Introduction: AI terms become much easier once you clear the 'katakana-English barrier'
As you follow generative AI topics, terms like Prompt, RAG, Fine-Tuning, and Agent appear one after another. They can be intimidating at first, but they are all terms that describe how you want AI to work—i.e., the roles and responsibilities.
This article organizes the important words you commonly use in the field in as approachable a way as possible, in the order: In a word, When to use, Benefits and caveats, How to differentiate.
Prompt (Prompt): The AI's "request"
In a word
Prompts are instruction sentences that tell a generative AI what you want it to do and how. They serve as the input for the AI and are a key factor that determines the quality of the output.
When to use?
- Text creation (emails, proposals, articles, summaries)
- Code generation (creating functions, debugging, reviews)
- Image generation (composition and style specification)
- Data organization (table formatting, classification, extraction)
Tips: Prompts become stronger when you include a "Role", "Constraints", and "Examples"
The three standard patterns that tend to work well are:
- Role: You are an expert in XX; please answer as such
- Constraints: character count, tone, output format (JSON/table/bullets)
- Examples: good examples, bad examples, sample of the expected output
Example: You are the PM for a B2B SaaS company. Please present three new feature ideas, each with its aim, target user, and KPIs, in bullet points. Briefly explain any jargon.
Notes
- Ambiguous instructions tend to yield ambiguous answers (the 'feels right' problem)
- Lack of shared assumptions leads to misalignment (target, purpose, prohibitions)
- Be mindful of information management, such as avoiding confidential information
RAG (Retrieval-Augmented Generation): A system for adding evidence by searching
In a word
RAG is a mechanism where, before the AI generates an answer, it searches for relevant information from internal documents or databases (Retrieval) and then generates (Generation) based on that content. In short, it’s like letting the AI consult your internal knowledge.
When to use?
- Chatbots that rely on internal rules, procedures, FAQs, and meeting notes
- Support responses while referencing product manuals
- Drafts of sales materials and proposals (referencing past cases)
- Initial checks for legal/compliance (with internal rules cited)
