檢索擴增生成(RAG)技術的發展現況與文本知識提取的應用 / Retrieval Augmented Generation (RAG) and Its Application in Text Knowledge Extraction

檢索擴增生成(RAG)現在已經是大型語言模型領域裡的新寵兒，而且是屬於易懂難經的技術。這篇文章內容是我在計劃書裡對於檢索擴增生成(RAG)技術的內涵和相關應用的介紹。在這個部落格也留存一份記錄吧。

Retrieval Augmented Generation (RAG) has become the new darling of the large language model domain, and it's one of those technologies that are easier to understand than to master. This post contains the introduction to RAG's core concepts and applications that I wrote for a proposal. I'm also keeping a copy here on the blog.

雜談：用Make設定了RSS傳送到Threads / TALK: Forword the RSS Feed to Threads in Make

在我寫完「RSS自動轉發到Threads跟X (Twitter)：使用dlvr.it」之後，dlvr.it就轉變成完全收費的制度，因此自動轉發也跟著無以後繼。最近我總算能騰出時間來撰寫Blog，同時也把Threads API整合到Make平臺裡，讓「布丁布丁吃什麼？」的新文章能夠自動轉發到Threads裡面。至於這個方法是否能夠成功，就要等時間來驗證了。以下就記錄我大致上的做法。

Home Assistant自動化的限制：無法偵測長時間狀態 / Limitations of Home Assistant Automation: Inability to Detect Long-Duration States

我本來想在Home Assistant設定自動化「如果實體的狀態維持[關閉]超過24小時，就把它設為[開啟]」，結果HA卻一直沒能觸發自動化。最後我把條件改成時間模式定期檢查、觸發，然後把條件設定在「且如果」裡面，這樣自動化才能夠正常運作。

I was trying to set up an automation in Home Assistant that would turn an entity [on] if its state remained [off] for more than 24 hours. However, the automation never triggered. I eventually got it working by changing it to a time-based automation that checks periodically and then using an condition to check the entity's state.

雜談：是時候該來處理一下Dify的問題了 / TALK: It's Time to Address the Issues with Dify

在「自行架設大型語言模式應用程式：Dify」這篇裡面，我用Dify在筆電架設了可客製化、具備RAG的大型語言模型應用程式。但這段期間用下來還是遭遇了很多問題。以下就稍微列舉一下我遭遇的狀況。

現代研究需要現代手段：研究工具彙整 / Modern Research Needs Modern Tools: A Compilation of Research Instruments

我在「現代研究需要現代手段：用知識管理和AI輕鬆做研究！」演講中使用了很多研究工具。為了方便讓大家找到這些工具，我把演講裡提到的工具都拉出來，額外放在下面進行介紹。這些工具包含了書籍、軟體、網站等等。有些是付費才能使用，有些是可以免費註冊帳號即可使用，使用之前請自行評估一下囉。

In my presentation, "Modern Research Demands Modern Tools: Leveraging Knowledge Management and AI for Effortless Research," I utilized a variety of research tools. For your convenience, I've listed those tools below with additional details. These resources include books, software, websites, and more. Some require payment while others offer free accounts. Please evaluate each option before using it.

雜談：用Chromecast接收器播放螢幕畫面並不容易 / Talk: Casting Your Screen to a Chromecast Device Can Be Challenging

由於我使用的設備是以Linux和Android為主，在螢幕投放的選擇上，Chromecast自然就是我的首選。雖然市面上很多號稱支援Chromecast的方案，但其實仍只有Google推出的Chromecast才具備完整的功能。不過，Chromecast的天性使得它難以移動，iOS使用者也不容易使用。目前要架設一個可移動的投影畫面接收器，仍然不是一件容易的事情。

現代研究需要現代手段：用知識管理和AI輕鬆做研究！ / Modern Research Demands Modern Tools: Leveraging Knowledge Management and AI for Effortless Research!

今年10月的時候我去南藝大為研究生們講述了活用圖書館資源、AI工具以及Zotero來撰寫研究論文的方法。有了內建大型語言模型資料庫的協助，例如Scopus、Semantic Scholar、Consensus、Elicit、SciSpace，現在作研究的方式已經有了很大的轉變了。投影片的內容分成了「研究搜尋篇」與「閱讀寫作篇」兩個部分，以下就來看看這次演講的投影片內容吧。

In October of this year, I gave a presentation at Tainan National University of the Arts to graduate students on how to leverage library resources, AI tools, and Zotero for research paper writing. With the help of databases incorporating large language models, such as Scopus, Semantic Scholar, Consensus, Elicit, and SciSpace, the way we conduct research has been significantly transformed. The presentation was divided into two parts: "Investigate Research" and "Reading and Writing." Let's take a look at the slides from the presentation below.