提示技術的五大支柱
除了簡單的對話介面之外，大型語言模型（LLMs）可作為結構化資料處理的精密引擎。本課程介紹五種功能「基本元件」，讓開發者能將人工智慧整合至專業工作流程中。
功能路徑圖
- 摘要：將大量文字（如產品評論）濃縮為簡明扼要的摘要，同時保留關鍵資訊。
- 推論：執行自動化分析，例如情感檢測、情緒辨識，或從文本中提取特定資料點（如品牌名稱）並轉換為 JSON 格式。
- 轉換：將文字從一種形式轉換為另一種形式——包括語言翻譯、語氣調整（例如由正式轉為口語），以及資料格式轉換（例如由 JSON 轉為 HTML）。
- 擴展：利用短訊號或項目清單生成長篇內容，例如客製化的客服電子郵件。
- 聊天機器人：運用對話歷史與預設角色（系統、使用者、助理）來建立互動式、具上下文意識的助手。
大型語言模型的無狀態特性
關鍵概念：與人類不同，大型語言模型是「無狀態」的。除非您在每次 API 呼叫時手動傳回完整的對話歷史，否則它們不會記得先前的互動。這段累積的歷史被稱為上下文。
TERMINALbash — 80x24
> Ready. Click "Run" to execute.
>
Question 1
A developer needs to extract the "Price" and "Discount Code" from 1,000 emails and save them to a database. Which approach and temperature setting are most appropriate?
Challenge: Tone Transformer
Apply the Transforming and Chatbot role concepts.
You are building a "Tone Transformer" for a corporate helpdesk. You need to convert a frustrated user's slang-heavy complaint into a "Formal" and "Polite" summary for a manager.
Task
Specify the role of the "System" message to ensure the model acts appropriately, and write the "User" prompt to translate: 'The app is buggy as heck and I want my money back ASAP!'
System Message:
User Prompt:
Expected Result:
"You are a professional executive assistant. Your task is to rephrase communication into a formal, corporate-appropriate tone."User Prompt:
"Translate the following text into a formal summary: 'The app is buggy as heck and I want my money back ASAP!'"Expected Result:
"The customer is reporting technical instability within the application and is requesting an immediate refund."