讓我們一同運用AI探索「推論」的世界!在接下來的12/26課程中,我們將探索從課程例題中得出結論的藝術,以及將調查報告翻譯成繁體中文。通過評估回饋感受、分析情感色彩和辨識任務中的關鍵要素,深入理解推論的精髓。加入我們,體驗引人入勝的推論技巧,將語言能力與批判性思維融合,開啟一場充滿探索與驚艷的旅程!
請攜帶你的手機,平板,或是筆電。(螢幕大會比較好用,不過手機也可以)
讓我們一同運用AI探索「推論」的世界!在接下來的12/26課程中,我們將探索從課程例題中得出結論的藝術,以及將調查報告翻譯成繁體中文。通過評估回饋感受、分析情感色彩和辨識任務中的關鍵要素,深入理解推論的精髓。加入我們,體驗引人入勝的推論技巧,將語言能力與批判性思維融合,開啟一場充滿探索與驚艷的旅程!
請攜帶你的手機,平板,或是筆電。(螢幕大會比較好用,不過手機也可以)
使用定界符標示不同部分:
可要求結構化輸出:
驗證條件是否符合:
少量示範:
明確步驟要求完成任務:
引導模型推理:
限制模型幻覺:
based on ChatGPT and the course in https://learn.deeplearning.ai/chatgpt-prompt-eng/lesson/3/iterative
提示(Prompt)撰寫其實是一個逐步修正的過程,不要期待第一次就上手
迭代式提示開發: 創建有效的提示通常需要多次迭代。初始提示可能無法完全符合預期的結果。因此,根據初始輸出對提示進行精煉至關重要。
清晰和具體: 在提示的指令中清晰和具體很重要。無論是字數、句子結構還是特定所需的詳細資訊,都提供清晰的指導對於模型獲得更準確的結果至關重要。
調整提示: 如果初始輸出不符合期望(例如,長度過長或缺少特定資訊),則需要修改指令。這可能涉及調整字數、添加或更改特定細節,或改變輸出的焦點。
複雜的提示: 隨著時間的推移,隨著迭代的進行,提示可能變得更加詳細和複雜。這種複雜性可能包括格式要求(如HTML輸出)或要求特定的數據安排(如表格)。
應用軟體的成熟度: 隨著應用的成熟,對提示進行評估,應對較大量的示例進行評估變得至關重要,以確保在不同的輸入下具有一致且準確的性能。
整體而言,透過迭代方法開發提示,從單一示例開始,逐步根據輸出精煉指令,有助於為利用LLM的各種應用程式打造有效的提示。
Example: Test of comprehension
Version 1: Generate the problems
Prepare 10 multiple choice problems from the paste text above to test comprehension.
Shani Dhanda's activism primarily focuses on:
A) Raising awareness for mental health issues
B) Advocating for environmental sustainability
C) Promoting cultural diversity
D) Advocating for the rights of people with disabilities
What year did Shani Dhanda organize the first event supporting disabled individuals and South Asian women?
A) 2018
B) 2019
C) 2020
D) 2021
What medical condition does Shani Dhanda have?
A) Rare genetic disorder
B) Brittle Bone Disease
C) Chronic fatigue syndrome
D) Muscular dystrophy
The "Diversability Card" app created by Shani Dhanda aims to provide:
A) Social media networking for disabled individuals
B) Travel guidance for people with disabilities
C) Restaurant ratings and food delivery services
D) Discounts for people with disabilities
Shani Dhanda's vision for the future primarily revolves around:
A) Global entrepreneurship programs
B) Financial independence for disabled individuals
C) Reducing carbon emissions
D) Increasing government funding for disability services
Version 2: Making it more interactive
Prepare 10 multiple choice problems from the paste text above to test comprehension. Generate one problem at a time. Then you request an answer. Then generate next problem.
Prepare 10 multiple choice problems from the paste text above to test comprehension. Generate one problem at a time. Then you request an answer. After you receive one, grade it. Then generate next problem.
Prepare 10 multiple choice problems from the paste text above to test comprehension. Generate one problem at a time. Then you request an answer. After you receive one, grade it. Then generate next problem. In the end, tell me out of the 10 questions above, how many questions are correctly answered?
Version 1 Generate the problems
Based on the vocabulary: (Exoskeleton, Myoelectric, Sensors, Hydraulic, Cylinder, Prosthesis, Rehabilitation, Paralysis, Spinal cord, Neural), prepare 5 fill the blank quiz questions.
Version 2 One at a time
Based on the vocabulary: (Exoskeleton, Myoelectric, Sensors, Hydraulic, Cylinder, Prosthesis, Rehabilitation, Paralysis, Spinal cord, Neural), prepare 5 fill the blank quiz questions. Generate one problem at a time. Then you request an answer. Then generate next problem.
Based on the vocabulary: (Exoskeleton, Myoelectric, Sensors, Hydraulic, Cylinder, Prosthesis, Rehabilitation, Paralysis, Spinal cord, Neural), prepare 5 fill the blank quiz questions. Generate one problem at a time. Then you request an answer. After you receive one, grade it. Then generate next problem.
Based on the vocabulary: (Exoskeleton, Myoelectric, Sensors, Hydraulic, Cylinder, Prosthesis, Rehabilitation, Paralysis, Spinal cord, Neural), prepare 5 fill the blank quiz questions. Each question has only one missing word. Generate one problem at a time. Then you request an answer. After you receive one, grade it. Then generate next problem. In the end tell me how many are correctly answered.
Based on the vocabulary: (Exoskeleton, Myoelectric, Sensors, Hydraulic, Cylinder, Prosthesis, Rehabilitation, Paralysis, Spinal cord, Neural), randomize the order of these words and then prepare 5 fill the blank quiz questions. Each question has only one missing word. Generate one problem at a time. Then you request an answer. After you receive one, grade it. Then generate next problem. In the end tell me how many are correctly answered.
based on ChatGPT and the course in https://learn.deeplearning.ai/chatgpt-prompt-eng/lesson/3/iterative
方便的話,可以攜帶筆電或是平板等螢幕比較大的設備,
AI工作坊主要內容請參看 https://nptechsl.blogspot.com/
裡面有七個程式實作單元,12/
有任何問題都可以隨時聯絡我
4 Sessions
請攜帶你的手機,平板,或是筆電。(螢幕大會比較好用,不過手機也可以)
https://learn.deeplearning.ai/chatgpt-prompt-eng/lesson/3/iterative
申請帳號一般隱私安全注意事項
1. 避免使用公用電腦登入或註冊帳號,以免隱私資料被可能潛在的惡意軟體盜取
2. 避免使用重複的帳號密碼,建議避免一組帳號走天下,尤其是不應與金融財務相關帳戶或是職務相關登入帳號密碼相同,以免帳號被外洩時遭到連帶連鎖影響
請使用上述課程的人工智慧程式回答以下問題,請用螢幕拍照,同時附上程式與結果
Iterative (逐步修正)12/12
1. 將產品規格翻譯成繁體中文(Traditional Chinese)
2. 依據辦公椅的技術規格,為行銷團隊產生一個350個中文字的產品文字,看看文字長度是否合宜,如果太長,試著縮短文字長度,是否變得比較專業簡潔有力
3. 檢查所產生的行銷文字,是否註明產地,如果沒有請重新產生
Summarize (摘要)12/19
1. 完成英文課程裡的例題。
2. 將四個客戶回饋(玩具,落地燈,電動牙刷,果汁機)翻譯成繁體中文
3. 將四個客戶回饋摘要為50個字,如果字數不足,請適度調整長度。需要翻譯成繁體中文
Inferring (推論)12/19, 12/26
1. 完成英文課程裡的例題。
2. 為前一題每個回饋內容產生二到五個感受,留下你認為最接近事實的感受,刪除多餘
3. 判斷每篇評論內容情緒為正面(快樂,滿意,激勵)或負面(憤怒,沮喪,失望)或中性(平和,平靜,事實陳述)
4. 不可使用ChatGPT,根據你的獨立思考為每個評論的翻譯,感受與情緒判斷進行綜合評分,0分(完全不能接受)到10分(出奇地好)
Inferring (推論)Part 2, 12/26
1. 將問卷調查報導翻譯成繁體中文
2. 為報導內容產生二到五個關鍵詞,留下最貼近事實的關鍵詞,刪除多餘
3. 這篇報導是否有關農業,地方政府,歷史,公部門,請逐一檢查
Transforming (轉換)
1. An old dude who invented AI says it is unsafe. 翻譯為中文與日文
2. 請將上述俚語改寫為正式英文商業書信。
Expanding (擴展)
1. 線上課程中客戶來信內容是正面還是負面?(sentiment 變數是多餘的,應刪除)
2. 改用另一封客戶來信,測試回信功能是否正常
3. 依據所提供的個人資料,撰寫一封求職信
以下練習需要用到ChatGPT, 進入點如下。第一次使用者需要先註冊一個帳號。
1. 使用ChatGPT以繁體中文回答Step 1中的五個簡答題
2. 使用ChatGPT為每個答題內容產生至多四個關鍵詞(Keywords)
3. 使用ChatGPT判斷每篇答題內容情緒為正面(快樂,滿意,激勵)或負面(憤怒,沮喪,失望)或中性(平和,平靜,事實陳述)
4. 不使用ChatGPT,為每題的簡答內容,關鍵詞與情緒判斷進行綜合評分,0分(完全不能接受)到10分(出奇地好)