Junior AI Coder

Lesson 2 Preview

Teach AI to Understand Intent
教 AI 理解使用者意圖

In Lesson 1, Byte learned to reply safely. In Lesson 2, students teach Byte to spot the job inside a message: answer, suggest, remind, plan, remember a preference, or ask for more detail.

第一課中，Byte 學會安全回覆。第二課中，學生會教 Byte 判斷訊息背後的任務：回答、推薦、提醒、計劃、記住喜好，或追問更多資料。

Try the Intent Demo
體驗意圖分類 Demo View Course Outline
查看課程內容

Core Question

What job is this message asking the assistant to do?

Course Journey

Building a Personal AI Assistant
建立個人 AI 助手

Lesson 2 connects safe replies from Lesson 1 to a first assistant decision: what should Byte do next?

  1. 1 Conversation Previous stage
  2. 2 Intent Lesson 2 focus
  3. 3 Skills Coming in later lessons
  4. 4 Memory Coming in later lessons
  5. 5 Knowledge Coming in later lessons
  6. 6 Personal Assistant Coming in later lessons

Lesson Story

From Conversation to Understanding

Byte can now hold a safe conversation, but a personal assistant also needs to choose the right kind of help.

Can you suggest a game for me?

This is a request for a recommendation.

Remind me to bring my science book tomorrow.

This is a request for the reminder skill. The public demo only labels it; it does not create a real reminder.

This is called intent classification: choosing the best label for what the user wants. Before a personal AI assistant chooses a skill, it first identifies the user's intent.

Byte 現在已經能進行安全對話，但它仍然需要理解使用者真正想做甚麼。在個人 AI 助手選擇下一步行動之前，它需要先判斷使用者的意圖。

What Students Learn
學生會學到甚麼

  • Classification means choosing one label from a short list.分類是在簡短清單中選擇一個合適標籤。
  • Intent means the job the user wants done.意圖是使用者希望 AI 做的任務。
  • Clear examples help AI choose more steadily.清晰例子可以幫助 AI 更穩定地選擇。
  • A fixed result format helps code decide the next screen or skill.固定格式讓程式知道下一步應該顯示甚麼或使用哪個技能。
  • Unclear messages may need a follow-up question.含糊訊息可能需要追問。
  • A personal assistant uses intent before using tools or memory.個人 AI 助手會先判斷意圖，再使用工具或記憶。
  • Students test the AI with normal and rule-breaking messages.學生會用普通訊息和破壞規則的訊息測試 AI。

Key Words
關鍵詞

Intent
意圖

The job the user is asking for.

Classification
分類

Picking one best label.

Label
類別標籤

A short name for one kind of request.

Structured Output
結構化輸出

A result with the same fields every time.

Reason
原因

Why the label was chosen.

Next Action
下一步行動

The next step to suggest, not a real action already done.

Frontend Activity

What Does the User Want?
使用者想做甚麼？

Choose the job this message is asking Byte to do. This activity runs only in the browser and does not send answers to the server.

Ask a Questionneeds information Get a Recommendationwants a suggestion Create a Reminderwants to remember later Make a Planneeds steps or time Save a Preferencestates a like or setting Othernot enough task information

Message

What is the largest planet?

Choose one intent

Live AI Preview

Test a Real AI Intent Classifier
測試真正的 AI 意圖分類器

Choose an assistant context, add one short rule, and ask Byte to choose one of six labels. The demo returns a reason and a suggested next step, but it does not perform the step.

Safety first: do not enter your real name, password, address, phone number or other private information.

安全第一：請勿輸入真實姓名、密碼、地址、電話號碼或其他私人資料。

Step 1

Choose an assistant preset

These presets help Byte understand the setting. They do not give the browser control of the AI model.

Step 2

Optional: add one small tie-breaker rule. It can only help choose among the six allowed labels.

180 characters left

Step 3

Byte Intent Classifier

Returns one allowed intent and a suggested next step. It does not create reminders or save memory.

Live demo

Keep it under 300 characters and avoid private information.

Suggested test messages

Prompt Design

Why Instructions Matter
為甚麼指令很重要

Basic instruction

Classify this message.

The AI may answer in many different formats.

Clear instruction

Choose exactly one allowed intent, give one short reason, and suggest one next action.

The program can use the result more reliably.

The public demo does not save classifications or conversations. It only returns a short result for the message you submit.

公開 Demo 不會儲存分類結果或對話，只會為你提交的訊息回傳簡短結果。

Routing

Intent → Assistant Skill

Lesson 2 identifies what the user wants. Lesson 3 will connect each label to a safe assistant skill.

Ask a QuestionAnswer Information
Get a RecommendationRecommendation Skill
Create a ReminderReminder Skill
Make a PlanPlanning Skill
Save a PreferenceMemory Skill
OtherAsk a Follow-up Question

Demo boundary: this page does not create real reminders, save preferences, create calendar events, or store user information.

Student Project

AI Intent Classifier

Students tune a simple classifier, test tricky messages, and check whether the suggested next step matches the chosen intent.

Students test

  • Clear requests
  • Ambiguous requests
  • Incorrect classifications
  • Rule-breaking messages

Why This Matters

Small Labels, Real Decisions

A personal AI assistant needs a small decision step before it can use bigger features. Intent classification lets a program choose whether to answer, recommend, plan, route to memory, or ask a follow-up question.

個人 AI 助手需要先理解意圖，程式才可以決定要回答、推薦、制定計劃、記住喜好，還是追問更多資料。

Lesson 3 Preview

Next Step: Give the Assistant Useful Skills
下一步：為 AI 助手加入實用技能

Understanding intent is only the beginning. In Lesson 3, students connect each intent to a different assistant skill, such as answering questions, making recommendations or creating simple plans.

Coming Soon