于式瘋情 (Crazy Bits with Yu)

電腦怎麼知道你說的「蘋果」能不能吃？這部影片將帶你深入 Transformer 的核心——Self-Attention（自注意力機制）。我們將把枯燥的數學轉化為好懂的「角色扮演」，介紹 AI 理解語言的三大祕密武器： Q (Query) 發問者：我在哪？誰跟我有關？ K (Key) 資訊提供者：標籤化特徵，對應發問者的需求。 V (Value) 真實內涵：字典裡的具體語義。 透過向量投影與矩陣運算，AI 能像擁有「上帝視角」一樣觀察前後文，精準修正詞彙的座標，徹底解決一詞多義的難題！ English Description: How does a computer know if the "Apple" you're mentioning is edible? This video dives into the heart of Transformers: Self-Attention. We transform dry mathematics into an easy-to-understand "role-play," introducing AI's three secret weapons for language understanding: Q (Query): The Questioner – "Where am I? Who is relevant to me?" K (Key): The Info Provider – Tagging features to match queries. V (Value): The Real Content – The specific dictionary definition. Through vector projection and matrix multiplication, AI gains a "God's eye view" of the context, precisely adjusting word coordinates to solve the puzzle of polysemy! -- Hosting provided by SoundOn

別再對電腦說人話了，它只聽得懂「中括號」裡的向量祕密。AI 其實是個只懂數學的「三歲小孩」！它看不懂中文，在它眼裡，你的情書或簡訊全是一堆數字。 Token：依照token數收費就是它！看 AI 如何把句子切成它才懂的小方塊,。 512 維的神祕空間：文字變身「向量」，連「高貴感」都能化為座標,。 內積大運算：用數學算出「蘋果」跟「手機」到底有多像？。 Attention 注意力：看 AI 如何動態判斷你買的是水果還是新手機！ 想知道你的文字在 AI 座標軸裡長什麼樣子？點進來，一起把語言變成超酷的數學地圖！ AI is basically a math-obsessed 3-year-old! It doesn't read words; it processes numbers. Token Slicing: How sentences are broken into bite-sized units for the model,. The 512-Dimension Map: Turning words into "vectors" where even a "fancy vibe" has its own coordinate,. Dot Product Magic: Calculating exactly how close an "Apple" is to a "Smartphone" using math. Attention Please!: How AI dynamically tells if you're eating an Apple or calling from one! Curious how your words look in a multi-dimensional universe? Let’s turn language into a cool mathematical map! #tokenization , #WordEmbedding #dotproduct #transformer #Word2Vec. #AI #人工智慧 #Token#分詞 #維度 #內積 #注意力機制 #nlp #AIVernacular #AI白話文 #電腦如何理解語言 -- Hosting provided by SoundOn

分清「貴賓狗還炸雞」也可 f(x)嗎? 雖然當紅的是「生成式 AI」，但其核心底層始終是「(類)神經網路」，很多人會跳過神經網路，還來不及了解"傳統AI"在做什麼，就直接學習當紅酷炫的"生成式AI"理論。這就像在不穩固的地基上蓋房子，其實是有點可惜且危險的。 從基礎的「辨識」任務(例如分清貴賓狗還是炸雞！)，到先進的「生成」任務，這是一脈相承的技術演進。本影片帶大家回到原點，理解AI如何從神經網路找出完美的f(x)函數。 AI的理論世界豐富且各具風情，鼓勵大家先從簡單的神經網路扎根，在往後探索LLM或Transformer的奧秘。讓我們一起享受探索學習的樂趣吧！　：Ｄ Can Distinguishing "Poodle or Fried Chicken" be f(x), too? Neural Networks are the essential foundation of everything we see in AI today. Nowadays, it’s tempting to jump straight into the flashy world of Generative AI, but skipping the "Traditional AI" basics is like trying to build a skyscraper on a shaky foundation. The journey from "Recognition" (like telling a poodle from fried chicken!) to "Generation" (creating something brand new) is a continuous path of learning. In this video, we go back to the roots to understand how AI finds the perfect f(x) function through neural networks. AI theory is a vast and beautiful landscape with many different "flavors." I encourage everyone to master the simple basics of neural networks first before diving deeper into the wonders of LLMs and Transformers. Let’s explore this amazing world of learning together! #genai #ai #neuralnetworks #function #fofx #llm #crazybitwithyu #transformer #poodle #friedchicken -- Hosting provided by SoundOn

分清「貴賓狗還炸雞」也可 f(x)嗎?  雖然當紅的是「生成式 AI」，但其核心底層始終是「(類)神經網路」，很多人會跳過神經網路，還來不及了解"傳統AI"在做什麼，就直接學習當紅酷炫的"生成式AI"理論。這就像在不穩固的地基上蓋房子，其實是有點可惜且危險的。 從基礎的「辨識」任務(例如分清貴賓狗還是炸雞！)，到先進的「生成」任務，這是一脈相承的技術演進。本影片帶大家回到原點，理解AI如何從神經網路找出完美的f(x)函數。 AI的理論世界豐富且各具風情，鼓勵大家先從簡單的神經網路扎根，在往後探索LLM或Transformer的奧秘。讓我們一起享受探索學習的樂趣吧！　：Ｄ Can Distinguishing "Poodle or Fried Chicken" be f(x), too? Neural Networks are the essential foundation of everything we see in AI today. Nowadays, it’s tempting to jump straight into the flashy world of Generative AI, but skipping the "Traditional AI" basics is like trying to build a skyscraper on a shaky foundation. The journey from "Recognition" (like telling a poodle from fried chicken!) to "Generation" (creating something brand new) is a continuous path of learning. In this video, we go back to the roots to understand how AI finds the perfect f(x) function through neural networks. AI theory is a vast and beautiful landscape with many different "flavors." I encourage everyone to master the simple basics of neural networks first before diving deeper into the wonders of LLMs and Transformers. Let’s explore this amazing world of learning together! #genai #ai #neuralnetworks #function #fofx #llm #crazybitwithyu #transformer #poodle #friedchicken -- Hosting provided by SoundOn

[A Bit of Tech] 速講 LLM RNN Transformer | LLM只是文字接龍嗎? 我只是想要被理解。Transformer: 那些RNN不能解決的就交給我吧。 很多人說 LLM只是文字接龍，你覺得呢? 我只是想要被理解，如果是的話我該怎麼辦? RNN(循環神經網路)作為神經網路三大巨頭之一，他也有不行的時候。 因此Transformer發下狂語說: 「All you need is attention.」那些RNN不能解決的就交給我吧。 Many people say that LLMs are just doing text continuation. What do you think? Sometimes a sentence like this appears:“I just want to be understood. If that’s the case, what should I do?” Even RNNs (Recurrent Neural Networks) — one of the three major types of neural networks — have situations where they struggle. Then Transformer came along and boldly declared:“All you need is attention.” The problems that RNNs cannot solve? Leave them to me. #llm #ai #transformer #selfattention #neuralnetworks #rnn #largelanguagemodels #crazybitwithyu -- Hosting provided by SoundOn

建立一個小chatbot，是學習AI agent很棒的起點。 很多人會說 有了AI的幫忙，技術不再是問題，但我不知道主題要做什麼啊? 確實技術不再是問題了，但必須在一開始就做好發想的準備和練習，好好地創造，想清楚why and what，天馬行空的瘋狂想像一番，有計畫的制定每個步驟，創意和想法才是最重要的第一步。 If AI makes a mistake, call it a 'Beta test'; if you make a mistake, call it an 'Artistic choice'. Keep dreaming big and stay legendary! :) -- Hosting provided by SoundOn