Hands-on with AI tools
AdvancedThe prompt decides the output. Four moves: be specific (topic, length, audience, format), assign a role, give a format or example, iterate with follow-ups. Text AI excels at explaining and rewriting, image AI at concept art. Output is a draft: verify facts, watch for hallucinations, never input personal data, and own the result.
- The prompt decides the output: the more specific the topic, length, audience and format, the better.
- Four moves: be specific, assign a role, give a format or example, iterate with follow-ups.
- Text AI for explaining and rewriting, image AI for concept art; pick tools per task.
文體小知識：方法指導課教你一套有系統的做法，照步驟練習，遇到問題就不會毫無頭緒。
§Text Passage
1Generative AI creates new content: text AIs write and answer, image AIs paint from a description, and there are music and video generators too. The instructions you type are the prompt — the language you speak to AI with, and its quality decides the output's quality. The same AI asked to 'write something about Tai O' versus 'write a 100-word tour-guide line about Tai O stilt houses for S3 students in a light tone' delivers worlds apart.
2Four prompt techniques: 1) be specific — state topic, length, audience, format; 2) assign a role — 'you are a science teacher explaining to S3 students…' tunes the wording and depth; 3) give an example or a format — 'three bullet points, one line each'; 4) iterate — unhappy with the first output? Do not quit; follow up with 'simpler, with a real-life example' and close in step by step. Good prompts are what make AI useful.
3Each tool has its strengths and blind spots: text AI (chatbots) excels at explaining, rewriting and brainstorming, but its maths and current-affairs facts can wobble; image AI produces illustration concepts fast but fumbles details — fingers and lettering go wrong constantly. Pick tools per task: the poster slogan comes from a text AI; the poster artwork is where an image AI steps in.
4The most important lesson is critical verification: AI output is always a draft. Check facts against textbooks and official sites, test whether the logic holds, stay alert for hallucinations — invented book titles, made-up events; never type personal data or passwords into AI tools; and the finished work is yours to answer for. AI is an assistant that amplifies your ability; it cannot replace your thinking — that sentence is the whole lesson.
1Summary 重點摘要
The prompt decides the output. Four moves: be specific (topic, length, audience, format), assign a role, give a format or example, iterate with follow-ups. Text AI excels at explaining and rewriting, image AI at concept art. Output is a draft: verify facts, watch for hallucinations, never input personal data, and own the result.
2Vocabulary cards Vocabulary
AI that creates new content — text, images, music, video — e.g. chatbots and text-to-image tools.
The text instruction given to generative AI; how specific it is directly shapes the output.
The craft of designing and refining prompts for better outputs — being specific, assigning roles, iterating.
AI tools generating images from text, e.g. 'Lion Rock at dusk, watercolour style'.
Repeated follow-ups and prompt tweaks, nudging the AI output toward the ideal step by step.
Verifying AI's facts and figures against reliable sources such as textbooks and official sites.
Knowing each AI tool's strengths and blind spots and choosing per task.
Treating AI output as a draft: check facts, test logic, watch for hallucinations before adopting.
Never type personal data, passwords or others' private details into AI tools — inputs may be retained.
3Step by step Steps
- Your first prompt: set a baselineA prompt is the text instruction given to generative AI, and how specific it is directly decides the output's quality. First experiment: type 'explain binary in one sentence' and the AI returns a generic definition. Inspect it closely — is the explanation clear, does it carry an example, does the wording suit your level? Keep it as your baseline version, then measure every later improvement against it.
- The specificity contrastRewrite the prompt as 'explain binary for an S3 student, within 50 words, with one everyday example', then compare the two outputs: the second has a concrete example, a controlled length and level-appropriate wording. Conclusion: same AI, different prompt, completely different quality. When writing prompts, state four things — the topic, the length, the audience, and any format or example required — and the output lands far closer to the mark.
- Assigning a roleAdding a role at the start of a prompt changes the angle and depth of the answer. Experiment: 'you are a basketball coach explaining to S3 players why knees bend before a shot' — the AI talks about stability and power from a coach's view; change it to 'you are a physics teacher' and the focus shifts to centre of gravity and mechanics. A role chooses the AI's knowledge frame, so for projects, appoint the professional role that matches the topic.
- Iterating with follow-upsGood output usually comes from narrowing round by round rather than one lucky shot. The routine: the first answer is too deep, follow with 'make it simpler'; still not right, add 'use cooking as the example'; the third version satisfies. Three exchanges beat one attempt, so when the first output disappoints, name the specific problem — too long, too deep, no example, no rhyme — and let the AI revise against it, raising the quality step by step.
- Text-to-image: the specificity testText-to-image tools generate pictures from descriptions, and the more concrete the description, the closer the picture. Prompt 'a painting' and the result is random and unusable; prompt 'Lion Rock at dusk, watercolour style, warm tones' and the image lands at once. Concrete description covers four dimensions: subject (what), style (watercolour, oil, cartoon), lighting (dusk, noon) and composition (wide shot, close-up). Still inspect the details — AI hands and lettering are often wrong.
- Choosing tools by taskEach AI tool has its own strength, and the wrong tool wastes effort. Scenario: the class committee is making a fair poster. Slogans and copy go to a text AI, which is strong at words; artwork concepts go to an image AI, which is strong at visuals; layout is finished in presentation or document software. Asking a text AI to draw, or an image AI to write an essay, only produces poor results. Before starting, ask: is the core of this task text, image or layout?
- The critical verification routineAI output is a draft, and every claim must be verified before adoption. Example one: the AI states 'Hong Kong became a SAR on 1 July 1997' — the textbook confirms it, so keep it. Example two: the AI dates a book to 1988, but the library catalogue says 1998 — a hallucination, discard it. Verify in three moves: check facts against reliable sources, test whether the logic holds, confirm no personal data was entered. With all three passed and the AI's role declared, the work is truly done.
4Diagrams Diagrams
5Worked example Example
Example: the class wants to promote the Christmas fair. Start from a vague prompt, improve it step by step into a good one, and verify the output.
- Prompt v1: 'write me some promotion' — the AI can only return generic filler, knowing nothing of the event, audience or requirements.
- Prompt v2 (specific): 'write three slogans for Class 3A's Christmas fair, 10 characters or fewer each, for the whole school, joyful mood' — the output lands at once, e.g. '3A Christmas fair, warm hands and hearts'.
- Prompt v3 (iterate): follow up 'make the second line rhyme better' — the AI polishes it; good prompts are usually ground out over three or four rounds.
- Verify the output: any typos? Do date and venue match the class committee's announcements? Was any student's name (personal data) typed in? Adopt only after all three checks pass.
- Answer: A good prompt = specific (event, count, length, audience, tone) + iterated + verified; a vague prompt buys only vague answers.
6Common mistakes Mistakes
7Quick check Quick check
Q1What is the text instruction typed into generative AI called?
Q2Which prompt is better? A. write something about the environment B. explain carbon neutrality to S3 students in 80 words with a school example
Q3The first output disappoints — what should you do?
Q4Which AI tool for the poster slogan, and which for the artwork?
Q5Name three checks before adopting AI output.
8Recap Recap
- The prompt decides the output: the more specific the topic, length, audience and format, the better.
- Four moves: be specific, assign a role, give a format or example, iterate with follow-ups.
- Text AI for explaining and rewriting, image AI for concept art; pick tools per task.
- AI output is a draft: verify facts, test logic, watch for hallucinations.
- Never type personal data or passwords into AI; the finished work is yours to answer for.