AIAI Agent Dev Labaiagentdev-lab.com
Sign in to save progressSign in
Menu

生成AIパスポート · LESSON 5

Chapter 5: Generative AI Practice and Use Cases19%（当サイト設定） · 6 topics

Learn prompt engineering and using generative AI in writing, planning, creative work, data analysis, development and work automation.

5.1Prompt engineering

Prompt engineering (basic components; zero-shot, few-shot, Chain-of-Thought; role setting and output format)

Concept

The basic components of a prompt are the instruction (what to do), context (background and purpose), input data and output format. Combine zero-shot (no examples), few-shot (a few examples), Chain-of-Thought (thinking step by step), role setting (“You are an expert in…”) and specifying output format such as tables, bullet points and length.

How it works / steps

  1. Clarify the purpose and readers
  2. Write needed background and conditions (length, tone)
  3. Show examples of expected output (few-shot)
  4. Check the output and improve the prompt

Workplace example

Give concrete instructions on readers, format and constraints: “For new hires, explain the expense claim steps in five bullet points without jargon.”

Common confusions and exam tips

Mini quiz

Q1. Which is the most appropriate combination of basic prompt components?

  1. Password, username, email address, phone number
  2. Font, color, margins, line spacing
  3. Model weights, learning rate, epochs, batch size
  4. Instruction, context (background and purpose), input data, output format
Show answer and explanation

Answer: D Instruction, context (background and purpose), input data, output format

Making these clear helps get output that matches your intent.

Q2. Which is the most appropriate view of “controlling specificity” in prompts?

  1. Convey purpose, target readers, constraints and evaluation criteria concretely, and improve interactively while viewing output
  2. Make prompts as short and vague as possible and let the AI interpret freely
  3. Never change a prompt once written; keep using it even if output is poor, to keep quality constant
  4. Always write explanations of 1,000 characters or more in prompts
Show answer and explanation

Answer: A Convey purpose, target readers, constraints and evaluation criteria concretely, and improve interactively while viewing output

Vague instructions tend to produce average output. Add needed information and improve by viewing output.

Reference links

5.2Use in writing and communication work

Concept

Generative AI helps with drafting email, reports and minutes, summarizing, rephrasing, translating and proofreading. Specify the recipient, purpose and tone, and check facts and figures yourself. For external documents, watch for entering confidential information and the risk of sending errors.

How it works / steps

  1. Give the key points and recipient and have it draft
  2. Check facts, figures and proper nouns
  3. Finalize in your own words before sending

Workplace example

From a meeting recording transcript, have it make minutes tabulating decisions, owners and deadlines, and share after participants confirm.

Common confusions and exam tips

Mini quiz

Q1. Which is the most appropriate precaution when using generative AI to write business email?

  1. Have AI rewrite signatures and greetings every time
  2. Check only that the honorific language is polite
  3. Check facts like recipients, dates and amounts against the source, and that no confidential data was entered
  4. Enter detailed recipient data to get a more accurate draft
Show answer and explanation

Answer: C Check facts like recipients, dates and amounts against the source, and that no confidential data was entered

It is good at drafting text, but people check facts and information handling. [Other options] Entering much personal data raises handling issues. Checking politeness alone misses factual errors. Fixed parts need not change.

Q2. Which is the most appropriate precaution when making meeting minutes with generative AI?

  1. Inform participants about using recordings and transcription, and have participants check decisions, owners and deadlines
  2. Send the generated minutes outside as-is without participants checking
  3. Have generative AI fill in things nobody said to make the minutes more readable
  4. Record without telling participants and share only the minutes made from the transcript
Show answer and explanation

Answer: A Inform participants about using recordings and transcription, and have participants check decisions, owners and deadlines

Consent to recording and checks that keep wrong records from spreading are important.

Reference links

5.3Use in planning and marketing work

Concept

In planning and marketing, generative AI can be used for brainstorming, creating personas, tagline ideas, ad copy variations and organizing competitor information. Having it produce many ideas for people to choose and refine is effective. Watch for advertising regulations such as Japan's Act against Unjustifiable Premiums and Misleading Representations, and for claims not based on fact.

How it works / steps

  1. Give the goal and target and have it produce many ideas
  2. People evaluate, select and refine
  3. Check that ad wording doesn't violate laws or facts

Workplace example

Have it produce 30 taglines for a new product, narrow them to 3 as a team and compare effects with an A/B test.

Common confusions and exam tips

Mini quiz

Q1. Which is the most appropriate precaution when using generative AI to write taglines and ad copy?

  1. Ideas from generative AI are original, so use them as-is without checking similarity to other companies
  2. Have it produce many ideas for people to choose from, and check for similarity to other companies' copy or trademarks and for exaggeration
  3. To boost effect, allow untrue or exaggerated expressions in ads
  4. Adopt the first idea and judge wording by the response after the ad is published
Show answer and explanation

Answer: B Have it produce many ideas for people to choose from, and check for similarity to other companies' copy or trademarks and for exaggeration

Ad wording must also respect regulations such as Japan's Act against Unjustifiable Premiums and Misleading Representations.

Q2. Marketing is about to publish a product description made with generative AI on the website. It says “industry No. 1 performance.” Which is the most appropriate next step?

  1. Check whether there is objective data supporting “industry No. 1,” and revise the wording if not
  2. Adding “written by AI” at the end lets you publish as-is without checking the basis
  3. To stand out more than competitors, strengthen it to “world No. 1 performance”
  4. Generative AI wrote it from lots of information, so the basis need not be checked
Show answer and explanation

Answer: A Check whether there is objective data supporting “industry No. 1,” and revise the wording if not

Unsupported superlatives risk misleading representation. Check the facts and appropriateness of wording.

Reference links

5.4Use in creative production

Concept

In producing images, video, music and design, generative AI can be used for rough drafts, generating variations and editing material. Watch rights (copyright, trademark, portrait rights) and terms of use, AI-generated labeling and provenance information (C2PA), and have people finish the work to ensure quality.

How it works / steps

  1. Specify the image, use and style of what you want concretely
  2. People select generated drafts, revise and finish them
  3. Check that it doesn't resemble existing works or real people

Workplace example

Make 10 rough website banner drafts with generative AI; a designer picks one and finishes it to brand guidelines.

Common confusions and exam tips

Mini quiz

Q1. Which is the most appropriate way to use generative AI to build a landing page (LP)?

  1. Leave handling and storage settings for personal information collected in forms to the generated code and publish
  2. Generated code is fine if it runs, so publish it to the production server without checking
  3. Enter the text and design of a competitor's LP as-is and have it make a page with the same structure and wording
  4. Have it draft structure, text and code, then publish (deploy) after checking display, facts and wording, and security
Show answer and explanation

Answer: D Have it draft structure, text and code, then publish (deploy) after checking display, facts and wording, and security

Both the code and the page content need checking. If collecting personal information, also set a handling policy.

Q2. Which is the most appropriate precaution when using generative AI to create variations of existing materials or localize them into multiple languages?

  1. More variations are better, so publish all without checking, not even asking each country's staff
  2. Check whether use is within the rights to the original material, and check each language's wording and cultural appropriateness
  3. They can be freely altered and distributed regardless of the original material's rights holder
  4. Translate mechanically with no local review
Show answer and explanation

Answer: B Check whether use is within the rights to the original material, and check each language's wording and cultural appropriateness

Check whether alterations (adaptations) are within rights and license terms.

Reference links

5.5Use in data analysis and research work

Concept

Generative AI helps aggregate tabular data and make charts (code execution features), suggest angles for analysis, classify free-text survey answers and organize literature and web research. Check calculation results and sources, and be careful handling confidential data.

How it works / steps

  1. Explain the purpose of the analysis and what the data columns mean
  2. Have it aggregate or chart, and verify results against the original data
  3. Open research sources and check their content

Workplace example

Classify 500 free-text survey answers by theme with generative AI; starting with the most common themes, the person in charge reads actual answers and considers improvements.

Common confusions and exam tips

Mini quiz

Q1. Which is the most appropriate way to use generative AI in survey design and analysis?

  1. Use it to draft questions, check for biased choices and classify free text; double-check tallies and let people interpret
  2. Since responses take time to collect, have generative AI make fictitious responses and add them to the tally
  3. Generative AI calculated the tallies, so report them as-is to the management meeting without checking
  4. Enter all personal information such as respondents' names and affiliations and have it rate each respondent
Show answer and explanation

Answer: A Use it to draft questions, check for biased choices and classify free text; double-check tallies and let people interpret

Treating fictitious data as real data is wrong. Also be careful with personal information.

Q2. Which is the most appropriate precaution when using generative AI for market and competitor research?

  1. Use its results as leads and back up key figures and facts with primary sources such as company announcements or public statistics
  2. Its results are based on the latest information, so use them in reports as-is without checking sources
  3. Even with no source given, treat figures as correct if they are specific and the text is consistent
  4. Have it guess competitors' non-public information and treat specific figures it produces as fact
Show answer and explanation

Answer: A Use its results as leads and back up key figures and facts with primary sources such as company announcements or public statistics

Research results can mix in hallucinations and outdated information.

Reference links

5.6Use in development and work automation, and by industry

Use in development and work automation, and use by industry

Concept

In development, it helps generate, explain and review code and write tests; in work automation, routine tasks (classifying inquiries, data entry, creating reports) can be automated by combining generative AI with tools. By industry, uses include healthcare (document writing support), finance (inquiry handling, assisting screening), manufacturing (searching maintenance procedures) and retail (writing product descriptions).

How it works / steps

  1. Look for automation candidates among highly repetitive work
  2. Pilot small and measure effects and error rates
  3. Check industry regulations (healthcare, finance, etc.) and keep human check steps

Workplace example

A call center automates call summaries and classification with generative AI, cutting operator after-call work by 3 minutes per call.

Common confusions and exam tips

Mini quiz

Q1. Which is the most appropriate way to proceed when automating routine work with generative AI or agents?

  1. Give only the goal and leave all step decisions and execution to the agent
  2. Automate all target work at once with no human checks
  3. Retry the same step automatically until it succeeds
  4. Define steps, start small, build in error handling and human checks, then expand
Show answer and explanation

Answer: D Define steps, start small, build in error handling and human checks, then expand

Task decomposition, permissions and check design determine automation quality. [Other options] A goal alone gives unstable results. Unlimited retries repeat failures. Automating everything at once magnifies problems.

Q2. Which is most important when automating complex tasks spanning multiple tools and steps?

  1. Don't record intermediate results; check only whether the final result is correct
  2. To save setup effort, share the same admin account across all tools
  3. Narrow each step's permissions, record intermediate results and add checks before important steps
  4. More steps take more time, so skip intermediate checks and prioritize speed
Show answer and explanation

Answer: C Narrow each step's permissions, record intermediate results and add checks before important steps

The more steps, the higher the risk of cascading errors.

Reference links

Lessons are written by this site based on public information and do not guarantee exam content. Product specifications change, so check the latest information at the reference links.

Notes (unofficial, original questions, no pass guarantee)