensive reading</strong> of English materials. By continuously exposing yourself to native texts, you’ll gradually reduce the sense of unfamiliarity and begin to internalize sentence structures and logical flow — ultimately helping you read with greater ease and confidence under exam conditions.</p><h4>Teach It Like a Teacher</h4><p>This is a strategy I began applying in early 2025. It came from a realization during the early stages of my Level 2 prep: the exam scope was incredibly broad, and it was easy to get lost in the technical details — especially when it came to formulas, accounting standards, and valuation methods.</p><p>But then I thought: <strong>if I were a teacher or a professional explaining these concepts to a client</strong>, I wouldn’t be able to use “It’s too hard” or “There’s just too much to cover” as an excuse. That’s when I began putting myself into the <strong>role of an instructor</strong> — someone who must explain each concept <strong>clearly, logically, and correctly</strong> to others.</p><p>A practical way to do this is by writing out each concept in your own words on paper, then reviewing what you wrote to see whether it’s accurate and logically sound. This becomes a powerful form of <strong>self-validation and output training</strong>.</p><p>An added bonus: those written explanations become <strong>personalized, high-impact notes</strong> you can review before the exam. I jokingly call this method the “Teacher’s Notebook Technique” (not an official term, of course).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6C5xXL4B2Dt_YuTrAvfXVA.jpeg" /><figcaption>Self-Recitation and Correction Using Notes or Lecture Recordings</figcaption></figure><h4>Clean &amp; Redo: Resetting for Deeper Learning</h4><p>In the early stages of preparing for CFA Level 2, many candidates — myself included — experience a significant sense of frustration. Compared to Level 1, the question format in Level 2 is a major shift. Level 1 questions are typically short and straightforward, often just a single sentence followed by three answer choices. As long as you’ve attended the lectures and grasped the basic concepts, you can usually maintain a decent accuracy rate.</p><p>Level 2, however, presents its first major challenge in the form of <strong>increased reading complexity</strong>. Questions evolve from brief statements to <strong>lengthy, scenario-based vignettes</strong>. In particular, the official CFA question bank features long-form case studies, each followed by multiple questions. The reading load and difficulty of comprehension rise sharply. Many questions require you to <strong>refer back to specific paragraphs</strong>, and it’s common in the beginning to feel completely lost — wondering which part of the text the question is referring to.</p><p>The <strong>second challenge</strong> is the gap between understanding and execution. Even after completing a lecture, many candidates find themselves unable to answer questions. Common reasons include:</p><ul><li>Forgetting the structure of a specific valuation formula</li><li>Not knowing which formula to apply in a given context</li><li>Misunderstanding a concept, leading to the wrong calculation approach</li></ul><p>To overcome these issues, I adopted a strategy of <strong>repeated practice and resetting the question environment</strong>. I did nearly all my practice using the <strong>CFA computer-based question system</strong>, and after completing each chapter or mock exam, I would <strong>reset the entire quiz and redo it from scratch</strong>. During the final stretch before the exam, I made a habit of <strong>resetting and retaking full mock exams every 2–3 days</strong>, reinforcing both familiarity and application skills.</p><p>Throughout this process, I rarely used printed materials. I only switched to paper when I needed to rest my eyes or when I was too mentally drained and wanted to lower the barrier to studying — using printed questions for a quick scan or casual review. It’s worth noting that the <strong>CFA official question bank</strong> presents each vignette as a full-length passage followed by multiple questions. At first, this format can feel overwhelming, with a flood of information and no clear indication of which paragraph to focus on. That’s why <strong>consistent repetition and controlled pacing</strong> in practice are absolutely essential.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8iZKIAUke6rT-GCNedQfWw.jpeg" /><figcaption>You can create your own quizs</figcaption></figure><p>As I entered the later stages of preparation, I experimented with more <strong>flexible — but still effective — review techniques</strong>. One of my go-to methods was printing out practice questions from the textbook and converting them into <strong>fill-in-the-blank exercises</strong>. Compared to rewriting full answers from scratch, this method felt lighter, but still required serious thinking — <strong>“Which formula goes here?”</strong> or <strong>“What concept applies in this context?”</strong> It struck a perfect balance between low burden and high efficiency.</p><p>I also organized these fill-in-the-blank sheets and <strong>brought them with me to the exam center</strong>. During the mid-session break, flipping through these personalized notes helped me quickly <strong>recall key concepts</strong> and also served as a way to <strong>calm my nerves</strong>, keeping me focused and mentally aligned for the second half of the exam.</p><h4>Collaborating with Generative AI</h4><p>During my Level 2 preparation, I also began using <strong>AI as a support tool</strong>, primarily for <strong>translation</strong>. Some of the materials and question prompts were quite lengthy, and reading them in English under time constraints could be taxing. By translating them into Chinese, I was able to more quickly grasp the key points and logical flow.</p><p>That said, I was also fully aware of <strong>AI’s limitations</strong>, especially the issue of <em>hallucinations</em> — such as mistranslations, omissions, or even entirely fabricated content. To avoid misunderstandings, I always <strong>cross-checked the AI-generated translation against the original English</strong>, verifying both accuracy and completeness.</p><p>Interestingly, this review process became a powerful learning tool in itself. Whenever I caught an incorrect translation — whether it was a misused concept or a poorly interpreted term — I paid closer attention to it, which helped solidify the correct meaning in my mind. In this way, combining <strong>AI assistance with active review</strong> unexpectedly became one of the <strong>most effective memory techniques</strong> in my study toolkit.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*JBwI8XkgeB1Ygp3n-GSo7Q.png" /><figcaption>ChatGPT that help me translate the CFA Program’s content</figcaption></figure><h4>Don’t Rush into Question Drilling</h4><p>If you’re someone who’s used to taking certification exams and aiming for fast-track success, you might be inclined to rely on <strong>heavy question drilling and reverse-engineering</strong> the test to identify key topics — perhaps even trying to predict the examiners’ preferences. For CFA Level 1, that approach might still work to some extent. Most questions there focus on testing basic concepts, and with enough repetition, you can start to recognize common patterns and directions.</p><p>However, <strong>this strategy quickly loses effectiveness at Level 2</strong>. First, the question format shifts from single-sentence prompts to <strong>lengthy, scenario-based vignettes</strong>. These aren’t simply testing one concept at a time. Instead, you’ll need to read and process a significant amount of information, determine what the question is truly asking, and decide which formula or valuation framework to apply. It challenges your comprehension, reasoning, and memory all at once.</p><p>Second, <strong>Level 2 emphasizes the <em>application</em> of concepts and the <em>logic</em> behind your answers</strong>. Mistakes often don’t come from misapplying a formula, but from misunderstanding the underlying scenario or misinterpreting the context — causing your entire line of reasoning to go off course. This can lead to repeated errors and a draining sense of discouragement that erodes motivation over time.</p><p>That’s why I strongly recommend adopting a <strong>textbook-centered deep learning strategy</strong> for Level 2. Rather than rushing through endless questions, go back to the core materials: work through the examples in the textbook, read the <strong>Learning Outcome Statements (LOS)</strong> multiple times, and build your understanding from the foundational concepts upward. While this may feel slower than brute-force question drilling, it results in <strong>deeper comprehension and stronger long-term retention</strong>.</p><p>For example, <strong>equity valuation</strong> was my weakest area. I often got the <strong>Terminal Value</strong> calculation wrong — either forgetting key steps or confusing the logic entirely. But instead of skipping ahead, I practiced similar question types <strong>seven to ten times</strong>, gradually untangling the relationships between present value and terminal value. Eventually, I gained a clear understanding and could apply the valuation concepts with confidence.</p><h3>Preparing for the Exam After a Deferral</h3><p><strong>Eligibility and Costs of Postponement</strong></p><p>This section is especially for those who may be <strong>facing a deferral</strong> of their CFA exam.</p><p>If you’re considering a deferral simply because you don’t feel ready, I urge you to think carefully. The CFA Institute has published data showing that the <strong>pass rate for deferred candidates is only around 35%</strong>. Their analysis suggests that many candidates defer due to lack of preparation — but fail to meaningfully improve their readiness during the extended time, leading to disappointing results. So if your decision to defer is driven purely by anxiety, it may actually trap you in a longer cycle of procrastination.</p><p>That said, <strong>life is unpredictable</strong>, and sometimes deferring the exam truly is the only option. That was certainly the case for me.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mKIZVJEM_LzPvcVpz86YZw.jpeg" /><figcaption>It is important to have a health body</figcaption></figure><p>After returning to Taiwan from a business trip in May 2024, I began experiencing <strong>severe lower back pain</strong> — so intense that even walking became a struggle. Within just two minutes of walking, I would develop <strong>intermittent claudication</strong>, making it difficult to move around normally. For the entire months of May and June, I found myself <strong>constantly going back and forth between hospitals and clinics</strong>.</p><p>At first, I assumed it was simply fatigue or a muscle strain. However, after undergoing an MRI in June and following up at the end of the month, I was shocked to learn that the condition was <strong>far more serious</strong> — a compressed nerve was causing the pain, and my doctor suspected that <strong>surgery might be necessary</strong>. On July 2, after a detailed consultation, surgery was officially recommended.</p><p>Naturally, my first thoughts were about my health — but immediately afterward came a wave of disbelief:<br> <strong>“What about the exam?”</strong><br> I couldn’t believe things had escalated to the point of needing surgery.</p><p>I underwent the operation on <strong>July 15, 2024</strong>, and during my hospital stay, one moment remains especially vivid: I was still lying in bed, listening to CFA Level 2 lecture recordings… and eventually drifted off to sleep with them playing in the background (<em>laughs</em>).</p><p>Unfortunately, my doctor advised that during the recovery period, I <strong>should not remain seated for more than 20 minutes at a time</strong>, and this restriction would last for at least three months. Under such conditions, it was clear I would be unable to take the exam as originally planned. I had no choice but to <strong>request a deferral from the CFA Institute</strong>. (You can find more about CFA’s exam deferral policy [here].)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*C2v5YBpiLi6PlKv_UXIwhg.png" /><figcaption>The content for requesting postpone</figcaption></figure><p>The <strong>CFA Institute has a clear and structured policy</strong> for exam deferrals. If you’re requesting a deferral due to health reasons, you’ll need to prepare the following documents:</p><ul><li>A <strong>medical certificate</strong> issued by your attending physician</li><li>A <strong>letter of explanation</strong> written by the doctor, outlining your condition, treatment plan, and why you are unable to sit for the exam (the CFA Institute provides a standard template for this)</li></ul><p>These documents must be uploaded through a <strong>designated CFA submission form</strong>. Once submitted, the request is reviewed by the Institute’s <strong>medical consultant team within three business days</strong>, and they will notify you whether your deferral has been approved.</p><p>In my case, the CFA Institute acknowledged that my condition <strong>met the criteria for deferral</strong>, but it did not qualify for a <strong>fee waiver</strong>. As a result, I was required to pay a <strong>$300 rescheduling fee</strong>. Fortunately, my surgery was covered by insurance, so I was able to manage the additional expense without financial strain.</p><blockquote>If you’re facing <strong>unforeseen health or life challenges</strong>, don’t blame yourself or fall into anxiety — <strong>the CFA system does allow room for flexibility and support</strong>. The key is to ensure that you have all the required documentation and submit your application <strong>as early as possible</strong>.</blockquote><p>However, if you’re thinking about deferring simply because you <strong>feel underprepared</strong>, I urge you to evaluate the decision carefully. <strong>Deferral is not a magic solution.</strong> Only consistent effort and follow-through will ultimately increase your chances of passing the exam.</p><h4>When You’ve Overprepared</h4><p>The term “<strong>overprepared</strong>” here doesn’t refer to anxious, obsessive cramming. Rather, it describes a <strong>rational yet disoriented state</strong> — where you’ve already reached a solid level of readiness, but still find yourself feeling uncertain or lost.</p><p>This experience is especially relevant for candidates who:</p><ul><li>Have a <strong>clear understanding of their strengths and weaknesses</strong></li><li>Score <strong>consistently well on mock exams</strong></li><li>And have a good grasp of the overall material</li></ul><p>Despite these indicators of readiness, you may still feel unsettled — unsure of what else to focus on, or worried that you’re overlooking something critical. If that sounds like you, the next section may offer helpful insights into how to regain clarity and confidence in the final stretch.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*nkOzu4ltpKMQOPBL6OsFEA.jpeg" /><figcaption>The score of the CFA mock exam</figcaption></figure><p>During my CFA Level 2 preparation, I personally encountered this exact situation. I began studying in <strong>August 2023</strong> and continued through <strong>April 2025</strong>. Although there was a four-month gap (from <strong>October 2024 to February 2025</strong>) due to surgery, work commitments, and the Lunar New Year, I still had more than enough overall study time.</p><p>By early <strong>May 2025</strong>, I started to feel a strange sense of <strong>uncertainty about what to do next</strong>. This feeling came from two main sources:</p><ol><li>I had already <strong>completed most of my study goals</strong></li><li>I had spent so much time reviewing the same materials that I could no longer tell whether I was truly <strong>understanding and able to apply the concepts</strong>, or merely <strong>recalling the correct answers from memory</strong></li></ol><p>I had experienced something similar in the past with shorter, more flexible certifications like the PMP. In those cases, once I felt prepared, I could simply schedule the exam and get it over with. But the <strong>CFA Program is a fixed-schedule, long-cycle exam</strong>, and even when you’re at your peak readiness, you may still have to wait weeks — or even months — to sit the test. That waiting period can lead to a frustrating sense of stagnation.</p><p>What eventually helped me was <strong>shifting my attention away from studying for a short period</strong>, and returning to it with a refreshed mindset. Specifically, at the end of April, I began <strong>more intensive physical rehab exercises</strong>, and added activities like <strong>listening to music and doing guided meditations</strong>. These breaks allowed me to <strong>step out of the constant loop</strong> of “work → CFA study → mock exam → back to work” and give my mind new forms of stimulation. It was a kind of mental reset that proved incredibly helpful.</p><p>That said, it’s important to clarify: <strong>this isn’t about breaking your study rhythm or giving up</strong>. It’s about introducing intentional variation to prevent burnout. Once I recalibrated my mental state, I was able to return to the materials with more focus and clarity — and a renewed sensitivity for absorbing and applying the content.</p><h3>Final Reflections</h3><p>Looking back at my Level 1 preparation, I can confidently say that <strong>Level 2 presented a much greater challenge</strong>. Fortunately, I was able to stay the course, complete the entire curriculum, and <strong>pass the exam on my first attempt</strong> — achieving the result I had hoped for.</p><p>I had mentally prepared myself for the possibility of retaking the exam, so on <strong>July 1</strong>, when the results were released, I wasn’t in a rush to check. At the time, I happened to be setting up for a <strong>live-streamed event for the AI community</strong>, and had decided not to look — until I accidentally opened Gmail and saw the subject line: <strong>“Congrats.”</strong> In that moment, I felt an overwhelming sense of joy and relief. Thankfully, that positive emotion didn’t disrupt the event or affect the speakers and audience — it only added a quiet boost of happiness.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*SbbbLA4fnXljWyFKhIczFg.png" /><figcaption>I just on the deck organize an AI agent cohort when I receieve the exam result</figcaption></figure><p>Among all the topics in Level 2, I found <strong>derivatives</strong> the most intellectually stimulating — though ironically, it was also one of the sections where I scored lower. Areas like <strong>corporate finance, equity valuation, and financial reporting</strong> go deeper than in Level 1, and I found that this level of depth encouraged me to think more critically about the structure of <strong>annual reports</strong> — especially how to produce financial statements that reflect <strong>high earnings quality</strong> and <strong>strong reporting quality</strong>, offering true value to the shareholders of a company.</p><p>The classes also stood out for their practical insights. Our instructor didn’t just explain the theories — he <strong>invested time in sharing background context and real-world case studies</strong>, helping us better connect abstract financial concepts to the realities of the market. These references added clarity and left a lasting impression.</p><p>Ultimately, this journey deepened my appreciation for what it takes to communicate financial value — <strong>how to craft an annual report that reflects both financial integrity and strategic vision</strong>, so that shareholders can genuinely see the company’s worth.</p><p>I’m deeply grateful to the instructors whose dedication made this learning experience so rich and impactful. Their thoughtful explanations and practical examples helped bridge the gap between theory and practice, and expanded my understanding of the financial world.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ed09d5ce33cf" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[CFA L2 心得]]></title>
            <link>https://mybaseball52.medium.com/perception-on-cfa-l2-c805c6fbc7dd?source=rss-4bcc562a9bb9------2</link>
            <guid isPermaLink="false">https://medium.com/p/c805c6fbc7dd</guid>
            <category><![CDATA[preparation]]></category>
            <category><![CDATA[cfa]]></category>
            <category><![CDATA[tips-to-pass-exam]]></category>
            <category><![CDATA[cfa-program]]></category>
            <category><![CDATA[cfa-exam]]></category>
            <dc:creator><![CDATA[(KJH) Kuan-Jung, Huang]]></dc:creator>
            <pubDate>Wed, 02 Jul 2025 15:50:30 GMT</pubDate>
            <atom:updated>2025-07-02T16:08:06.993Z</atom:updated>
            <content:encoded><![CDATA[<h4>After many months of studying, I am very excited to share that I have passed the #CFA L2 examination today.</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/612/1*1-vUrsm6Yk-s_3rRvzoCsA.png" /><figcaption>終於通過 L2 考試了</figcaption></figure><h3>前言</h3><p>如同我在 <a href="https://medium.com/p/perception-on-cfa-l1-7c87b8c8fb7b">CFA Level 1 的心得</a>中提到的，我的背景是工程師，平時主要從事程式開發的工作。儘管如此，在努力準備一年的時間後，我順利通過了 L1。作為一名已通過 L1 的考生，加上自身在經營公司過程中累積了一些財務知識，雖然對金融領域仍屬新手，但我仍決定繼續挑戰 L2。</p><p>在 2023 年 8 月完成 L1 考試後，我便無縫銜接 L2 的準備課程，期間也多次與授課老師通話，針對準備策略進行討論，確保能以最佳狀態應對考試。</p><p>或許讀者會好奇，既然我從 2023 年 8 月就開始準備 L2，為何直到 2025 年 5 月才參加考試？實際上，我在 2024 年因突發的身體狀況，不得不緊急接受手術治療，導致無法如期參加原訂考試。當時 2024 年 11 月的考期也已額滿，因此最早能報名的場次便是 2025 年 5 月。</p><p>在這篇文章中，我也將分享我如何向 CFA 協會申請延後考試的流程、申請後的等待與結果，以及在重新調整備考節奏的過程中，我所遇到的挑戰與心境轉變。希望這些經驗能對正在備考或遇到類似狀況的你有所幫助。</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mZ3OnemBE1kOZcTCjWemww.png" /></figure><h3>準備方式</h3><p>有一點非常重要，請務必牢記在心：<strong>在準備 L2 前，一定要將 L1 的內容重新完整複習，並且理解得相當透徹</strong>。尤其是如果你在 L1 的準備過程中沒有投入足夠的時間或沒有系統性地掌握每個主題，那麼這一步就更為關鍵。</p><p>許多 L2 的教材會直接跳過對某些觀念的重點說明，因為 CFA 協會預設這些知識你在 L1 就已經熟悉了。換句話說，這些概念在 L2 是作為<strong>基本知識</strong>存在，鮮少甚至不會再次詳細解釋。</p><p>因此，如果你發現自己對 L1 的內容已經遺忘得差不多，我會強烈建議你<strong>安排至少一個月的時間，系統性地複習 L1 的所有核心主題</strong>。這段時間可以搭配我在後面會提到的「準備技巧 1」，確保你不只是重讀，而是真正理解並建立起與 L2 接軌的基礎觀念。</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*JwRILqK1U7lPpmDvw6BeoA.png" /><figcaption>考試的 MPS 落在 2650</figcaption></figure><p>和 L1 一樣，我在 L2 的準備過程中依然選擇上王祥老師的課程。當時正逢 2024 年教材改版，加上我平日的工作時數經常超過 12 小時，能夠有效分配學習時間變得格外重要。補習班提供的講義和老師的系統性講解，幫助我在有限的時間內快速掌握脈絡，不只見樹也能見林，更清楚 CFA L2 希望我們掌握的核心觀念。特別是老師課堂的錄音檔，也讓我能反覆複習，強化吸收。</p><p>整體來說，我的準備資源非常精簡。除了王老師的講義外，僅搭配 CFA 協會提供的原文教材與官方題庫。我沒有額外購買其他參考書籍或題庫，選擇聚焦在官方內容與課程講解。</p><p>如果說 L1 是為金融領域新手鋪陳的一幅全貌藍圖，那麼 L2 就是深入進入「<strong>評價（Valuation）</strong>」的世界 — — 包括股票的估值、債券的評價、公司價值的判斷，甚至連房地產評價的邏輯都涵蓋其中。以前總覺得分析師給出的股票目標價難以捉摸，直到深入 L2，才真正理解背後的評價架構與理論基礎。</p><p>在 L1 的準備心得中，我已分享了一些基本的學習方法。而面對 L2，更高的難度與更繁重的內容量，讓我不得不最佳化自己的讀書策略。特別是在 2023 年生成式 AI 快速崛起的背景下，我也開始善用這些新工具，在高度壓縮的時間內提升學習效率與輸出能力。後續我會分享幾個具體的方法，希望能提供你實用的參考。</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*B6Uiku9lZIcBDltRvchsyA.png" /><figcaption>製作心智圖讓自己方便整理重點</figcaption></figure><p>在 L2 的準備過程中，除了延續我在 L1 中建立的學習基礎，我也進一步採用了更多結構化的策略來強化理解與記憶。具體而言，我製作了心智圖來統整各章節重點，重新規劃讀書時間並進行自我進度管理，也再次針對自身的優勢與弱點進行 SWOT 分析，幫助我聚焦在關鍵待加強的領域。各科準備方向可以參考王祥老師上課所講的，因為協會的確也會有一定的比例覺得學生要會哪些內容。</p><p>同時，對於錯誤題目的處理，我採取了差異分析的方式，不只是記下正確答案，而是深入理解自己錯在哪裡、為什麼錯、以及如何避免類似的誤區。</p><p>接下來的篇幅，我將詳細分享我在這段準備期間實際採取的策略與具體做法，希望能提供你一些實用的參考與啟發。</p><h3>準備技巧</h3><h4>英文的邏輯閱讀</h4><p>在準備各類英文考試的過程中，對許多台灣考生而言，「閱讀理解」始終是一項不小的挑戰。尤其是 CFA Level 2 的題型設計，往往會以一段篇幅不長、但資訊密度高的案例文章作為題幹，要求考生在短時間內理解情境並選出正確答案，這對英文邏輯閱讀能力提出了更高的要求。</p><p>我在另一篇文章<a href="https://medium.com/@mybaseball52/english-reading-comprehension-e4237715ae4c">英文的閱讀理解</a>中有詳細分享自己是如何訓練這項能力的，歡迎有需要的同學參考。</p><p>當然，提升英文閱讀能力的方法有很多。除了我使用的方法之外，最基本也最有效的方式之一，就是<strong>大量閱讀英文原文文章</strong>。透過持續接觸英文文本，可以幫助你降低閱讀時的陌生感，漸漸建立起對英文語句結構與邏輯的熟悉度，進而在應試時更為從容。</p><h4>把自己當講師</h4><p>這個學習策略是我在 2025 年初開始實行的，起因是我在準備 L2 初期，感受到考試範圍相當廣泛，尤其在各種計算公式、會計認列標準等技術性內容上，容易產生混淆與遺漏。</p><p>但我也意識到，若我們是作為一位老師、或是要向客戶解釋這些觀念的專業人士，是無法用「這太難了」或「內容太多」作為藉口的。我開始嘗試將自己轉換為「教學者的角色」：當有人向你提問，你必須能夠清楚、有邏輯地解釋這個觀念，而且要解釋得正確。</p><p>一個有效的練習方式是：<strong>把這些觀念用自己的話寫在紙上</strong>，再自行檢查是否表達正確、有無邏輯漏洞。這不僅是一種自我驗證的過程，也是一種高效的輸出練習。</p><p>額外的好處是，這些寫下來的內容可以在考前帶著複習，形成個人化的快速筆記。我戲稱這套方法為「<strong>講師筆記法</strong>」（誤。</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6C5xXL4B2Dt_YuTrAvfXVA.jpeg" /><figcaption>自我默寫&amp;看著書或聽錄音檔訂正</figcaption></figure><h4>Clean&amp;Redo</h4><p>在準備 CFA L2 的初期，許多考生會感受到明顯的挫折，這在我自身的經驗中也不例外。L1 的題型相對簡潔，多數題目是一句話描述搭配單一選項選擇，因此即使只是聽完課程、理解基本觀念，作答時仍能維持一定的正確率。</p><p>然而，L2 的第一個難關就是<strong>英文閱讀的轉變</strong>。題型從簡短句子擴展為完整情境文章，尤其 CFA 協會提供的官方題庫，每一題的篇幅都非常長，閱讀負擔與理解難度顯著上升。許多題目甚至必須回頭對照文章段落才能作答，初期常會有「不知道題目問哪一段」的挫敗感。</p><p>第二個挑戰則是<strong>理解到作答的落差</strong>。即使聽完課，也常出現「下不了筆」的狀況，背後常見原因包括：</p><ul><li>忘記評價公式的具體結構</li><li>不知道應該套用哪個公式</li><li>觀念錯誤，導致選錯方法計算</li></ul><p>面對這些問題，我採取的方式是<strong>重複練習與重置題目練習環境</strong>。我幾乎全程使用電腦題庫系統練習，每完成一個章節或模擬試卷後，會選擇「Reset」整個題目，再次從頭作答。特別是在考前階段，我每隔 2～3 天就會重設一次模擬試卷，全新作答，強化熟悉度與應用力。</p><p>這次準備過程中，我很少使用紙本教材。除非在眼睛需要休息時，我才會暫時改用紙本練習；或在較疲憊、想降低學習門檻時，用紙本快速瀏覽文章與題目。值得一提的是，CFA 官方電腦題庫的題目都是<strong>完整文章 + 多題設計</strong>，初次接觸時確實會覺得資訊量龐大、難以定位題目對應的段落，因此練習的次數與節奏控制就變得更加重要。</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8iZKIAUke6rT-GCNedQfWw.jpeg" /><figcaption>可自己製作題目或填空題</figcaption></figure><p>到了備考後期，我也嘗試了一些更具彈性、但依然有效的複習方式。像是我會直接將課本中的題目印出來，以填空的形式進行練習。這種方式相較於完整重寫每一題來得輕鬆，但實際上仍需動腦思考「這裡應該填入哪個公式、哪個觀念才是正確的」，算是一種<strong>低負擔但高效率的複習手段</strong>。</p><p>這些填空練習的紙張，我也會整理起來，在考試當天帶到考場。中場休息時翻閱這些筆記與題目，不只能再次喚起關鍵觀念，也能幫助我穩定情緒，確保下半場保持專注與熟悉度。</p><h4>與生成式 AI 協作</h4><p>在準備 L2 的過程中，我也開始使用 AI 作為輔助工具，主要的應用是在<strong>翻譯</strong>方面。由於部分教材或題目篇幅較長，閱讀全英文內容在時間有限的情況下會造成一定的負擔，因此我會將內容轉換成中文，幫助自己更快速地掌握重點與邏輯。</p><p>不過，我也非常清楚 AI 目前仍存在「幻覺（hallucination）」的問題 — — 例如錯譯、漏字，甚至自行補內容。為了避免錯誤理解，我在使用 AI 翻譯後，<strong>一定會重新對照原文</strong>，檢查翻譯是否正確，有無遺漏或曲解。</p><p>反而也因為這樣的過程，在我「<strong>檢查 AI 的錯誤</strong>」時，更容易留下深刻印象。每當發現某個觀念或詞彙被誤譯，我就會特別留意，進而加深記憶。因此，這種結合 AI 與主動審核的學習方式，對我來說意外成為一種更有效的記憶機制。</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*JBwI8XkgeB1Ygp3n-GSo7Q.png" /><figcaption>ChatGPT 搭配 CFA Program 課本的內容翻譯</figcaption></figure><h4>千萬不要一開始就想著刷題</h4><p>如果你是一位習慣考各類認證、追求快速通關的人，可能會傾向使用「刷題 + 逆向工程」的方式來找出考點，甚至試圖猜測出題者的偏好。在 CFA Level 1，這樣的策略或許還能奏效，因為大多數題目聚焦在基本觀念的測驗，透過大量練習可以快速掌握出題方向。</p><p>但進入 L2 後，這樣的方式就顯得力不從心了。第一，<strong>題型從單句轉變為篇幅較長的案例文章</strong>，不再只是對單一知識點的檢核，而是要能在閱讀完一大段資訊後，判斷題目問的是什麼、該套用哪個公式或哪個評價架構。這對理解力、邏輯推演與記憶力都是挑戰。</p><p>第二，<strong>L2 著重的是「概念的應用」與「邏輯的正確性」</strong>，錯誤往往不只是單一公式的失誤，而是因為前面理解錯誤、判斷失焦，導致整體作答方向偏離。很多時候，一整天的學習熱情可能就在反覆出錯中被磨損，讓人感到沮喪。</p><p>因此，我會更推薦在 L2 階段採取<strong><em>「以教材為核心」的深度學習策略</em></strong>。與其一味地刷題，不如回到課本中，多做例題、多讀幾遍 LOS（Learning Outcome Statement）內容，並練習從底層觀念出發，建立正確的邏輯路徑。這種做法雖然不像刷題來得快速，但對長期記憶與理解力的提升更為扎實。</p><p>以我自己為例，最不熟悉的部分是「股票評價」。一開始連 Terminal Value 的計算都常常出錯，不是忘了步驟，就是概念混淆。但我選擇<strong>反覆練習同一類型題目 7 到 10 次以上</strong>，逐步釐清現值與終值的邏輯關係，最終也能熟練運用並理解其背後的評價思維。</p><h3>延期後的考試準備</h3><h4>延期的資格跟延期的費用</h4><p>這一段，是特別寫給<strong>那些不得不延期考試的朋友</strong>。</p><p>如果你是因為擔心自己準備不夠而主動選擇延期，其實需要特別謹慎考慮。CFA 協會曾公開一項統計，<strong>延期考試的通過率</strong><a href="https://www.cfainstitute.org/about/press-room/2025/cfa-program-results-level-ii-may-2025"><strong>大概只有 35% 左右</strong></a>。他們的分析認為，許多考生因為準備不充分而延期，但延期後也未真正加強準備，導致成效有限。因此，若只是出於焦慮而延期，反而可能讓備考陷入更長期的拖延。</p><p>不過，人生總有意外，<strong>有時延期確實是唯一的選項</strong>。我自己就是如此。</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*mKIZVJEM_LzPvcVpz86YZw.jpeg" /><figcaption>身體要健康，才有本錢做更多事</figcaption></figure><p>在 2024 年 5 月出差返台後，我突然出現嚴重的腰部疼痛，甚至連正常行走都成問題。每走不到兩分鐘，便會出現間歇性跛行的症狀。整個 5 月與 6 月，我幾乎都在醫院與診所間來回奔波。</p><p>一開始我以為只是過勞或肌肉拉傷，但 6 月進行 MRI 檢查後，直到 6 月底回診才確認傷勢嚴重 — — 神經受到壓迫，醫師研判可能需要手術。2024 年 7 月 2 日經過會診後，確定必須開刀治療。當時我第一個想到的，除了健康，就是「考試怎麼辦？」也感到難以置信，自己竟然嚴重到需要動手術。</p><p>我在 <a href="https://www.facebook.com/share/v/16hnnLAdCJ/">7/15 進行手術治療</a>，術後住院期間印象最深刻的，是我還躺在病床上聽 L2 的錄音檔聽到睡著（笑）。但醫師評估我在術後恢復期內，<strong>無法久坐超過 20 分鐘</strong>，這樣的情況至少會持續 3 個月，因此我不得不決定向 CFA 協會申請考試延期，<a href="https://www.cfainstitute.org/about/governance/policies/cfa-program-deferral-policy-faq">考試延期相關的政策在這邊</a>。</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*C2v5YBpiLi6PlKv_UXIwhg.png" /><figcaption>請假信件內容</figcaption></figure><p>CFA 協會對於延期申請有明確政策。若是因<strong>健康因素</strong>申請，需準備以下文件：</p><ol><li><strong>主治醫師出具的診斷證明</strong></li><li><strong>醫師撰寫的說明信</strong>（說明病況、治療計畫、為何無法參加考試，協會會提供格式）</li></ol><p>這些資料需一併上傳至協會指定的表單中。提交後，協會會在 <strong>三天內由其合作的醫療顧問審核</strong>，並回覆是否核准延期。</p><p>在我的案例中，協會認為我的情況雖符合延期條件，但<strong>不屬於可免費延期的範疇</strong>，因此仍需支付 <strong>300 美元的改期費用</strong>。所幸當時手術有保險給付，才得以順利應對這筆支出。</p><h3>如果你真的遇到無法控制的身體或生活狀況，不用過度自責或焦慮，CFA 的制度仍有提供協助的空間。關鍵在於要準備好所需文件、並儘早申請。</h3><p>但如果你只是因為「擔心沒準備好」而考慮延期，請務必謹慎評估。畢竟延期不是萬靈丹，<strong>只有行動和努力，才能真正提升通過率。</strong></p><h4>準備過頭的情況</h4><p>這裡所說的「準備過頭」，並不是指過度焦慮地反覆讀書，而是<strong>一種理性地評估自己已達到一定水準，卻仍感到無所適從的狀態</strong>。這段內容特別適用於那些已經清楚自己強弱項，模擬考成績穩定，並且對整體內容掌握度不錯的考生。</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*nkOzu4ltpKMQOPBL6OsFEA.jpeg" /><figcaption>CFA 模擬考試分數</figcaption></figure><p>我自己在準備 CFA L2 的過程中，就遇到了這種情況。從 2023 年 8 月開始準備，一直到 2025 年 4 月，雖然中間因為手術、工作和過年，大約有四個月（2024/10 ～ 2025/2）沒有密集讀書，但整體準備時間已經相當充足。</p><p>到了 2025 年 5 月初，我開始出現一種「<strong>不知道接下來要做什麼</strong>」的感覺。這部分來自於兩個原因：</p><ol><li>設定的學習目標幾乎都已達成</li><li>長時間反覆閱讀同樣的內容，開始無法辨別自己到底是「真正理解並能延伸應用」，還是只是「記得答案」。</li></ol><p>這種狀況我過去也曾經歷過，當時面對的是如 PMP 等短期可彈性安排的證照考試。當準備到某個階段覺得差不多了，就可以立即報考。然而 CFA Program 屬於<strong>定期舉辦的長期考試</strong>，即使在狀態最佳時也無法立即應考，這反而讓人進入一種空轉的焦躁感。</p><p>我後來的做法是：<strong>暫時把注意力轉移到其他事物上，再回來看書</strong>。具體來說，我在 4 月底開始進行強度較高的復健運動，並加入聽音樂、放鬆冥想等活動。這些安排有助於我跳脫原本「工作 → 讀 CFA → 解模擬題 → 再工作」的循環節奏，給大腦一些不同的刺激，讓心態稍微重置。</p><p>需要特別提醒的是，這樣的調整<strong>不代表放棄節奏，而是適度變化，不要打亂整體備考進度</strong>。當狀態重新調整好，再回頭看書時，會更有餘裕也更能保持學習的敏感度。</p><h3>心得總結</h3><p>回頭看 L1 的準備過程，其實 L2 帶來的挑戰更大，但幸運的是，我順利走完整個 L2 課程，並在第一次考試就達成了目標，拿到了理想的結果。原本還有心理準備要重考，結果在 7/1 放榜那天，當下正在準備 <a href="https://www.youtube.com/watch?v=nv-FGf-aNdg">AI 社群的活動直播</a>，本來不想看成績，還是不小心打開 Gmail，看到「Congrats」的那一刻，真的非常開心也鬆了一口氣，還好心情不會影響到來分享的講師跟聽眾。</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*SbbbLA4fnXljWyFKhIczFg.png" /><figcaption>主持每個月 AI 社群活動，剛好講 AI Agent 議題</figcaption></figure><p>L2 的課程上我覺得最有趣的就是衍生性金融商品，可惜我拿到的是分數較低的。在 L2 中的公司財管跟股票評價、財報等等幾乎是更深入的探討。對我來說感覺在 L2，把更細節的內容更深入探討，讓我有著更細節的知識思考年報要怎麼出，才會有好的盈餘品質跟好的財報品質，給投資我司的股東閱覽，上課過程中也能感受到老師花了很多時間在講一些背景知識，讓我們可以有案例參考，加深印象。</p><blockquote><em>要怎麼寫出一份具備「盈餘品質」與「財報品質」兼具的年報，才能真正讓股東看到公司經營的價值。</em></blockquote><p>過程中也非常感謝授課老師們的用心，不僅釐清觀念，還補充了許多實務案例，幫助我們把抽象理論連結到現實世界的金融環境，加深了對財務世界的理解。</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c805c6fbc7dd" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[最近的 MCP 紅什麼？]]></title>
            <link>https://mybaseball52.medium.com/what-is-mcp-in-ai-bb09fab5b2ec?source=rss-4bcc562a9bb9------2</link>
            <guid isPermaLink="false">https://medium.com/p/bb09fab5b2ec</guid>
            <category><![CDATA[mcps]]></category>
            <category><![CDATA[llm-applications]]></category>
            <category><![CDATA[llm-function-calling]]></category>
            <category><![CDATA[claude]]></category>
            <category><![CDATA[ai]]></category>
            <dc:creator><![CDATA[(KJH) Kuan-Jung, Huang]]></dc:creator>
            <pubDate>Fri, 14 Mar 2025 06:23:28 GMT</pubDate>
            <atom:updated>2025-04-02T06:38:08.119Z</atom:updated>
            <content:encoded><![CDATA[<h4>讓 AI 模型輕鬆對接外部資料</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*GJJC0B7qsnAg5VqZ6GCuvA.png" /><figcaption>Source: <a href="https://github.com/modelcontextprotocol">https://github.com/modelcontextprotocol</a></figcaption></figure><h3>MCP 簡介</h3><p>這是 Anthropic Claude 推出的一套新通訊協定，全名是 <strong>Model Context Protocol</strong>。它的目的是定義一個標準，讓負責執行大型語言模型（LLM）的程式（稱為 MCP host）能夠和提供特定功能的外部服務（稱為 MCP server）進行通訊。這樣的設計能讓 LLM 獲得延伸感知與行動的能力，同時讓 MCP server 不需要了解 LLM 的運作，就能提供自然語言驅動的操作介面。</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*S0dII5FO0H48sD1KeXkhFw.png" /><figcaption>Source: <a href="https://modelcontextprotocol.io/introduction">https://modelcontextprotocol.io/introduction</a></figcaption></figure><p>舉例來說，假設某知名電商平台支援 MCP 協定並成為一個 MCP server，那你就可以在一些有 MCP host 功能的應用程式上，直接用自然語言查詢商品、追蹤訂單或加入購物車。又或者，如果該電商網站本身沒有支援 MCP，你也可以自行撰寫一個 MCP server，它透過該網站的 API 來實作這些操作，仍然可以達成相同的效果。</p><h3>跟 OpenAI 之前的 function calling 差異在哪？</h3><p>同樣地，如果你想在自己含有 LLM 的應用程式中提供這些電商功能，只要呼叫使用這個 MCP server 就行了，不需要重新整合 API 或寫新的 function。這與傳統的 function calling 不同，function calling 通常需要把特定函式直接綁進你的程式，這樣的耦合度會很高，變成要一個功能就要寫對應的一個程式。而 MCP 目前看起來的好處就是可以解耦，MCP host 和 MCP server 可以獨立存在，只要雙方都遵守定義好的 protocol 就能串接使用。你寫的 MCP server，其他 MCP host 都能用；你寫的 MCP host，也能輕鬆擴展去使用更多不同的 MCP server 上的程式。</p><p>如果換個場景，像是銀行服務支援 MCP，那你也可以用自然語言查詢帳戶餘額、轉帳、或設定金融提醒，而不必進入繁瑣的網銀流程。這讓 LLM 的應用可以真正介接到日常生活的實際服務。</p><h3><strong>哪些廠商採用 MCP?</strong></h3><p>目前 MCP 協定仍以 Anthropic 為主力推動，其他 LLM 廠商尚未說明是否跟進。但一旦這套標準被廣泛採納，整個 AI 生態系將能實現模組化、可組合、自然語言驅動的服務體驗，讓更多人更方便地進入 AI 世界。</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=bb09fab5b2ec" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[敏捷開發還是敏捷自助餐？]]></title>
            <link>https://mybaseball52.medium.com/agile-is-not-buffet-796cf4e18932?source=rss-4bcc562a9bb9------2</link>
            <guid isPermaLink="false">https://medium.com/p/796cf4e18932</guid>
            <category><![CDATA[mygo]]></category>
            <category><![CDATA[myscrum]]></category>
            <category><![CDATA[screamguide]]></category>
            <category><![CDATA[agile]]></category>
            <category><![CDATA[scrum]]></category>
            <dc:creator><![CDATA[(KJH) Kuan-Jung, Huang]]></dc:creator>
            <pubDate>Sun, 26 Jan 2025 17:57:35 GMT</pubDate>
            <atom:updated>2025-01-26T17:57:35.276Z</atom:updated>
            <content:encoded><![CDATA[<h4>只想挑幾道菜，卻忘了主菜是什麼</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/1*w5YNtcaRenSsmvJr2xEhAw.png" /><figcaption>2025 年，敏捷還在 GO，大概。《圖源：MyGO!!!!!》</figcaption></figure><h3>前言</h3><p>敏捷開發常被視為現代軟體開發的黃金標準，強調快速迭代、靈活調整和高度適應性。然而，在實際運作中，有些團隊的敏捷實踐卻成了「buffet-style agile」（自助餐式敏捷）：挑自己喜歡的吃，卻忽略了整體的平衡與意圖。</p><p>敏捷開發的原則清晰地寫在《敏捷宣言》中，四個價值觀和十二條原則明確指出了敏捷的核心精神：重視人與互動、專注於交付有價值的產品、擁抱變化，以及與客戶緊密合作。然而，這些核心理念在某些團隊或個人手中，卻被變形成了「取用方便的開發菜單」。</p><h3>敏捷=混亂？錯誤的敏捷實踐如何拖垮團隊</h3><ol><li><strong>Stand-up變成主管盯進度</strong><br>敏捷開發中的 Stand-up 會議，本來是團隊成員之間的一種快速同步工具，用來分享進度、解決障礙，並確保團隊朝著共同的目標前進。然而，在許多公司和團隊中，Stand-up 卻逐漸變質，變成了主管每天盯進度的「例行公事」。這樣的變化，不僅違背了敏捷的核心精神，長遠對團隊士氣和效率造成損害。</li><li><strong>過度追求變化，忘了穩定性</strong><br>敏捷提倡對需求變更的彈性，但不代表有隨意變更需求的情況，這種結果往往是：沒有真正完成的功能，只有永遠在修改的半成品。特別是 Product Owner 成了不斷「塞需求」的角色</li><li><strong>Scrum 只剩框架，沒有靈魂</strong><br>有些團隊打著敏捷的旗號，卻只是敷衍地遵循 Scrum 的形式：有衝刺、有燃盡圖，但沒有效能提升或團隊合作的實質變化。</li></ol><h3>敏捷不是要挑最方便的，而是要挑最適合的</h3><p>敏捷不是讓每個人都拿著自己的盤子，想夾什麼就夾什麼，而是要從整體目標出發，根據團隊的實際情況選擇合適的工具和流程。如果團隊的核心價值跟共同目標都被拋到腦後，敏捷反而會造成混亂。</p><p>我們最終的目的依舊在談論怎麼交付價值，而不是追求快速；過程中互相了解，提升團隊協作。缺乏整體規劃的敏捷實踐，會讓團隊成員迷失方向，最終導致開發速度雖然快，但結果毫無價值。</p><p>敏捷中每一道程序跟儀式都有其意圖，不能只是挑最省力，或是變相的滿足私慾的做法，卻忽視了整體效益。</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=796cf4e18932" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A Half-Point Victory: Analyzing the Iconic Go Match from Hikaru no Go]]></title>
            <link>https://mybaseball52.medium.com/a-half-point-victory-analyzing-the-iconic-go-match-from-hikaru-no-go-977f8818a7c6?source=rss-4bcc562a9bb9------2</link>
            <guid isPermaLink="false">https://medium.com/p/977f8818a7c6</guid>
            <category><![CDATA[go]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[gemini]]></category>
            <category><![CDATA[chatgpt]]></category>
            <category><![CDATA[claude]]></category>
            <dc:creator><![CDATA[(KJH) Kuan-Jung, Huang]]></dc:creator>
            <pubDate>Sun, 22 Sep 2024 13:24:33 GMT</pubDate>
            <atom:updated>2024-09-22T13:25:15.583Z</atom:updated>
            <content:encoded><![CDATA[<h4>Exploring the Dramatic Game Between Sai and Toya Meijin Through AI-Powered Analysis</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*TaK_gHFiG792pUM3qYt33g.png" /><figcaption>Scene of Toya Koyo resign</figcaption></figure><p><strong>Hikaru no Go</strong> holds nostalgic value for many of my peers, and I’ve often seen posts on some game or animate forums where people mention that their introduction to Go came from this series.</p><p>One of the most iconic matches in the series is from Episodes 55 and 56, where Sai (playing white) faces off against Toya Meijin (playing black). This match marks the climax of the entire series, with two Go masters battling it out, ending in a half-point difference.</p><p>To this day, I still can’t fully grasp the intricacies of that game. I can only rely on the background music to figure out who won or lost. 😂</p><p>That made me think of my strongest AI friend: ChatGPT. I used GPT-4 to analyze the match, and it turned out to be quite interesting. Since GPT-4 can recognize images, I uploaded the game board. Based on the analysis, the result was the same as in the anime: white wins by half a point. Unfortunately, I can’t share a link directly since I uploaded an image, but here’s a screenshot of the result.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*b49eAnaQ8p3xCpK_U8tM6Q.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hX51RBfqiYoq2Weh7z9sPg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Nntq7Cqhrv53QTB1wpiDTg.png" /></figure><p>With GPT-4o (a non-image-recognizing version), I had to manually recreate the game for analysis. Interestingly, the logic and reasoning felt a bit off, and it concluded with white winning by a whopping 18.5 points: <a href="https://chatgpt.com/share/66f01480-6464-8007-97dc-815406590a91">Link to the result</a>.</p><p>I posed the same question to Google Gemini, but it only told me, “Go scoring is a task that requires patience and attention to detail. A deeper understanding of Go rules and scoring methods will help you analyze games more accurately.” It felt like a lazy answer that kept going in circles. 😂<br><a href="https://g.co/gemini/share/877647aa1303">Link to Gemini result</a>.</p><p>When I asked Claude the same question, it took multiple attempts before it finally gave me the correct answer: white wins by half a point. The first two tries said black won by half a point. Unfortunately, Claude doesn’t provide shareable links, so I’ve attached a screenshot.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1kao-9vM0CB6e8Rbl8J8jA.png" /></figure><p>In the image, it said:</p><p>Let’s analyze this Go game:</p><p><strong>Territory Calculation:</strong></p><ul><li>White stones: approximately 82 points</li><li>Black stones: approximately 87 points</li></ul><p><strong>Komi Rules:</strong></p><ul><li>You mentioned that black has to give 5.5 points as komi.</li></ul><p><strong>Result Calculation:</strong></p><ul><li>Black’s actual score = 87–5.5 = 81.5 points</li><li>White’s score = 82 points</li></ul><p><strong>Outcome:</strong></p><ul><li>White 82 points &gt; Black 81.5 points</li></ul><p>Thus, with black giving 5.5 points of komi, white wins by a narrow margin of 0.5 points.</p><p>So for the Claude, it tries 3 times to get correct answer.</p><p>As for Perplexity, I tried using the math mode for analysis. Since its CoT (Chain of Thought) feature doesn’t support Chinese, I tried using English on my second attempt. Unfortunately, the calculations were way off, with wildly different results:<br><a href="https://www.perplexity.ai/search/00000000000-10000-110-0000-1-1-mSNvuNAeQ0G1cgRt1_HO0g">Link to Perplexity result</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=977f8818a7c6" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[利用 LM Studio 跟 AnythingLLM 實作 RAG 系統]]></title>
            <link>https://mybaseball52.medium.com/using-anythingllm-and-lm-studio-to-do-rag-79a962095da1?source=rss-4bcc562a9bb9------2</link>
            <guid isPermaLink="false">https://medium.com/p/79a962095da1</guid>
            <category><![CDATA[ai-tools]]></category>
            <category><![CDATA[retrieval-augmented-gen]]></category>
            <category><![CDATA[anythingllm]]></category>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[lm-studio]]></category>
            <dc:creator><![CDATA[(KJH) Kuan-Jung, Huang]]></dc:creator>
            <pubDate>Fri, 14 Jun 2024 14:41:31 GMT</pubDate>
            <atom:updated>2025-08-01T12:00:35.436Z</atom:updated>
            <content:encoded><![CDATA[<h4>兩套工具打出個人化 AI 模型的組合拳</h4><h3>前言</h3><p>很常有人問我怎麼訓練自己的 AI 模型，但是又不懂怎麼訓練模型。或是不想要把自己公司的資料放在雲端上而苦惱著。</p><p>所以這次我想介紹兩個很簡單的工具，LM Studio 和 Anything LLM 設置檢索增強生成 (RAG) 系統。</p><p>在 2025 年 8 月 1 日發現 LM studio 有更新 UI，文章會以新版的 UI 為主進行安裝教學。新版的 LM studio 本身就有對話的功能了，所以安裝 AnythingLLM 變成可選的選項。</p><h3>什麼是 RAG（檢索增強生成）？</h3><p>檢索增強生成（RAG Retrieval Augmented Generation）是一種結合了檢索技術和生成模型的方法。它的主要目的是使用者提供的相關資訊，透過一些技術來增強內容的輸出。讓我們在檢索資訊的過程中，提高生成內容的準確性和提問的相關性。</p><p>在檢索的過程中，系統會從預定的檔案文件、資料庫或網路資源中檢索與使用者查詢相關的資訊。這些資訊可以是文本段落、文件或網頁內容等。例如我想查詢：3/15 日有什麼行程，一個理想的 AI 模型就會去檢索我的 Google 日曆，或是某些電子信箱，幫我整理出有 3/15 相關的內容。我們會透過 Embedding，也就是向量化的方式來表示文本，目前最簡單的向量化是將文本資料轉換為以數字表達向量距離的過程。我們就可以透過向量距離（相關性）來回傳相似度最高的答案。</p><p>完成了檢索階段，我們就進入到內容生成階段，模型會利用這些前面檢索到的內容，來生成更為精確和有意義的回答或文本。</p><p>這樣的過程，我們不直接調整模型參數取得資訊的方式，我們就稱之為檢索增強生成。</p><h3>設定與安裝環境</h3><p>接下來的示範我是以 Macbook Pro M1 為例，我確認過各種作業系統的設定大同小異。</p><h4><a href="https://lmstudio.ai/">安裝 LM studio</a></h4><p>使用 LM Studio 的好處在於，它可以在個人電腦上使用下載的 LLM 模型，確保隱私和資料控制。這些 LLM 模型所檢索的資料不用上傳到網際網路就可以使用。</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*jh5KUHnhsl5dQB8nXtwmYw.png" /></figure><p>在新版的 LM studio UI，如果非工程師，可以選擇 User，然後安裝想要使用的模型。</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*8fTJqekdk2hWHbh9WPYY1A.png" /></figure><p>在畫面的正上方，可以選擇模型，例如 Gemma or Deepseek 等等</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*S-isBePjnYWK4fhVgJn40A.png" /></figure><p>挑好模型後下載模型。</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VcTArYkKnJg5yWP1n8mVCQ.png" /></figure><p>下載完成後，就可以在正上方看到 UI 可以選擇模型的選項：</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1LM9pgghHzCJXhiPd1fGHA.png" /></figure><p>接著側邊欄可以開啟對話模式</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rV2wph1BAiOfmEsp6i40MA.png" /></figure><p>執行結果</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*joCZKxePZkOgH2_4zpXy4Q.png" /></figure><h4>設定 LM Studio Developer 模式</h4><p>如果你想要把 LM studio 變成類似 AI server 的功能，步驟也很輕鬆</p><p>首先在最左下方的 mode 改為 Developer</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*a75tg3c9kGzJ-AR3ZVpMgg.png" /></figure><p>左側邊會出現 4 個 icon，選擇第二個。接著在上方把 toggle 改為啟動， local server 就建置完畢了！</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*j2guG5v0Hn5EJ-RQbh2Khw.png" /></figure><p>接著安裝 <a href="https://useanything.com/download">AnythingLLM</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*0AR7au1weYgenRQMWpAGMw.png" /></figure><h3>透過 AnythingLLM 達成 RAG</h3><p>完成 AnythingLLM 的設定之後，你會看到這樣個畫面：</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*TAZLiDUo0JQoxVkHQXNwgg.png" /></figure><h4>上傳檔案</h4><p>上傳檔案這個功能，可以讓你把你想要 Embedding 的文本讓 LLM 來做資訊檢索。點擊畫面中的 Upload a document 就可以開始上傳檔案了。</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*NeSDDLwiLRT50LQrqp4XDQ.png" /></figure><h4>問問題</h4><p>完成檔案上傳之後，就可以開始問問題了。當然準確度可能就需要由你做最後決定，如果表現不如預期，可能還是需要整理一下文件內容，或是換個模型，例如從 7B 換成 13B 的模型之類的。</p><h3>小記</h3><p>通過 AnythingLLM 來做 RAG 除了免除寫程式的挑戰之外，即使不具有機器學習專業知識的使用者，也能在自己的電腦中，透過這兩個工具來體會到 RAG 的威力，以及客製化自己的 AI 模型。而且運用這樣可以在本機執行的 LLM 工具，也確保了資料隱私和資料可以由自己控制跟提供。從而生成更準確、資訊更豐富且最佳的輸出。</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=79a962095da1" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI Users Community 2023 致社員信]]></title>
            <link>https://mybaseball52.medium.com/2023-retro-letter-to-ai-users-community-members-ab43cc098422?source=rss-4bcc562a9bb9------2</link>
            <guid isPermaLink="false">https://medium.com/p/ab43cc098422</guid>
            <category><![CDATA[community]]></category>
            <category><![CDATA[yearly-review]]></category>
            <category><![CDATA[management]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[engagement]]></category>
            <dc:creator><![CDATA[(KJH) Kuan-Jung, Huang]]></dc:creator>
            <pubDate>Fri, 29 Dec 2023 13:10:42 GMT</pubDate>
            <atom:updated>2023-12-29T13:12:04.403Z</atom:updated>
            <content:encoded><![CDATA[<h4>Kevin 個人 2023 的 AI 社群營運報告與反思</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*v6Ay4lpqoVvMa00KCMbYZg.png" /></figure><h3>致 AI 社群的所有朋友：</h3><p>人們在過往普遍相信年輕時多打拼，期望能按著過往傳統的路線升遷或是成長，並期望在一定時間能有一定的名聲或成就。這個理論是過往百年或千年所流傳的。並且這些過程也會「繼承給後代」。</p><p>但在今年 ChatGPT 出來之後，人們似乎又覺得不一樣了，在元宇宙的虛實整合的話題性，突然又回到「 AI 可能要取代工作了，我的生活要一團糟了」的想法。在彼時各種牛鬼蛇神出來，準備在蕭條的寒冬出來散播恐懼。我個人不是很喜歡這樣的情況，但就是因為沒有出來講，所以才讓這些人出來。決定探索新技術的旅程同時以自身專業回饋社會，並且獲得一定的成就。</p><h3>我還有優勢嗎？</h3><p>一筆對自己有幫助的投資，很多時間是考慮這筆投資在能承受一定風險下可以有多少超額報酬？而在面對再次讓人們陷入焦慮的 AI 新技術中，同樣的概念也是適用的。一年下來人們從「 AI 可能要取代工作了」，透過各種學習與探索，逐漸理解到「 AI 不會取代工作，但會用 AI 的人一定會」。這些在 2023 年就開始想方設法增強自身能力的人，可以預見是後面取得超額報酬的人。</p><h3>當知識可以用複利思考</h3><p>如果是社團創立以來就加入的朋友，一定看過我寫的<a href="https://www.facebook.com/groups/aigcapplied/posts/570527625202075/">這兩段話</a>：</p><p>* Kevin 認為並沒有 100% 可以得到 ROI 的東西，然而最後一點一定 100% 會有投資報酬</p><p>* 學習，學習，學習。透過學習變得更聰明，變成某個領域專家，驅動 AI 進行任何事物才是好的 AI assitant intellgent。這是我(Kevin) 從 2018 大學畢業以來持續至今的信念。</p><p>這段話在前面一個段落剛好應驗了「 AI 不會取代工作，但會用 AI 的人一定會」。多花點時間想一下自己的工作有什麼地方要調整，或是可以用 AI 協助完成。</p><h3>當生產力可以用複利思考</h3><p>一天多完成一個任務可能沒有多大的感覺，但是在操作 AI 工作的時候，可能一天完成 100 件事情不一定是紙上談兵或天方夜譚。如果你將這件事持續個 5 年將會是多可觀的產出？另外一件事情就是持續做，做久了就會越快。一個好的 prompt 寫作者花費時間讓 LLM 不胡說八道，短期也讓自己的生產力大幅提升。當生產力提升之後，就有更多的空閒時間可以提升自己或做更好的規劃。</p><h3>當人際關係因前面兩個而複利成長</h3><p>來我們社群的朋友，可能是來學習怎麼用 AI，或是來認識新朋友，也有來找尋商機的朋友，在社群的互動過程中，新的橫向或是縱向關係被建立，是一個很有趣的事情，沒有人知道下一個認識的人會幫助你達到什麼樣的新的高度。</p><h3>我們社群的工作</h3><p>我們建立這個社群的目的在於促進使用者之間的互動和交流，讓使用者可以分享問題、交換知識和建立關係。我們相信你已經意識到，人工智慧是一個充滿潛力和機遇的領域，我們也相信你已經開始了解 AI 技術以及其對未來的影響。在經過一年的社團密集籌辦活動之後，我們除了向大家爭取到不少講師的時間以及錄影留存知識之外，更多的是開放不同的知識探索家在 Facebook 上分享他們自己對於 AI 的洞察，或是知識分享。</p><p>在 2023 年中，我們的 FB 約有 340,000 人次的瀏覽量，社團短短 9 個月的時間，從零成長到 1.7 萬人，當然相較於同性質的社團來說我們的社團人數很少，但是我們從問卷中發現：</p><p>每場演講中，平均聽分享的11% 的企業主，38% 的中高階主管 31% 的專業技術人員，對於社群想要認識企業端人脈或者，對於講師來說都是很棒的表現或交流。</p><p>在 Youtube 經營方面，我們有 23K 的總瀏覽量，以及 1440 位訂閱者。令我們感到開心的事情是我們並非娛樂頻道，內容也是非常困難的，居然總計有 3200 小時的總時數，也說明社群的朋友人對於議題是真心關注的。而且在直播後都還有將近 200 次以上的回訪量，且這些回訪量約有 10% 的人是從頭看到尾的。</p><p>我們也持續開始探索更多合作的可能與長期關係，這包含台灣的雲端業者們，以及技術專家們。</p><h3>我們的幹部</h3><p>我想，以一個完全沒有任何機構所援助的社群能有將近 1.7 萬人，這個經營成果是令人驚艷的。同時我們在跟敏捷社群合辦的 Line 社群雖然從 5 千人降至 4.8 K 的人數。這樣的成功肯定是一個能幹、聰明並賣力的團隊所組成的。身為這樣團隊其中一份子，我感到非常驕傲。</p><p>當然經營一個社群是真的很不容易。在一個經營績效良好的企業中你的團隊可以是能幹、聰明並賣力的的三種其中兩種。並且以高報酬獎勵之，但是在我們社群，我們並需三者皆有才有辦法。</p><p>這些源自於我們除了對於成為一個卓越的知識社群的追求以外，我們更致力於打造一個重要，有價值的東西給我們每一位參與社群的朋友們。</p><p>如果我自己本身打造的<a href="http://metablox.co/"> metablox.co</a>，它是一個保存回憶，人生中最珍貴價值的平台。進一步地，希望打造出來的社群，是我們可以在後面很驕傲地向後人所述說的「想當年啊…」</p><p>當然這樣的工作很不容易，一年「大咖講座（感謝明遠副社長的命名）」與「QA 時間的場次」合計下來剛好 71 場次，沒有這樣的恆毅力是達不到的。我要特別感謝 Darren 與 Jimmy，沒有他們兩個我想 71 場次這件事是完全做不到的。也要感謝各位海報組的志工持續製作精緻的海報資訊。</p><p>這種社群經營的過程讓人看到更真實的人性。在幹部們孜孜不倦的讓社群茁壯的同時，我們社團也受到很多前輩的支持，投入寶貴時間來分享一場又一場的演講，直到今天我們社團內容積累起來，也開始對一些來學 AI 的人有影響力。</p><h3>我們打算做的</h3><p>AI 終究有其限制，AIGC 不可能完全不需要人類介入，一定是 IA intellgent assitant，這個概念在我創立社團並對外公開演講的就說明了，這是借用 《謝謝你遲到了》 ㄧ書作者在第九章的標題概念。沒想到幾天後微軟把他的 AI 整合的生態系產品命名為 copilot。</p><p>我們來社團的每個人都是一起預見未來 AI 是我們生活中的一部分，而我們幹部則是往落實 AI 真的是我們生活中的一部分而努力進行宣揚與教育。再放到更大的，很多台灣的菁英份子也在這個浪潮中努力做出貢獻，讓大家能更快地適應並融入新科技的發展。</p><h3>2024 年的目標</h3><p>我們一樣處於面對環境的高度變動，我們致力於了解新的技術與新的應用，讓社群朋友持續取得最新的資訊以及思維。我們的目標依舊是持續舉辦活動之外，同時讓大家更理解怎樣用 AI，這需要大量時間的投入以及投資。</p><p>在 2023 年的簡易調查中，居然還是有高達 60% 的企業停留在評估或測試 AI，還有 25% 的企業甚至尚未開始使用 AI。算下來平均只有 15% 的企業正式使用 AI 在日常工作中。所以本社群的第一筆最大的投資在於建置論壇，這樣的論壇有顯著的機會可以改善企業對於資訊搜尋的便利性以及對於工具的比較性，從中帶來的些許收入支撐社群未來可能所需的經費。同時有關注到女性的互動偏少，這塊我也會持續加強，讓女性願意參加，或是提供一些幫助讓女性也能參與到 AI 的話題。再來是議題的多元性，我們在 2024 也會關注 HR, ESG, DevOps 跟資安等等企業持續面臨的威脅與挑戰。</p><p>2023 年是充滿各種挑戰與驚喜的一年，我們由衷感謝各位帶給我們的鼓勵、資訊與信賴，感謝所有團隊成員的努力，感謝家人的支持。</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ab43cc098422" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[為什麼使用 User story]]></title>
            <link>https://mybaseball52.medium.com/why-user-story-2e6349a9cf5d?source=rss-4bcc562a9bb9------2</link>
            <guid isPermaLink="false">https://medium.com/p/2e6349a9cf5d</guid>
            <category><![CDATA[scrum]]></category>
            <category><![CDATA[product]]></category>
            <category><![CDATA[agile]]></category>
            <category><![CDATA[user-stories]]></category>
            <category><![CDATA[product-management]]></category>
            <dc:creator><![CDATA[(KJH) Kuan-Jung, Huang]]></dc:creator>
            <pubDate>Wed, 27 Dec 2023 13:55:55 GMT</pubDate>
            <atom:updated>2023-12-27T13:55:55.374Z</atom:updated>
            <content:encoded><![CDATA[<h4>A story is a placeholder for a conversation.</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*PVnVjUIwMQHNn79skMX_2Q.png" /><figcaption>Image powered by DALLE</figcaption></figure><h3>User Story 的概念</h3><p>過往我們對於提需求這件事存在著一個固定的ㄧ想法：先把需求規格定好，我們就照這個規格做好軟體給使用者。當然想也知道通常規格一定是一直改來改去的，所以有很多有趣的文章都在說明規格跟工程師之間的戰爭。</p><p>User Story 這一概念的出現就是希望解決前面的問題，User Story 不是一個刻板的、詳細的需求說明，而是一個高層次、靈活的描述，用以引導後續的開發工作。所以 User Story 的目的就是希望促進互動，並延遲到必要時才提供細節。</p><p>User Story 通過簡潔的敘述方式，鼓勵團隊成員之間的對話和討論。這種描述方式專注於用戶需求和期望的「什麼」和「為什麼」，而非具體的實現方式或技術細節。這種高層次的描述使得非技術人員也能參與到討論中，從而確保每個人對產品目標和功能有一致的理解。另外因為還不是很具體的敘述內容，可以引發更多人的討論跟對於 User story 內容的理解，確保在這個 User story 討論好以後是團隊的每個人皆有共識。</p><p>User Story 的另一個關鍵特點是其對細節的延遲提供。在初期階段，過於詳細的技術或設計規格可能限制思維，並可能在未來的開發過程中導致不必要的更改。隨著產品的進展和獲取到更多的資訊時，團隊可以逐步完善和細化需求。</p><h3>有效的 User Story</h3><p>User Story 通常遵循一個基本的格式：“As a [Role], I want [Feature], so that [Benefit].” 這裡的 [Role] 代表著一個特定的使用者群體而非個別使用者，這有助於從這個群體的特性出發，考慮其在系統中的利益。相對於 [Role]，我們還有 [Persona]，這是一個更加具體化的角色，代表個別的使用者，通常會有一個背景故事來描述這個角色的特點和需求。這些角色的描述幫助開發團隊理解和同理使用者，從而設計出更加符合使用者需求的功能。</p><p>有效的 User Story 應該是獨立的、可協商的、有價值的、可估計的、小而精的、並且可測試的（INVEST）。例如，一個過於龐大的 User Story 可能是：“As a user, I want a video app that lets me play, download, share, and create playlists of videos。” 這樣的故事過於廣泛，難以管理和實施。相反，它應該被分解成更小、更具體的故事，例如分開處理播放和下載功能。</p><p>在 User Story 的開發過程中，我們還需要考慮如何切割故事。橫向切割，即按照專業領域（如 UI, server side, DB）來分割故事，這種方法雖然在技術上可行，但有可能無法交付完整的功能，比如只有 UI 但是什麼功能都沒有，難以帶來真正的價值。相反的，所謂功能的縱向切割，即按照物件的功能來切割，我們交付較小的功能片段，使用者真正能使用這個功能，並使這個故事更容易被用來測試。</p><h3>User story 的階層關係</h3><p>User Story 還應該包括「Epic」和「Theme」。Epic 是一系列 User Story 的集合，用於描述達成特定目標的完整工作流程。而 Theme 則是追蹤一系列相關的故事，但這些故事可以獨立完成。例如，在我們的學習平台軟體中，一個 Epic 可能是關於建立和管理影片學習清單等一系列 User Story，而一個 Theme 則可能是關於提高使用者體驗的不同故事。</p><h3>User story 的驗收</h3><p>我們需要為制定的功能考慮「驗收標準（Acceptance Criteria, AC）」和「完成標準（Definition of Done, DoD）」。驗收標準為開發團隊提供具體的實作細節，幫助團隊明確理解何時 User Story 被視為完成。例如，在我們的前面的學習平台軟體產品中，驗收標準可能是：Given the user has selected a video, When they press the play button, Then the video should start playing without undue delay.。 DoD 確保產品符合品質標準，像是達成一開始團隊共同協商的測試覆蓋率、Code Reivew、部署和滿足 User story 驗收標準等等。</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2e6349a9cf5d" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[到底從事軟體開發的學生缺乏了甚麼？]]></title>
            <link>https://mybaseball52.medium.com/does-algorithms-and-datastructures-matters-in-software-development-careers-ff54cb684926?source=rss-4bcc562a9bb9------2</link>
            <guid isPermaLink="false">https://medium.com/p/ff54cb684926</guid>
            <category><![CDATA[skills-development]]></category>
            <category><![CDATA[datastrucutre]]></category>
            <category><![CDATA[software-development]]></category>
            <category><![CDATA[algorithms]]></category>
            <dc:creator><![CDATA[(KJH) Kuan-Jung, Huang]]></dc:creator>
            <pubDate>Wed, 27 Dec 2023 10:58:02 GMT</pubDate>
            <atom:updated>2023-12-27T10:58:47.342Z</atom:updated>
            <content:encoded><![CDATA[<h4>資料結構與演算法篇</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*uYyRI70ddBqOeROfkh5gcQ.png" /><figcaption>Source: <a href="https://pbs.twimg.com/media/Bv5xTwpIMAAx9hu.png">https://pbs.twimg.com/media/Bv5xTwpIMAAx9hu.png</a></figcaption></figure><h3>前言</h3><p>在 Scrum community 有篇很熱門的貼文討論，到底軟體業需要什麼樣才能的人才，議論紛紛，很多人指出軟體產業需要有溝通力的工程師；其次是熟悉資料結構和演算法的人才；第三多的票數則投給了能接受專案不斷變動，學校教育夠紮實的能力。</p><p>身為一個還在就學的學生，當然會覺得這個東西很特別。因為未來我就業，大家看的居然是這些能力！之前我已經提過溝通的重要性了，這次我試著來解釋為甚麼資料結構與演算法很重要！</p><h3>為甚麼資料結構與演算法極度重要？</h3><h3>甚至軟體工程師面試都直接考這些東西？</h3><p>可能有些讀者看過一則新聞：</p><blockquote>Max Howell在 OS X/iOS 開發領域是一位相當出名的軟體工程師，過去曾開發過網路電臺 Last.fm 的客戶端App、Twitter 的應用程式 TweetDeck 的開發者之一。他在 google 面試時因為反轉二元樹(是一個非常基本的資料結構)的解法不符合面試者期待因而落選。</blockquote><p>所以那時資料結構與演算法又再度成為了熱門話題。當然身為一個資訊科學系的學生，我們最重要的工作就是處理資料，有玩過解題系統或網站的人應該會知道，我們基本上都利用三個階段來處理資料：</p><p>1. 輸入要處理的資料 2. 運算這些資料 3. 輸出我們預期的結果</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*OmLtMy0P6aBQzfzbZewlAA.png" /><figcaption>來源：<a href="https://upload.wikimedia.org/wikipedia/commons/thumb/3/3b/Function_machine2.svg/1200px-Function_machine2.svg.png">https://upload.wikimedia.org/wikipedia/commons/thumb/3/3b/Function_machine2.svg/1200px-Function_machine2.svg.png</a></figcaption></figure><p>這一系列處理資料的工作其實一直充斥在人們日常生活之中，例如你遇到問題，你會想很多步驟去解決問題，電腦也是，例如我們在臉書上輸入一些文字，理所當然的，我們預期這些文字會顯示在我們的個人頁上。但這個過程並不是那麼理所當然，因為背後還有網站在處理我們所輸入的資料，然後轉換成網頁該顯示的結果，也就是我們所輸入的文字。這個過程我們一般稱為「運算」。</p><p>我們希望讓這些「運算」能夠更快，所以我們不斷優化從輸入到產出之間處理步驟，就像你每天都要做很多事情，我們總會希望減少過程中不必要的麻煩來加快速度。</p><p>所以我們採取「資料結構」設計以及「演算法」設計，正是為了讓電腦能夠更快地處理完資料顯示結果，優化我們從輸入到輸出的過程。通常我們可以優化的是運算階段﹝也稱做第二階段﹞。</p><h3>到底甚麼是資料結構？</h3><p>資料結構對應到我們如何在電腦上呈現我們組織資訊的形式，讓電腦可以有效地使用這些經過整理的資料。假設我們要儲存一組聯絡人的資訊，我們會去想像要如何儲存：是一筆聯絡人紀錄就儲存全部的資料，包括姓名、性別、電話號碼；還是分門別類儲存，像是姓名一組、性別一組、；電話號碼一組等等。當然後面所述的方法，找這個人的全部資料會浪費時間在資料重組上面。所以回來仔細想一下，其實你會發現使用不同的資料結構會有截然不同的效能。</p><p>舉例來說，我們會希望有的資料結構可以展現資訊該有的樣子，意思是說讓先進來的資料可以先進行運算。又或者是讓資料經過整理後像一張心智圖一樣排序，如下圖</p><h3>資料結構的重要性</h3><p>1. 每個程式都會需要用到資料結構，因為你會需要整理輸入的資料</p><p>2. 有些特別的資料結構是組成重要演算法的基礎物件，也因為輸入的資料經過開發者的設計，在呈現上，讓使用者管理大量資料時，會變得非常簡單與有效率。</p><p>3. 有些程式語言很強調資料結構，甚至當成是軟體設計的基石，這時反而不是把演算法當作設計基礎進行開發。</p><h3>到底甚麼是演算法？</h3><p>有人說，演算法是工程師選擇不解釋的一種說詞。</p><p>但實際上，演算法是思考有效能地問題解決方式。</p><p>舉例來說， 1 + 2 + 3 + 4 + … + n，有些人用迴圈去做計算，只需要 O(n) 的時間（甚麼是 O 的時間 http://ppt.cc/kFULt），看起來非常地好，簡單易懂。</p><p>但是為何不用高斯公式 ((1+n)*n)/2，這個只要 O(1) 的時間就可以處理完畢了，更快，電腦不必一個一個去加總。</p><p>這樣差在哪裡？我們要先知道電腦是一個指令一個動作，如果你用迴圈來進行計算，對電腦來說我要每次提取一個數字，相加之後再存回記憶體中，要如此進行 n 次。但如果用高斯公式的話，電腦只要先讀取 1 ，與 n 相加後，再跟 n 相乘，接著除以 2 ，就出現結果了，這樣的步驟快上很多。</p><p>甚至，你還要考慮有些程式語言只支援遞迴（就是各位高中學的遞迴），沒有所謂的迴圈運算，這時候你必須用一個叫做 reduce 的作法或者使用 tail recursion 模擬迴圈去解決問題。演算法的重要性在這邊顯示出來。</p><h3>演算法的重要性</h3><p>為了讓各位體會演算法的重要性，我將用這兩個程式來做示範：</p><p>遞迴處理費伯納西數列：</p><pre>def fibonacci_recursive(n):<br>    if n &lt;= 1:<br>        return n<br>    else:<br>        return fibonacci_recursive(n-1) + fibonacci_recursive(n-2)</pre><p>利用帶有快取的資料結構來運算費伯納西數列：</p><pre>def fibonacci_with_cache(n, cache={0: 0, 1: 1}):<br>    if n not in cache:<br>        cache[n] = fibonacci_with_cache(n-1, cache) + fibonacci_with_cache(n-2, cache)<br>    return cache[n]</pre><p>詳細原理是因為我觀察到如果利用遞迴處理費伯納西數列，會一直不斷出現 return 1 的情況，因為遞迴並不會記錄運算的結果，算完就「忘記」這些資料。</p><p>所以為了解決這個情況，我就利用一個資料結構把運算結果做儲存，等到我需要用到的時候就可以直接取得資料，而不是一直重複運算一些已知的結果</p><h3>我知道他們很重要，但是我還是無法體會</h3><h3>增加硬體不也可以加快運算速度嗎？</h3><p>有個行銷語句叫做：用免費的方式獲得更好的運算效能。這句話聽起來像是非常差勁的行銷廣告一樣，但是一旦這句話應用在資料結構與演算法上面就會變得非常地契合。</p><p>只要動腦筋改改資料結構或者演算法，就不用購置硬體設備增強處理效能！</p><p>許多人一旦電腦程式運算效能變慢，就會一直購買新世代的硬體來增強效能。</p><p>他們常常認為程式本身沒有問題，一定是硬體太差跟不上時代了，所以買了很多昂貴的硬體來增強效能，殊不知其實有可能把程式運作邏輯稍作修改就可以大幅改善效能。</p><p>附註：請從巨觀角度去想問題，學校的案例都是為了學習用而縮小，請想想看當你的網站或程式從 5 人到 1000000 人同時登入使用，根據本文的想法，你覺得會出現甚麼情況，增加硬體會是一個好的解決方案嗎？</p><h3>那我該怎麼開始學習這些東西？</h3><p>學習資料結構與演算法幾乎是軟體工程師的命，不只是為了完成各種軟體公司的面試關卡，也是為了更有效率地解決日常的軟體開發過程中的各種問題。</p><p>所以我們要問問自己兩個問題：如何用有效率的方式解決問題，以及如何使用正確的工具解決問題。不只是看到問題，只想到用什麼合適的演算法來處理特定的問題，也要想著有沒有對應的工具跟解法來達到最佳解。</p><p>市面上已經有很多教你怎麼學好資料結構跟演算法的課程，也有不少人已經做了心得分享，這邊我也提出我一些當時訓練自己思考並應用資料結構跟演算法的一些方向：</p><ul><li><strong>上解題網站練習</strong>：平台如 UVA、eTutor、CodeWar、Leetcode，他們都提供了大量的面試可能被問到的題目等等，從面試心得分享到面試中遇到的基礎到進階的各種問題。這些練習不僅可以幫助我們理解和實踐演算法和資料結構的概念，還可以提高我們解決問題的速度和效率。</li><li><strong>寫部落格分享心路歷程</strong>：這不僅僅是一種分享知識的方式，同時也是加深自己理解的過程。有醫生說過學習並且適當地寫作，我們可以更好地理解自己所學的內容，並且加深對於新知識的印象，部落格還有對外宣傳的效果，如果取得社群的各種回饋是可以獲得新的視角和建議。</li><li><strong>參加討論或加入開源專案</strong>：參與社群討論或貢獻開源專案可以讓我們在實際的開發環境中應用所學。這種實踐經驗是非常寶貴的，它不僅可以幫助我們應用和鞏固知識，還可以讓我們學習如何與其他開發者互相合作。</li><li><strong>尋找實習機會</strong>：直接投身於工作環境是學習和成長的絕佳方式。實習可以提供實際的工作經驗，讓我們在專業指導下應用和提高我們的技能。</li></ul><h3>就這樣嗎？</h3><p>以下是我寫作時參考的資料：</p><p><a href="https://www.slideshare.net/ccckmit/ss-56891871">陳鍾誠 — 用十分鐘 學會《資料結構、演算法和計算理論》</a></p><p>延伸閱讀：</p><p><a href="https://www.happybearsoftware.com/how-learning-data-structures-and-algorithms-makes-you-a-better-developer">How learning data structures and algorithms make you a better developer</a></p><h3>系列文</h3><p><a href="https://medium.com/@mybaseball52/%E5%88%B0%E5%BA%95%E5%BE%9E%E4%BA%8B%E8%BB%9F%E9%AB%94%E9%96%8B%E7%99%BC%E7%9A%84%E5%AD%B8%E7%94%9F%E7%BC%BA%E4%B9%8F%E4%BA%86%E7%94%9A%E9%BA%BC-%E4%B8%8A-%E6%BA%9D%E9%80%9A%E8%83%BD%E5%8A%9B%E7%AF%87-adb656728ab6?source=your_stories_page-------------------------------------">到底從事軟體開發的學生缺乏了甚麼？溝通能力篇</a></p><p><a href="https://mybaseball52.medium.com/does-algorithms-and-datastructures-matters-in-software-development-careers-ff54cb684926">到底從事軟體開發的學生缺乏了甚麼？資料結構與演算法篇</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ff54cb684926" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>