Structure: Section A has 20 multiple-choice questions, 3 marks each (60 marks). Section B has 2 structured tasks (40 marks). The answer key and marking guide begin on a new page after Section B.試卷結構:A部分為20題選擇題,每題3分(共60分)。B部分為2題結構化任務題(共40分)。答案與評分指南列於B部分之後的新頁。
第一部分・選擇題 Section A · Multiple-Choice
Choose the one best answer for each question. Each question is worth 3 marks.
每題請選出一個最佳答案,每題3分。
Q13 分
Using A=1, B=2, C=3, D=4 …, write the word “BAD” as numbers.
用 A=1、B=2、C=3、D=4……的方式,把「BAD」這個字轉換成數字。
B A D → ❓ ❓ ❓
A
2 1 4
B
1 2 3
C
4 1 2
D
2 4 1
A正解 Correct Answer
B=2, A=1, D=4, so BAD = 2 1 4. Letters can be encoded as numbers.
B=2、A=1、D=4,所以BAD=2 1 4。文字可以被編碼成數字。
LO: LP.P1.3
Q23 分
A robot reaches a dead-end while trying to get through a maze. The smart thing to do is…
機器人在走迷宮時碰到死路,聰明的做法是……
A
try the exact same path again
再試一次一模一樣的路徑
B
go back and try a different path
退回去,換一條路試試看
C
smash through the wall
直接把牆撞破
D
switch itself off
把自己關機
B正解 Correct Answer
Backing up and trying another path is the seed of search - try, check, try again.
退回去換一條路試試看,正是「搜尋」的雛形——嘗試、檢查、再嘗試。
LO: LP.P2.4
Q33 分
Which AI can make a NEW picture when you describe it in words?
當你用文字描述一張圖,哪一種AI可以「畫出一張全新的圖」?
A
A photo album application
相簿應用程式
B
A camera filter effect
相機濾鏡效果
C
An image generator
圖像生成器
D
A printed colouring book
印刷版著色本
C正解 Correct Answer
An image generator is generative AI that draws a brand-new picture from your words; the others only show or change pictures that already exist.
How does an AI learn to know what a cat looks like?
AI是如何學會「認識貓長什麼樣子」的?
A
It is born already knowing
它一出生就知道了
B
It asks a real cat itself
它自己去問一隻真的貓
C
It reads the word ‘cat’ once
它只讀過一次「貓」這個字
D
It looks at many cat pictures
它看過很多貓的照片
D正解 Correct Answer
AI learns from lots of examples (called data). Many cat pictures help it learn cats.
AI是從大量的範例(稱為資料)中學習的。很多貓的照片能幫助它學會認識貓。
LO: LP.P2.1
Q53 分
A chatbot builds a whole sentence by…
聊天機器人組出一整句話的方式是……
A
copying a full page from a book
直接抄一整頁書
B
drawing small pictures
畫一些小圖案
C
adding the most likely next word, again and again
不斷加上「最可能出現的下一個字」
D
counting to ten first
先從一數到十
C正解 Correct Answer
Generating text = repeatedly guessing the next word - a “what comes next?” game.
生成文字=不斷猜測下一個字——就像玩「接下來是什麼」的遊戲。
LO: LP.P3.2
Q63 分
Find the repeating rule, then say what colour the 8th bead is.
找出重複出現的規律,說出第8顆珠子是什麼顏色。
①🔴 ②🟡 ③🔵 ④🔴 ⑤🟡 ⑥🔵 ⑦❓ ⑧❓
A
🔴 red
🔴 紅色
B
🟡 yellow
🟡 黃色
C
🔵 blue
🔵 藍色
D
🟢 green
🟢 綠色
B正解 Correct Answer
The rule repeats every 3 beads: red, yellow, blue. The 7th starts a new group (red), so the 8th is yellow. Finding the repeating unit lets you predict any position, not just the next.
Should you tell a chatbot your home address and phone number?
你應該告訴聊天機器人你家的地址和電話號碼嗎?
A
Yes, so it can help you better
應該,這樣它才能幫你幫得更好
B
Yes, but only your address
應該,但只講地址就好
C
No, keep private things to yourself
不應該,隱私的事要自己保管好
D
Only if a friend tells you to
只有朋友叫我講才講
C正解 Correct Answer
Keep private information safe. Never share your address, phone number or passwords with a chatbot.
要保護好私人資訊。絕對不要把地址、電話號碼或密碼告訴聊天機器人。
LO: LP.P5.1
Q83 分
A chatbot sometimes says something that is NOT true. What should you do?
聊天機器人有時候會說出不是真的的內容,這時候你該怎麼做?
A
Believe it because computers are clever
相信它,因為電腦很聰明
B
Use it in your homework anyway
還是把它寫進作業裡
C
Share it with all your friends
分享給所有朋友
D
Check it with a trusted grown-up
找信任的大人一起確認
D正解 Correct Answer
AI can make mistakes. Always check important things with a trusted person or source.
AI也會犯錯,重要的事一定要找信任的人或可靠來源確認。
LO: LP.P5.2
Q93 分
Which of these can AI NOT really do?
下列哪一項是AI真正做不到的?
A
Sort pictures into groups
把圖片分類
B
Write a short story
寫一篇短篇故事
C
Truly feel happy or sad
真正感受到開心或難過
D
Answer simple questions
回答簡單的問題
C正解 Correct Answer
AI has no real feelings. It can use the word “happy,” but it does not feel.
AI沒有真正的情緒。它可以使用「開心」這個詞,但並不會真的感受到。
LO: LP.P1.1, LP.P1.4
Q103 分
To help an AI sort photos into “dog” and “not a dog,” we are teaching it to…
要讓AI把照片分成「狗」和「不是狗」,我們是在教它……
A
classify (put into groups)
分類(把東西歸類)
B
search (look things up)
搜尋(查找資料)
C
save (keep a copy on disk)
儲存(把資料存起來)
D
draw (make a brand-new picture)
繪圖(畫出全新的圖)
A正解 Correct Answer
Putting things into groups like dog / not-a-dog is called classifying.
把東西分成「狗/不是狗」這樣的組別,稱為分類。
LO: LP.P1.4
Q113 分
Your friend asks you to use AI to write their homework and pretend they did it. Is that fair?
朋友請你用AI幫他寫作業,然後假裝是他自己寫的,這樣公平嗎?
A
Yes, it saves a lot of time
公平,這樣可以省很多時間
B
Only for the hard subjects
只有難科目才可以這樣
C
Yes, if the work looks neat
公平,只要看起來整齊就好
D
No, that is not honest
不公平,這樣不誠實
D正解 Correct Answer
Pretending AI’s work is your own is not honest. AI can help you learn, but the work should be yours.
把AI做的成果假裝成自己的,這樣不誠實。AI可以幫助你學習,但成果應該是你自己的。
LO: LP.P5.3
Q123 分
A video app keeps showing you more cat videos after you watch one. What is it doing?
影片App在你看完一支貓咪影片後,一直推薦更多貓咪影片給你,它在做什麼?
A
picking videos completely at random
完全隨機挑影片
B
deleting the videos you skip past
刪除你跳過的影片
C
copying videos straight from your friends
直接複製朋友的影片
D
learning what you like and suggesting more of it
學習你的喜好,推薦更多同類型的內容
D正解 Correct Answer
A recommendation system learns from what you watch and suggests similar things, handy, but it can trap you in a bubble.
推薦系統會根據你看過的內容學習,並推薦類似的東西——很方便,但也可能把你困在同溫層裡。
LO: LP.P2.1
Q133 分
Inside a computer, a photo is really stored as…
在電腦裡面,一張照片其實是被儲存成……
A
lots of numbers
一大堆數字
B
a single letter of the alphabet
一個英文字母
C
the name of its main colour
它主要顏色的名稱
D
a tiny copy of the real scene
真實場景的小型複製品
A正解 Correct Answer
Computers store pictures (and words) as numbers, not paint or letters.
電腦儲存圖片(和文字)的方式是數字,不是顏料或字母。
LO: LP.P1.2
Q143 分
If an AI only ever saw RED apples, what might it get wrong?
如果一個AI只看過紅色的蘋果,它可能會弄錯什麼?
A
It will refuse to look at apples
它會拒絕看蘋果
B
It will recognise every kind of fruit
它會認得每一種水果
C
It might think all apples are red
它可能會以為所有蘋果都是紅色的
D
It will always give correct answers
它會永遠給出正確答案
C正解 Correct Answer
If the examples are not fair and varied, the AI learns a biased idea - here, that all apples are red.
如果範例不夠公平多元,AI就會學到有偏差的想法——這裡就是「所有蘋果都是紅色的」。
LO: LP.P2.1
Q153 分
What is a good way to use generative AI for your ideas?
使用生成式AI來幫你想點子時,好的做法是什麼?
A
Let it do all the thinking for you
讓它幫你想完所有事情
B
Use it only to play games
只拿來玩遊戲
C
Copy it exactly without reading
完全照抄,不用看內容
D
Use it for ideas, then think too
拿它來激發想法,自己也要動腦想
D正解 Correct Answer
AI can help you brainstorm, but you should still use your own brain and check the ideas.
AI可以幫你腦力激盪,但你仍然要自己動腦並檢查這些想法。
LO: LP.P3.1
Q163 分
To get good help from an AI assistant, the best way to ask is…
想從AI助理那裡得到好的幫助,最好的提問方式是……
A
in as few words as you can
用越少字越好
B
clearly, saying exactly what you want
說清楚,明確講出你要的是什麼
C
by repeating it many times over
重複講很多次
D
using only one single word
只用一個字
B正解 Correct Answer
Asking clearly and kindly is the seed of good prompting.
清楚、友善地提問,正是好提問(prompting)的雛形。
LO: LP.P4.1
Q173 分
Who makes AI programs?
AI程式是誰做出來的?
A
Other robots build them
由其他機器人打造
B
People write the programs
由人類撰寫程式
C
They form all by themselves
它們自己形成的
D
They come already inside every computer
每台電腦本來就內建好了
B正解 Correct Answer
People design and build AI. That means people are responsible for using it well.
AI是由人設計和打造出來的,這也代表人類要為妥善使用AI負責。
LO: LP.P1.1
Q183 分
Be the robot and follow the rule exactly: “IF a word starts with S, THEN put it in the STAR box; OTHERWISE put it in the MOON box.” Where do the words “sun” and “cat” go?
“sun” starts with S, so it goes in the STAR box; “cat” does not, so it goes in the MOON box. Following a rule exactly, step by step, is just what a computer does.
A helper robot wants to water a thirsty plant. Which step comes FIRST in its helper loop?
一台幫手機器人想幫口渴的植物澆水,在牠的「幫手循環」中,哪一步應該最先做?
A
pour water straight away
立刻倒水
B
check if the soil is dry
先檢查土壤是不是乾的
C
decide how much water to pour
決定要倒多少水
D
wait for a person to say go
等人說「開始」
B正解 Correct Answer
A helper works in a loop: Sense → Think → Act. It must sense (check the soil) first, before deciding and acting.
幫手是以「感知→思考→行動」的循環運作。它必須先感知(檢查土壤),才能決定並行動。
LO: LP.P4.2, LP.P4.3
第二部分・結構化任務 Section B · Structured Tasks
Answer all parts. Show your working or reasoning where asked.
請作答所有小題,若題目要求請寫出計算過程或推理。
任務一・扮演分類機器人20 分
Task 1 · Be the Sorting Robot
A robot sorts fruit using only two rules: • Rule 1: IF the fruit is yellow, THEN put it in Basket A. • Rule 2: IF the fruit is red, THEN put it in Basket B. The fruits are: 🍌 banana (yellow), 🍎 apple (red), 🍋 lemon (yellow), 🍓 strawberry (red), 🍏 green apple (green).
For each fruit, write which basket it goes in. (Be the robot - follow the rules exactly.)
請寫出每種水果應該放進哪個籃子。(扮演機器人,完全照規則做。)
b)
The green apple does not match any rule. What should the robot do? Explain your idea.
青蘋果不符合任何一條規則,機器人該怎麼做?請說明你的想法。
c)
Why does a robot need clear rules to do its job well?
為什麼機器人需要清楚的規則才能把工作做好?
a)
8 marks: Banana → Basket A · Lemon → Basket A · Apple → Basket B · Strawberry → Basket B. (2 marks each correct match.)
8分:香蕉→A籃・檸檬→A籃・蘋果→B籃・草莓→B籃。(每對答對得2分。)
Why this earns marks: Rewards following an algorithm step by step, exactly as written - the heart of how computers run rules.
得分原因:獎勵能一步一步完全照著寫好的演算法執行——這正是電腦執行規則的核心精神。
b)
6 marks: Any sensible answer: the robot should NOT guess and throw it away. It could leave it aside, ask a person, or a new rule could be added (e.g., “IF green, THEN Basket C”).
Why this earns marks: Shows the child understands that AI fails safely when it meets something outside its rules, instead of pretending to know.
得分原因:顯示學生理解AI在遇到規則之外的情況時,應該安全地不作為,而不是假裝自己知道答案。
c)
6 marks: Clear rules let the robot make the same correct choice every time and not get confused. Without rules it would not know what to do.
6分:清楚的規則能讓機器人每次都做出相同且正確的選擇,不會搞混。沒有規則,它就不知道該做什麼。
Why this earns marks: Introduces the idea that AI behaviour comes from its instructions/data - a first step toward understanding bias and error.
得分原因:引導學生理解AI的行為來自它的指令/資料——這是理解偏誤與錯誤的第一步。
满分條件 For full marks
Full marks need correct sorting (a), a safe idea for the unknown fruit (b), and a reason rules matter (c).分類正確(a)、對未知水果提出安全的處理方式(b),並說明規則為什麼重要(c),即可獲得满分。
LO: a) LP.P2.3 · b) LP.P2.3, LP.P1.4 · c) LP.P2.3 / a) LP.P2.3・b) LP.P2.3, LP.P1.4・c) LP.P2.3
任務二・設計一台幫手機器人(AI代理人)20 分
Task 2 · Design a Helper Robot (AI Agent)
An AI agent is a helper that can do three things by itself to reach a goal: SENSE (look or listen) → THINK (decide what to do) → ACT (do something). Imagine a friendly helper robot for your classroom whose goal is to keep the classroom tidy.
Draw or describe your helper robot. What can it sense (see/hear)?
畫出或描述你的幫手機器人。牠可以感知(看見/聽見)什麼?
b)
Write its three steps for tidying up: SENSE → THINK → ACT.
寫出牠整理教室的三個步驟:感知→思考→行動。
c)
Think carefully: Your robot sees a drawing lying on the floor. It is told to throw away rubbish, but this might be someone’s special artwork. What should the robot do, and why?
Why this earns marks: Checks the child grasps that an agent needs input from the world before it can act.
得分原因:檢驗學生是否理解代理人需要先從外界獲得輸入,才能採取行動。
b)
9 marks: SENSE: sees a toy/paper on the floor. THINK: decides it is out of place and where it belongs. ACT: picks it up and puts it in the right spot. (3 marks per stage.)
6 marks: The robot should NOT just throw it away. It should pause and check or ask a person first (“Is this rubbish or something you want to keep?”), because it could be someone’s treasured work.
Why this earns marks: A real dilemma: following an order vs. respecting people. Good answers show the agent should ask when unsure rather than cause harm - an early lesson in AI acting responsibly.
得分原因:這是一個真實的兩難:服從指令 vs. 尊重他人。好的答案會顯示,代理人在不確定時應該先詢問,而不是造成傷害——這是AI負責任行動的初步概念。
满分條件 For full marks
Award up to full marks for a clear agent (a), a complete sense→think→act loop (b), and a kind, careful choice with a reason (c).清楚描述代理人(a)、完整的感知→思考→行動循環(b),以及一個善良、謹慎且有理由的選擇(c),即可獲得满分。
LO: a) LP.P4.2, LP.P4.3 · b) LP.P4.2, LP.P4.3 · c) LP.P4.2, LP.P4.3 / a) LP.P4.2, LP.P4.3・b) LP.P4.2, LP.P4.3・c) LP.P4.2, LP.P4.3
國小高年級 Grades 4–6
Structure: Section A has 30 multiple-choice questions, 2 marks each (60 marks). Section B has 2 structured tasks (40 marks). The answer key and marking guide begin on a new page after Section B.試卷結構:A部分為30題選擇題,每題2分(共60分)。B部分為2題結構化任務題(共40分)。答案與評分指南列於B部分之後的新頁。
第一部分・選擇題 Section A · Multiple-Choice
Choose the one best answer for each question. Each question is worth 2 marks.
每題請選出一個最佳答案,每題2分。
Q12 分
What can generative AI do that a normal calculator cannot?
生成式AI能做到一般計算機做不到的是什麼?
A
create brand-new text, images or music
創造全新的文字、圖像或音樂
B
add a long list of numbers
把一長串數字加起來
C
work out long sums very fast
很快算出長串的加總
D
show the answer on a small screen
在小螢幕上顯示答案
A正解 Correct Answer
Generative AI produces brand-new content based on patterns it learned; the other options are ordinary computer tasks.
生成式AI會根據學到的模式產生全新的內容;其他選項都只是一般電腦就能做到的工作。
LO: UP.P3.1
Q22 分
You want a clearer answer from a chatbot. Which change to your typed request (your prompt) helps most?
你想從聊天機器人得到更清楚的答案,對你輸入的請求(提示詞)做哪種調整最有幫助?
A
repeat the same request several times
把同樣的請求重複打好幾次
B
add a role, a clear task and a limit
加入角色設定、明確任務與限制條件
C
make the request as short as possible
把請求寫得越短越好
D
write the whole request in capitals
整段請求都用大寫字母
B正解 Correct Answer
A prompt works best with a role, a specific task and a constraint; vague prompts give vague answers.
提示詞加上角色、具體任務與限制條件效果最好;模糊的提示只會得到模糊的答案。
LO: UP.P3.3
Q32 分
A chatbot like a Large Language Model mainly works by…
像大型語言模型這樣的聊天機器人,主要的運作方式是……
A
storing every possible answer in a database
把所有可能的答案都存在資料庫裡
B
following a fixed set of grammar rules
遵循一套固定的文法規則
C
searching the live web for each reply
每次回覆都即時搜尋網路
D
predicting the next likely word, step by step
一步一步預測最可能出現的下一個字
D正解 Correct Answer
An LLM predicts the next word again and again, using patterns from huge amounts of text.
大型語言模型會運用從海量文字中學到的模式,不斷反覆預測下一個字。
LO: UP.P3.1
Q42 分
AI systems learn their skills mainly from…
AI系統的能力主要是從哪裡學來的?
A
a single built-in fixed rule
一條內建的固定規則
B
random guessing each time
每次隨機亂猜
C
advice from one human expert
一位人類專家的建議
D
lots of example data
大量的範例資料
D正解 Correct Answer
AI learns patterns from training data. The data shapes what it can and cannot do.
AI從訓練資料中學習模式,資料的內容決定了它能做什麼、不能做什麼。
LO: UP.P2.1
Q52 分
An AI turns words into number-lists (vectors) so that words with similar meanings get…
AI把文字轉換成數字列表(向量),讓意思相近的字詞會得到……
A
completely random numbers
完全隨機的數字
B
the same number as every other word
和其他所有字一模一樣的數字
C
similar numbers, sitting close together
相近的數字,彼此靠得很近
D
no numbers at all
完全沒有數字
C正解 Correct Answer
Similar meanings get similar numbers - the idea behind embeddings.
意思相近的字會得到相近的數字——這正是「詞嵌入(embeddings)」背後的概念。
LO: UP.P1.1
Q62 分
To sort fruit photos into groups when no names are given, the AI looks for…
在沒有標示名稱的情況下,要把水果照片分組,AI會尋找的是……
A
shared features such as colour and shape
共同的特徵,例如顏色和形狀
B
the name of the photographer
攝影者的名字
C
the date each file was saved
每個檔案的儲存日期
D
the order they were uploaded
上傳的先後順序
A正解 Correct Answer
Grouping by shared features is how unlabelled data is organised.
根據共同特徵分組,正是未標記資料被整理的方式。
LO: UP.P1.2, UP.P2.4
Q72 分
A robot dog earns a point each time it takes a step without falling, and loses a point when it falls. After many tries it walks smoothly. How did it learn?
一隻機器狗每走一步沒跌倒就得一分,跌倒就扣一分。經過很多次嘗試後,牠走得很順了。牠是怎麼學會的?
A
by copying a rule book written by its makers
靠抄襲製造者寫好的規則手冊
B
by memorising a video of another robot walking
靠背下另一台機器人走路的影片
C
by asking its owner before every single step
每走一步之前都先問主人
D
by trial and error, keeping the actions that earned rewards
靠反覆嘗試錯誤,保留能獲得獎勵的動作
D正解 Correct Answer
Learning from rewards by trial and error is called reinforcement learning - moves that earn points get repeated, and moves that lose points get dropped.
靠獎勵透過試錯來學習,稱為強化學習——能得分的動作會被重複,會扣分的動作會被淘汰。
LO: UP.P2.5
Q82 分
A shocking video shows a real politician saying something they never said; it was AI-made. Before reacting, the wise step is to…
check whether it is a genuine clip from a trusted source
先確認這是不是來自可信來源的真實片段
C
assume it is real because it looks real
因為看起來很真實就認定它是真的
D
add your own caption and repost it
加上自己的說明文字後轉發出去
B正解 Correct Answer
AI-made fakes of real people (deepfakes) can look convincing; verify with trusted sources before believing or sharing.
AI偽造真人的內容(深偽影片)可能看起來很真實;在相信或分享之前,要先用可信來源查證。
LO: UP.P5.1, UP.P5.2
Q92 分
You sketch an AI “homework checker”. A sensible FAILURE to plan for - with a fix - is…
你設計了一個AI「作業檢查員」。一個應該事先規劃、並附上解決辦法的合理「失敗情境」是……
A
it confuses two pupils' similar handwriting; fix: ban handwriting
它把兩位學生相似的筆跡搞混了;解法:禁止手寫
B
it works slowly at busy times; fix: skip the checking step
忙碌時運作變慢;解法:直接跳過檢查步驟
C
it marks a messy but correct answer wrong; fix: allow human review
它把字跡潦草但答案正確的作業判錯;解法:允許人工複核
D
it needs many example answers; fix: train it on none
它需要很多範例答案;解法:完全不給它任何訓練資料
C正解 Correct Answer
Naming a likely failure and a fix (human review) is good design.
事先指出可能的失敗情境並提出解法(人工複核),是良好的設計方式。
LO: UP.P4.2, UP.P4.3
Q102 分
You use AI to make a picture, then enter it in an art contest as if you drew it yourself. This is mainly a problem of…
你用AI畫了一張圖,然後拿去參加美術比賽,假裝是自己畫的。這主要是什麼問題?
A
the image resolution chosen
選擇的圖片解析度
B
the colour balance settings
色彩平衡的設定
C
honesty and giving credit
誠實與歸功於原作者的問題
D
the saved file format
儲存的檔案格式
C正解 Correct Answer
Being honest about AI’s help matters. Claiming AI work as fully your own is unfair to others.
誠實說明AI提供的協助很重要。把AI的成果完全說成是自己的,對其他人並不公平。
LO: UP.P5.3
Q112 分
These examples follow one rule. Work out the rule, then give the output for 10.
這些範例遵循同一條規則。找出這條規則,並算出輸入為10時的輸出。
2 → 3 3 → 5 4 → 7 10 → ?
2 → 3 3 → 5 4 → 7 10 → ?
A
19
B
21
C
20
D
11
A正解 Correct Answer
Each output is double the input minus one (2→3, 3→5, 4→7), so 10→19. Inferring a rule from examples and applying it to a new input is the core idea behind machine learning.
2 + 1 = 3. Adding up rule points step by step is how a simple AI scores things.
2+1=3。一步一步把規則的分數加起來,正是簡單AI評分的方式。
LO: UP.P2.2
Q132 分
An AI agent is different from a plain chatbot because it can…
AI代理人和一般聊天機器人不同,因為牠可以……
A
store far more of your past conversations
儲存多得多的過往對話紀錄
B
reply with much longer, more detailed messages
回覆更長、更詳細的訊息
C
run without needing any electricity at all
完全不需要用電就能運作
D
plan steps and use tools to reach a goal
規劃步驟、使用工具來達成目標
D正解 Correct Answer
An agent plans, takes actions and can use tools (like searching or a calculator) to finish a task.
代理人會規劃、採取行動,並能使用工具(例如搜尋或計算機)來完成任務。
LO: UP.P4.1
Q142 分
Which is the SAFEST thing to paste into a public chatbot?
貼進公開聊天機器人裡「最安全」的內容是哪一個?
A
Your friend’s home address
朋友家的地址
B
A made-up practice question
一道自己編的練習題
C
A family member’s password
家人的密碼
D
A classmate’s medical note
同學的病歷資料
B正解 Correct Answer
Never share private information about yourself or others. Made-up practice content is fine.
絕對不要分享自己或他人的私人資訊。自己編的練習內容則沒有問題。
LO: UP.P5.2
Q152 分
Before trusting an important fact from AI, you should…
在相信AI提供的重要事實之前,你應該……
A
repeat it so you remember it
重複唸幾次好記住它
B
share it with your classmates
分享給同學
C
check it against a reliable source
對照可靠的來源加以查證
D
save it as a screenshot
把它截圖存起來
C正解 Correct Answer
AI can be wrong, so verify important facts with a trusted source.
AI可能會出錯,所以重要的事實要用可信的來源查證。
LO: UP.P5.2
Q162 分
What is the difference between classifying and generating?
分類(classifying)和生成(generating)有什麼不同?
A
Classifying sorts items; generating makes new content
分類是把東西歸類;生成是創造新的內容
B
Both simply mean creating brand-new pictures
兩者都只是指創造全新的圖片
C
Classifying writes text while generating deletes it
分類是寫文字,生成是刪除文字
D
They are two different names for the same thing
兩者其實是同一件事的不同名稱
A正解 Correct Answer
Classification labels/sorts existing things; generation creates new things.
分類是替既有的東西加標籤/歸類;生成則是創造出新的東西。
LO: UP.P2.2
Q172 分
Which prompt is likely to give the BEST result?
哪一個提示詞(prompt)最可能得到最好的結果?
A
“Write a poem about an animal.”
「寫一首關於動物的詩。」
B
“Make a cat poem, any length is fine.”
「寫一首貓的詩,長度不拘。」
C
“Do a nice poem for me, please and thanks.”
「幫我寫首好詩,拜託謝謝。」
D
“Write a funny 4-line poem about a sleepy cat.”
「寫一首關於一隻愛睏的貓、有趣的四行詩。」
D正解 Correct Answer
A specific, clear prompt (topic, length, style, audience) gives better generative results than a vague one.
具體清楚的提示詞(主題、長度、風格、對象)會比模糊的提示詞得到更好的生成結果。
LO: UP.P3.1, UP.P3.2
Q182 分
Some artists are upset that AI learned from their drawings without asking. This is a debate about…
有些藝術家不滿AI在未經同意的情況下學習了他們的畫作。這是關於什麼的爭議?
A
fairness and consent for training data
訓練資料的公平性與同意權
B
the layout of the app's menus
應用程式選單的排版
C
the speed of the internet
網路的速度
D
which file format to use
該使用哪種檔案格式
A正解 Correct Answer
Who owns the data AI learns from, and whether permission was given, is a real ethical debate.
AI學習所用的資料屬於誰、是否經過同意,是一個真實存在的倫理爭議。
LO: UP.P2.3
Q192 分
Does a chatbot truly “understand” your words the way a human friend does?
聊天機器人真的能像人類朋友一樣「理解」你說的話嗎?
A
Yes, exactly like a human friend does
會,和人類朋友一模一樣
B
Yes, it genuinely feels real emotions
會,它真的會感受到真實的情緒
C
No, it only predicts language patterns
不會,它只是在預測語言的模式
D
Only for very simple, everyday questions
只有非常簡單的日常問題才會
C正解 Correct Answer
AI matches patterns in language. It does not understand or feel as people do.
AI是在比對語言中的模式,並不像人類一樣真正理解或感受。
LO: UP.P3.1
Q202 分
Which of these is a real-world cost of training a big AI model?
下列哪一項是訓練大型AI模型在現實世界中要付出的成本?
A
it wears out the user's keyboard
會磨損使用者的鍵盤
B
it uses large amounts of electricity and water
會消耗大量的電力與水資源
C
it uses up the internet's words
會把網路上的文字用光
D
it makes other computers run slower worldwide
會讓全世界其他電腦變慢
B正解 Correct Answer
Training and serving big models runs on data centres that draw electricity and cooling water - a real cost someone pays.
訓練與運行大型模型都依靠資料中心,而資料中心需要消耗電力與冷卻用水——這是有人要真實承擔的成本。
LO: UP.P5.4
Q212 分
You read an amazing online story “written by a student.” A clue it might be AI-made is that it…
你在網路上讀到一篇「由學生撰寫」的精彩故事。它可能是AI寫的線索是……
A
was posted on the site fairly recently
是最近才貼上網站的
B
is neatly divided into several paragraphs
整齊地分成好幾段
C
has a clear, descriptive title at the top
開頭有一個清楚、有描述性的標題
D
is polished yet gets simple facts wrong
文筆流暢,卻把簡單的事實弄錯
D正解 Correct Answer
AI text can read smoothly yet contain made-up facts or a generic tone - stay curious and check.
AI寫的文字讀起來可能很流暢,卻可能包含捏造的事實或千篇一律的語氣——保持好奇心並加以查證。
LO: UP.P5.1
Q222 分
A tiny “points machine” decides if an email is spam. +3 if it says “free prize”, +2 if it has many links, −1 if from a known friend. Email: “FREE PRIZE!! click these links” from a stranger. Score?
RULES: +3 “free prize” · +2 many links · −1 known friend
規則:+3「免費獎品」・+2很多連結・−1認識的朋友
A
4
B
5
C
3
D
6
B正解 Correct Answer
3 + 2 = 5. A higher score means “more likely spam.” You computed a simple model yourself.
3+2=5。分數越高代表「越可能是垃圾郵件」。你剛剛親手計算了一個簡單的模型。
LO: UP.P2.2
Q232 分
You have checked carefully and are now SURE a shocking video of a classmate is an AI-made fake. The responsible next step is to…
你已經仔細查證過,確定一支關於同學的震驚影片是AI做出來的假影片。負責任的下一步是……
A
repost it with a warning caption
加上警告文字後轉發出去
B
report it to a trusted adult or the platform
通報給信任的大人或該平台
C
keep a copy to show friends later
留一份備份,以後給朋友看
D
do nothing - fakes are harmless
什麼都不做——假影片沒有危害
B正解 Correct Answer
When you find a harmful fake, report it so it can be taken down. Spreading it further - even with a warning - still increases the harm.
發現有害的假內容時,要通報讓它被下架。就算加上警告文字繼續散播,仍然會擴大傷害。
LO: UP.P5.1, UP.P5.2
Q242 分
Generative AI can also help programmers by…
生成式AI也能這樣幫助程式設計師……
A
guaranteeing the code has no mistakes
保證程式碼完全沒有錯誤
B
finding every bug with no human checking
找出所有錯誤,完全不需要人工檢查
C
drafting code for a person to check and test
起草程式碼草稿,讓人來檢查與測試
D
running finished code faster
讓寫好的程式跑得更快
C正解 Correct Answer
AI can draft code, but humans must still check and test it.
AI可以起草程式碼,但人類仍然必須檢查與測試它。
LO: UP.P3.4, UP.P3.1
Q252 分
When a voice assistant hears you speak, turning the sound into words it can work with happens in its…
當語音助理聽到你說話,把聲音轉換成牠可以處理的文字,這發生在牠的哪個步驟?
A
acting step
行動步驟
B
sensing step
感知步驟
C
reward step
獎勵步驟
D
sleeping step
休眠步驟
B正解 Correct Answer
Hearing and converting your speech is the agent's sense step - input first, then think, then act.
聽到並轉換你的語音,是代理人的「感知」步驟——先輸入,再思考,最後行動。
LO: UP.P4.1
Q262 分
Who often prepares the labelled examples that AI learns from?
通常是誰在準備AI學習所需的「已標記範例」?
A
the model invents all labels itself
模型自己發明所有標籤
B
people who label the data manually
由人工替資料貼上標籤
C
labels are copied automatically online
標籤是從網路上自動複製來的
D
no labels are needed at all
完全不需要標籤
B正解 Correct Answer
People label data (e.g., “this is a cat”), so human choices shape the AI.
人類會替資料貼標籤(例如「這是一隻貓」),所以人類的選擇會形塑AI。
LO: UP.P1.3, UP.P2.3
Q272 分
Your friend wants to ask a chatbot for medical advice about a real illness. The wisest thing is to…
朋友想問聊天機器人關於真實疾病的醫療建議。最明智的做法是……
A
fully trust the chatbot’s diagnosis
完全相信聊天機器人的診斷
B
see a doctor; AI is not a substitute
去看醫生;AI無法取代醫生
C
ask it exactly which medicine to take
直接問它該吃哪種藥
D
just wait and see if it passes on its own
什麼都不做,等它自己好
B正解 Correct Answer
For health, money or safety, rely on qualified people. AI can give wrong or unsafe answers.
涉及健康、金錢或安全的事,要依靠合格的專業人員。AI可能給出錯誤或不安全的答案。
LO: UP.P5.2
Q282 分
A GOOD use of AI that helps people is…
AI一個能幫助他人的「良好用途」是……
A
describing images for the blind
替視障者描述圖像內容
B
writing fake product reviews for money
為了錢寫假的商品評論
C
automatically liking all your own posts
自動幫自己的貼文按讚
D
secretly copying answers during a test
在考試中偷偷抄答案
A正解 Correct Answer
AI can improve accessibility, such as describing images for blind users - a positive use.
AI可以提升無障礙近用性,例如替視障使用者描述圖像——這是一種正向的用途。
LO: UP.P4.4
Q292 分
A face-recognition tool works well for some people but makes more mistakes for others. This shows the importance of…
一款人臉辨識工具對某些人效果很好,對另一些人卻容易出錯。這凸顯了什麼的重要性?
A
using much higher-resolution cameras
使用解析度更高的攝影機
B
training the staff to type much faster
訓練工作人員打字更快
C
adding several more display screens
增加更多顯示螢幕
D
testing AI for fairness
測試AI的公平性
D正解 Correct Answer
AI should be tested for fairness so it does not disadvantage some groups.
AI應該接受公平性測試,才不會讓某些群體處於不利地位。
LO: UP.P1.4, UP.P2.3
Q302 分
The most responsible way to use generative AI for schoolwork is to…
在課業上使用生成式AI最負責任的方式是……
A
hand in its output as if it were your own work
把它產出的內容當成自己的作業交出去
B
use it but never bother to fact-check what it says
使用它,但從不查證它說的內容
C
learn from it, then verify
從中學習,並加以查證
D
let it complete your entire exam for you
讓它幫你把整場考試都寫完
C正解 Correct Answer
Responsible use = learn, verify, and be honest about AI’s help.
負責任的使用方式=從中學習、加以查證,並誠實說明AI提供的協助。
LO: UP.P3.1, UP.P5.2
第二部分・結構化任務 Section B · Structured Tasks
Answer all parts. Show your working or reasoning where asked.
請作答所有小題,若題目要求請寫出計算過程或推理。
任務一・從基本原理打造一台「積分機器」20 分
Task 1 · Build a “Points Machine” from First Principles
A school wants a simple AI to flag possibly unkind messages for a teacher to check. The “points machine” gives each message a score: • +3 if it contains a mean name• +2 if it is in ALL CAPITAL LETTERS• +1 if it has 3 or more “!”• −2 if the sender often sends kind messages Flag a message if the score is 3 or more.
Score this message from a usually-kind sender: “YOU ARE A LOSER!!!” Show your working.
請為這則來自平常很友善的寄件人的訊息計分:「YOU ARE A LOSER!!!」請寫出計算過程。
b)
Should the message be flagged? Explain using your score.
這則訊息應該被標記嗎?請用你算出的分數說明。
c)
The machine flags a message that was actually a joke between best friends. Why can a simple rule machine make this mistake, and why should a human still check before anyone is punished?
Why this earns marks: Rewards executing a scoring model step by step and combining positive and negative weights correctly.
得分原因:獎勵能一步一步執行評分模型,並正確結合正負權重。
b)
4 marks: Yes - 4 is ≥ 3, so it is flagged for a teacher to look at.
4分:應該——4分≥3分,所以會被標記,交給老師查看。
Why this earns marks: Checks the learner applies the decision threshold, the core of a simple classifier.
得分原因:檢驗學生是否能套用判斷門檻,這正是簡單分類器的核心。
c)
8 marks: The machine only counts words/CAPS; it cannot understand tone, friendship or jokes (context). So it gives false positives. A human should make the final call so no one is unfairly punished by a rule that missed the meaning.
Why this earns marks: Surfaces the key dilemma: automated systems lack context and can be unfair, so meaningful human oversight protects people.
得分原因:點出關鍵的兩難:自動化系統缺乏脈絡判斷、可能不公平,因此有意義的人工監督能保護人們。
满分條件 For full marks
Full marks: correct score (a), correct flag decision (b), and a clear explanation of why human oversight matters (c).計分正確(a)、標記判斷正確(b),並清楚說明人工監督為何重要(c),即可獲得满分。
LO: a) UP.P2.2 · b) UP.P2.2 · c) UP.P2.2 / a) UP.P2.2・b) UP.P2.2・c) UP.P2.2
任務二・設計一個負責任的作業幫手代理人20 分
Task 2 · Design a Responsible Homework-Helper Agent
An AI agent can sense a request, plan steps, use tools (like a calculator or web search), and act to help reach a goal. Design a generative-AI “study buddy” agent whose goal is to help a Grade 5 student learn, not to do the work for them.
List two tools your agent could use and what each is for.
列出你的代理人可以使用的兩種工具,並說明各自的用途。
b)
Write the agent’s plan as 3–4 steps for helping with a maths question (sense → plan → act).
寫出這個代理人幫忙解一道數學題的3到4個步驟(感知→規劃→行動)。
c)
Dilemma: The student types, “Just give me all the answers to my graded test so I can copy them.” What should a responsible study-buddy do and say, and why?
4 marks: Any two sensible tools, e.g., a calculator (to check arithmetic) and a search/reference tool (to find an explanation or example). (2 marks each.)
Why this earns marks: Tests understanding that agents extend their abilities with external tools rather than ‘knowing everything’.
得分原因:測驗學生是否理解代理人是靠外部工具擴展能力,而不是「無所不知」。
b)
8 marks: e.g., 1) Sense/read the question; 2) Plan: break it into steps; 3) Act: guide the student with hints and one worked example; 4) Check the student’s answer and give feedback. (2 marks per clear stage.)
Why this earns marks: Rewards a genuine sense→plan→act loop aimed at teaching, not just outputting an answer.
得分原因:獎勵一個真正以教學為目的的感知→規劃→行動循環,而不是只給出答案。
c)
8 marks: It should decline to hand over graded-test answers and explain why (that would be cheating and would stop real learning). Instead it offers to teach the method, give practice questions, or hints. A good answer notes honesty/academic integrity and that the helper’s goal is learning.
Why this earns marks: A real dilemma between being ‘helpful’ and being ethical. Strong answers show the agent must refuse harm and redirect toward legitimate help - early alignment thinking.
Full marks: useful tools (a), a teaching-focused agent loop (b), and a principled refusal with a helpful alternative (c).提出有用的工具(a)、以教學為核心的代理人循環(b),以及有原則地拒絕並提供有幫助的替代方案(c),即可獲得满分。
LO: a) UP.P4.4 · b) UP.P4.4, UP.P4.1 · c) UP.P4.4, UP.P5.3 / a) UP.P4.4・b) UP.P4.4, UP.P4.1・c) UP.P4.4, UP.P5.3
國中 Grades 7–8
Structure: Section A has 20 multiple-choice questions, 2 marks each (40 marks). Section B has 3 structured tasks (60 marks). The answer key and marking guide begin on a new page after Section B.試卷結構:A部分為20題選擇題,每題2分(共40分)。B部分為3題結構化任務題(共60分)。答案與評分指南列於B部分之後的新頁。
第一部分・選擇題 Section A · Multiple-Choice
Choose the one best answer for each question. Each question is worth 2 marks.
每題請選出一個最佳答案,每題2分。
Q12 分
To a computer, a photo is really a grid of…
對電腦來說,一張照片其實是一個由……組成的網格。
A
numbers (one or more per pixel)
數字(每個像素一個或多個數字)
B
letters, one per object
字母,每個物件一個
C
short text labels describing it
描述它的簡短文字標籤
D
recorded sound waves
錄下的聲波
A正解 Correct Answer
An image is a grid of number-pixels - the start of computer vision.
一張影像就是一個由數字像素組成的網格——這是電腦視覺的起點。
LO: LS.P1.5
Q22 分
A model splits your sentence into sub-word pieces before processing. A practical consequence of working on these pieces (not letters) is that the model…
模型在處理你的句子前,會先把它拆成「子詞片段」。用這些片段(而非字母)來運作,實際上會造成模型……
A
can miscount the letters in a word
可能會數錯一個字裡有幾個字母
B
always spells perfectly
永遠都能完美拼字
C
reads one character at a time
一次只讀一個字元
D
cannot handle any new words
完全無法處理任何新字詞
A正解 Correct Answer
Working on tokens (sub-word pieces) explains why a model can miscount letters or mis-split rare words.
用詞元(子詞片段)來運作,說明了為什麼模型可能會數錯字母數量,或把罕見字詞拆分錯誤。
LO: LS.P1.1, LS.P1.2, LS.P1.3
Q32 分
Raising the temperature setting of a generative model usually makes its output…
提高生成模型的「溫度(temperature)」設定,通常會讓輸出變得……
A
more factually accurate
事實上更準確
B
strictly shorter in length
長度嚴格變短
C
more random and varied
更隨機、更多樣
D
faster to generate
生成速度更快
C正解 Correct Answer
Higher temperature increases randomness/creativity; lower temperature makes output more focused and predictable.
較高的溫度會增加隨機性/創意;較低的溫度則讓輸出更聚焦、更可預測。
LO: LS.P3.1
Q42 分
Word embeddings represent words as vectors so that…
詞嵌入(word embeddings)把文字表示成向量,目的是讓……
A
every single word receives an identical vector
每一個字都得到完全相同的向量
B
words get directly converted into images
文字直接被轉換成圖片
C
vectors are always limited to two numbers each
每個向量永遠只限定兩個數字
D
words with similar meanings sit close together
意思相近的字彼此靠得很近
D正解 Correct Answer
Embeddings place similar meanings near each other, letting models measure how related words are.
詞嵌入會把意思相近的字放得很近,讓模型能衡量字詞之間的相關程度。
LO: LS.P1.4
Q52 分
In supervised learning, the training data includes…
Count the pairs: “we”→play appears 2 times, “we”→win once and “we”→sing once. A bigram model predicts the most frequent next word - play (2 of 4). You just generated text the way a tiny language model does.
Answer all parts. Show your working or reasoning where asked.
請作答所有小題,若題目要求請寫出計算過程或推理。
任務一・追蹤一個神經元與一次決策(基本原理)20 分
Task 1 · Trace a Neuron and a Decision (First Principles)
A tiny model decides whether to recommend a video to a user. It is a single neuron: z = (w₁·x₁) + (w₂·x₂) + b, then it applies a step activation: recommend if z > 0, else skip.(A step activation is just an on/off switch. First work out the score z. If z is bigger than 0, the neuron switches ON → recommend. If not, it stays OFF → skip. Nothing in between.) Features: x₁ = “matches user’s topic” (1 yes / 0 no), x₂ = “video length in minutes ÷ 10”. Weights: w₁ = 4, w₂ = −1, bias b = 1.
Why this earns marks: Confirms the learner handles a 0 feature and decimals correctly.
得分原因:確認學生能正確處理數值為0的特徵以及小數計算。
c)
4 marks: A negative w₂ means longer videos lower the score, so the model is biased toward shorter videos.
4分:負的w₂代表影片越長,分數越低,所以這個模型偏向較短的影片。
Why this earns marks: Tests interpretation of weight sign - connecting maths to model behaviour.
得分原因:測驗學生是否能解讀權重的正負號——把數學和模型的行為連結起來。
d)
4 marks: Risk: it ignores quality/accuracy and can create filter bubbles or push only short content. Improvement: add more features, use human feedback, or include a diversity term.
Why this earns marks: Links a simple model to real recommender-system harms and mitigation - early systems thinking.
得分原因:把一個簡單的模型連結到真實推薦系統可能造成的傷害與改善方式——這是系統性思考的初步練習。
满分條件 For full marks
Full marks: correct z and decisions (a,b), correct interpretation of the negative weight (c), and a sensible risk + improvement (d).z值與判斷都正確(a、b)、正確解讀負權重(c),並提出合理的風險與改善方式(d),即可獲得满分。
LO: a) LS.P2.1 · b) LS.P2.1 · c) LS.P2.1 · d) LS.P2.1, LS.P5.4 / a) LS.P2.1・b) LS.P2.1・c) LS.P2.1・d) LS.P2.1, LS.P5.4
任務二・寫出一個生成式AI學習代理人的程式(實作題)20 分
Task 2 · Program a Generative-AI Study Agent (Practical)
You will design an AI agent that helps a student revise. The agent can call these tools: search(query) → returns notes · quiz(topic) → returns a practice question · ask_LLM(prompt) → returns generated text. Write your answer as pseudocode (Python-like is fine).
Add a loop so the agent keeps quizzing until the student answers 3 questions correctly. Track the count.
加入一個迴圈,讓代理人持續出題,直到學生答對3題為止。請記錄答對的題數。
c)
Dilemma: The student types “just tell me the exam answers.” Add a check in your code that refuses this and explains why, while still offering legitimate help. Briefly justify your design choice.
Why this earns marks: Tests control flow and state - moving from a single call to an iterative agent loop.
得分原因:測驗控制流程與狀態管理——從單一次呼叫進展到會反覆執行的代理人循環。
c)
EN
6 marks: e.g.if "exam answers" in request: return "I can’t give graded-test answers - that’s cheating. Want practice questions or an explanation instead?"Justification: the agent should refuse harmful/unethical requests and redirect to legitimate help.
中文
6分:例如:
if "考試答案" in request:
return "我不能直接給你計分考試的答案——那樣算作弊。要不要改成練習題或講解?"
理由說明:代理人應該拒絕有害/不符倫理的請求,並引導到正當的協助。
Why this earns marks: Embeds a guardrail in code and asks the student to justify it - practical ethics and the start of alignment thinking.
得分原因:在程式碼中埋入防護機制,並要求學生說明理由——這是實務倫理與AI對齊思維的起點。
满分條件 For full marks
Full marks: working tool orchestration (a), a correct quiz loop with counter (b), and a coded refusal with justification (c).能正確調度工具(a)、有正確計數器的出題迴圈(b),以及有理由說明的程式化拒絕機制(c),即可獲得满分。
LO: a) LS.P3.1, LS.P4.3 · b) LS.P4.3 · c) LS.P4.3 / a) LS.P3.1, LS.P4.3・b) LS.P4.3・c) LS.P4.3
任務三・深偽影片的兩難(理論+倫理)20 分
Task 3 · The Deepfake Dilemma (Theory + Ethics)
A student newspaper receives a shocking AI-generated video appearing to show the school principal admitting to cheating in a competition. It could go viral before tomorrow’s big match. The editor must decide what to do tonight.
Identify three things that could go wrong if the paper publishes the video without checking.
指出如果報社在未經查證的情況下就刊登這支影片,可能會出什麼問題(三項)。
b)
Describe two practical methods the team could use to check whether the video is real.
描述兩種團隊可以用來查證這支影片是否為真的實用方法。
c)
The student who supplied the video says, “Even if it’s fake, it makes people talk about cheating, which is a good cause.” Evaluate this argument - is a good cause a valid reason to publish a possible deepfake? Argue your position.
State one school-wide policy you would recommend for handling AI-generated media, with a reason.
提出一項你會建議全校採用、用來處理AI生成媒體內容的政策,並說明理由。
a)
5 marks: Any three: defaming an innocent person; spreading misinformation; legal/disciplinary trouble; eroding trust in the paper; causing real harm before the truth emerges. (≈2 each, cap 5.)
Why this earns marks: Checks the student can foresee concrete harms of unverified generative media.
得分原因:檢驗學生是否能預見未經查證的生成式媒體內容可能造成的具體傷害。
b)
5 marks: Any two: check provenance/metadata or watermarks; look for visual/audio artefacts; seek the original source; ask the principal directly; use a detection tool; corroborate with other evidence. (≈2–3 each.)
Why this earns marks: Rewards practical media-verification literacy rather than vague ‘ask someone’.
得分原因:獎勵實用的媒體查證素養,而不是籠統地說「去問問看」。
c)
6 marks: A strong answer argues that ends do not justify spreading falsehood: publishing a likely fake harms a real person, deceives readers, and damages credibility - the ‘good cause’ can be pursued honestly instead. Credit is given for a clear position supported by reasons; acknowledging the counter-view strengthens it.
4 marks: e.g., ‘verify-before-publish with a second source and a label for any AI-generated content,’ with a reason such as protecting people and reader trust.
Why this earns marks: Moves from judgement to governance - designing a rule that prevents the harm.
得分原因:從個人判斷進一步走向治理層面——設計出一條能預防傷害的規則。
满分條件 For full marks
Full marks: foreseeable harms (a), real verification methods (b), a reasoned stance on the dilemma engaging the counter-argument (c), and a justified policy (d).指出可預見的傷害(a)、提出真正可行的查證方法(b)、對兩難情境提出有理由且回應反方觀點的立場(c),以及提出有理由支持的政策(d),即可獲得满分。
LO: a) LS.P5.2 · b) LS.P5.2 · c) LS.P5.2 · d) LS.P5.2 / a) LS.P5.2・b) LS.P5.2・c) LS.P5.2・d) LS.P5.2
高中職 Grades 9–12
Structure: Section A has 20 multiple-choice questions, 2 marks each (40 marks). Section B has 3 structured tasks (60 marks). The answer key and marking guide begin on a new page after Section B.試卷結構:A部分為20題選擇題,每題2分(共40分)。B部分為3題結構化任務題(共60分)。答案與評分指南列於B部分之後的新頁。
第一部分・選擇題 Section A · Multiple-Choice
Choose the one best answer for each question. Each question is worth 2 marks.
每題請選出一個最佳答案,每題2分。
Q12 分
In a transformer, self-attention computes how much each token should attend to others using…
The cosine similarity of orthogonal embeddings a=[1,0] and b=[0,1] is…
兩個互相正交的嵌入向量a=[1,0]與b=[0,1]之間的餘弦相似度是多少?
cos = (a·b) / (‖a‖‖b‖) = 0 / (1×1)
A
1
B
0
C
−1
D
2
B正解 Correct Answer
Their dot product is 0, so cosine similarity is 0 - the vectors are unrelated in direction.
它們的內積是0,所以餘弦相似度是0——代表這兩個向量的方向彼此無關。
LO: US.P1.1
Q52 分
Setting a model's sampling temperature very close to 0 makes its output…
把模型的取樣溫度設得非常接近0,會讓輸出變得……
A
always factually correct
永遠符合事實
B
much longer on average
平均長度變得長很多
C
nearly deterministic (the most likely tokens win almost every time)
幾乎變成決定性的(幾乎每次都選出機率最高的詞元)
D
unable to stop generating
無法停止生成
C正解 Correct Answer
Very low temperature makes sampling nearly greedy - the highest-probability token wins almost every time, so output becomes stable and repeatable. Note this makes it consistent, not necessarily true.
In a convolutional neural network (CNN) processing an image, the deeper layers typically learn to detect…
在處理影像的卷積神經網路(CNN)中,較深層通常會學會偵測……
A
simple local features such as edges and corners
簡單的局部特徵,例如邊緣和角點
B
whole objects such as faces and cars
完整的物件,例如人臉和汽車
C
the caption text attached to the image
附加在影像上的說明文字
D
the file format the image was saved in
影像儲存時使用的檔案格式
B正解 Correct Answer
A CNN reads the image as a grid of pixel numbers; early layers detect simple local patterns (edges, corners) and deeper layers combine them into parts and whole objects.
Fairness metrics quantify disparate outcomes so bias can be detected and addressed.
公平性指標能量化不同群體之間結果的落差,讓偏誤得以被偵測並加以處理。
LO: US.P1.3, US.P5.1, US.P5.4
Q152 分
LLMs can sometimes reproduce verbatim text or personal data from training. The main concern this raises is…
大型語言模型有時會逐字重現訓練資料中的文字或個人資料。這主要引發的疑慮是……
A
noticeably slower inference latency overall
整體推論延遲明顯變慢
B
privacy leakage from memorised data
因記憶下來的資料而導致隱私外洩
C
higher per-token serving costs
每個詞元的服務成本變高
D
a reduced effective context length
有效上下文長度變短
B正解 Correct Answer
Memorisation can leak private data; mitigations include de-duplication, filtering and differential privacy.
模型的記憶行為可能洩漏私人資料;緩解方式包括去重複、資料過濾與差分隱私技術。
LO: US.P1.2, US.P2.2, US.P5.3
Q162 分
Which is the STRONGEST way to reduce hallucinations in a deployed assistant?
對於已上線的助理來說,減少幻覺(hallucination)最有效的方式是哪一個?
A
ground answers in cited sources
讓答案依據有引用出處的資料來源
B
substantially increase the sampling temperature
大幅提高取樣溫度
C
always generate much longer responses
永遠生成長很多的回應
D
remove all of the system instructions
移除所有的系統指令
A正解 Correct Answer
Grounding + citations let users verify claims and reduce unsupported assertions.
依據可查核的資料來源作答,加上引用出處,能讓使用者自行查證內容,並減少沒有根據的論述。
LO: US.P4.3, US.P5.3
Q172 分
Generated images and text raise unsettled questions about…
AI生成的圖像與文字,引發了尚未有定論的爭議,是關於……
A
the screen resolution to view them
觀看時的螢幕解析度
B
which keyboard layout to use
該使用哪種鍵盤配置
C
how fast they download
下載速度有多快
D
copyright and ownership of data and outputs
訓練資料與生成結果的著作權與所有權
D正解 Correct Answer
IP and copyright of both training inputs and AI outputs are active legal and ethical debates.
訓練資料與AI生成結果的智慧財產權與著作權,都是目前仍在進行中的法律與倫理爭議。
LO: US.P5.2
Q182 分
Which statement about the environmental cost of large AI models is most accurate?
關於大型AI模型的環境成本,下列哪一項敘述最正確?
A
Serving millions of daily queries adds energy and water use on top of the one-off training cost
每天服務數百萬次查詢,會在一次性的訓練成本之外,額外增加能源與用水消耗
B
Once a model is trained, running it consumes no meaningful energy at all
模型訓練完成後,運行時幾乎不消耗任何有意義的能源
C
Data centres use no water, so only their electricity use matters in practice
資料中心不用水,實務上只需要考慮用電量
D
The environmental cost of a model depends only on how accurate it is
模型的環境成本只取決於它的準確度
A正解 Correct Answer
Both training and inference at scale consume energy, and data centres use water for cooling. Weighing these costs - and who bears them - against the benefits is part of a responsible deployment decision.
A system uses a planner agent, a researcher agent and a coder agent that pass work to each other. What is the main advantage of this multi-agent design over one big prompt?
Why this earns marks: Confirms the weighted-sum-of-values step that yields the attended representation.
得分原因:確認學生能算出「值的加權總和」這個步驟,這正是產生注意力表徵的關鍵。
d)
4 marks: Self-attention lets every token dynamically weight and incorporate information from all other tokens in context (e.g., resolving what a pronoun refers to), rather than treating each word independently or in fixed order.
Full marks: correct scores (a), correct ranking incl. tie (b), correct weighted sum = 20.0 (c), and a clear conceptual explanation (d).分數計算正確(a)、排序正確且處理好平手情況(b)、加權總和正確算出20.0(c),並提出清楚的概念性解釋(d),即可獲得满分。
LO: a) US.P3.1 · b) US.P3.1 · c) US.P3.1 · d) US.P3.1 / a) US.P3.1・b) US.P3.1・c) US.P3.1・d) US.P3.1
任務二・設計挑戰——具代理能力的研究助理20 分
Task 2 · Design Challenge - An Agentic Research Assistant
Design an autonomous AI agent that helps a journalist research and draft factual articles. It may use tools: web_search, read_url, summarise(LLM), save_draft. It should plan, act, verify, and know its limits. This is open-ended - there is no single right answer; you are marked on the quality of your design and reasoning.
Dilemma: The agent can finish a story faster by quoting an unverified but very plausible source that supports a popular narrative. As the designer, what should the agent be built to do, and how do you weigh speed/engagement against truth and harm?
6 marks: A clear loop: plan sub-questions → web_search → read_url → summarise → check/verify across sources → save_draft, iterating until evidence is sufficient. Credit a sensible decomposition and correct tool sequencing.
5 marks: Any two concrete guardrails: require ≥2 independent sources per claim; attach citations; flag low-confidence claims for human review; restrict to reputable domains. (≈2–3 each.)
Why this earns marks: Rewards specific, implementable safeguards tied to the failure being mitigated.
得分原因:獎勵具體、可實作,且能對應到所要緩解的失敗情境的防護措施。
c)
4 marks: e.g., prompt injection: treat fetched page text as untrusted, never execute embedded instructions, sandbox tool use, and validate that quotes exist in the cited source. Credit a named failure + a matching mitigation.
Why this earns marks: Tests security/robustness awareness specific to tool-using agents.
得分原因:測驗學生對於「會使用工具的代理人」在安全性/穩健性上的意識。
d)
5 marks: The agent should be built to refuse to publish unverified quotes - verification and avoiding harm outweigh speed/engagement; it should mark the claim as unconfirmed or seek another source. Strong answers explicitly weigh the trade-off and acknowledge the pull of engagement metrics while prioritising truth and accountability. Also credit any principled design that prevents unverified claims being presented as fact (e.g., publishing only with an explicit ‘unconfirmed’ label), provided the speed-vs-truth trade-off is engaged.
Why this earns marks: The core dilemma: optimisation pressure (speed/engagement) vs. integrity. Marks reward a reasoned, principled stance that engages the trade-off rather than dismissing it.
Full marks: coherent agent loop (a), two real guardrails (b), a named failure mode + containment (c), and a reasoned ethical stance on the dilemma (d).提出連貫的代理人循環(a)、兩個真正的防護機制(b)、指出具體的失敗模式與防堵方式(c),並對兩難情境提出有理有據的倫理立場(d),即可獲得满分。
LO: a) US.P4.2, US.P4.4 · b) US.P4.2, US.P4.3, US.P5.2 · c) US.P4.1, US.P4.2, US.P4.3 · d) US.P4.2, US.P4.3 / a) US.P4.2, US.P4.4・b) US.P4.2, US.P4.3, US.P5.2・c) US.P4.1, US.P4.2, US.P4.3・d) US.P4.2, US.P4.3
Implement the core of a small RAG pipeline in Python-like pseudocode. You are given query and document embeddings (lists of numbers) and a function generate(prompt).
Dilemma: A keyword guardrail that blocks ‘self-harm’ also blocks a student researching a mental-health essay (a false refusal). Briefly: how would you make the guardrail safer AND less censoring, and why does over-blocking also cause harm?
5 marks: def cosine(a,b):
dot = sum(x*y for x,y in zip(a,b))
na = sqrt(sum(x*x for x in a))
nb = sqrt(sum(y*y for y in b))
return dot/(na*nb)(Marks: correct dot product and normalisation.)
中文
5分:def cosine(a, b):
dot = sum(x*y for x, y in zip(a, b))
na = sqrt(sum(x*x for x in a))
nb = sqrt(sum(y*y for y in b))
return dot / (na*nb)
(評分依據:內積與正規化計算正確。)
Why this earns marks: Verifies the learner can implement the similarity metric that powers retrieval.
得分原因:驗證學生能實作出驅動檢索功能的相似度指標。
b)
EN
5 marks: def retrieve(q,docs,k):
scored=[(cosine(q,d['emb']),d) for d in docs]
scored.sort(reverse=True, key=lambda t:t[0])
return [d for _,d in scored[:k]](Marks: scores all docs, sorts, returns k.)
中文
5分:def retrieve(q, docs, k):
scored = [(cosine(q, d['emb']), d) for d in docs]
scored.sort(reverse=True, key=lambda t: t[0])
return [d for _, d in scored[:k]]
(評分依據:對所有文件計分、排序,並回傳前k筆。)
Why this earns marks: Tests ranking/selection logic - the retriever stage of RAG.
得分原因:測驗排序/篩選邏輯——這正是RAG流程中「檢索」的階段。
c)
EN
6 marks: def answer(query,q_emb,docs):
top = retrieve(q_emb,docs,3)
context = "\n".join(f"[{i}] {d['text']}" for i,d in enumerate(top))
prompt = ("Answer using ONLY the sources; cite [n].\n"
f"Sources:\n{context}\nQ: {query}")
return generate(prompt)(Marks: uses retrieve, builds grounded prompt, instructs citation / only-context.)
中文
6分:def answer(query, q_emb, docs):
top = retrieve(q_emb, docs, 3)
context = "\n".join(f"[{i}] {d['text']}" for i, d in enumerate(top))
prompt = ("請只根據以下資料來源作答,並標註引用編號[n]。\n"
f"資料來源:\n{context}\n問題:{query}")
return generate(prompt)
(評分依據:有使用retrieve、建構出有依據的提示詞、並要求引用出處/只能使用提供的內容作答。)
Why this earns marks: Assesses grounding the generator in retrieved context with citations - the anti-hallucination payoff of RAG.
4 marks: Use intent/context-aware classification instead of raw keyword blocking; distinguish research/help-seeking from harmful instructions; route sensitive queries to safe, supportive resources rather than a blank refusal; allow appeal. Over-blocking harms by denying legitimate help and information (access-to-information and wellbeing costs).
Why this earns marks: The dilemma weighs two harms - enabling danger vs. censoring legitimate need. Marks reward a nuanced design that reduces both, not a simple ‘block more’ answer.
Full marks: correct cosine (a), correct top-k retrieval (b), a grounded cited-answer function (c), and a nuanced guardrail design weighing both harms (d).cosine函式正確(a)、前k筆檢索正確(b)、能依據檢索內容並附引用出處作答的函式(c),以及能同時權衡兩種傷害的細膩防護設計(d),即可獲得满分。
LO: a) US.P1.1, US.P4.2 · b) US.P1.1, US.P2.3, US.P4.2 · c) US.P4.2 · d) US.P4.2 / a) US.P1.1, US.P4.2・b) US.P1.1, US.P2.3, US.P4.2・c) US.P4.2・d) US.P4.2