Learning Map | Crafting Clear Instructions

Strategy 1: Craft Clear Instructions - Add Details and Elaborate

In OpenAI's official documentation (Documentation - GPT best practices), several strategies are provided to help users better utilize ChatGPT.

Core Point: Large language models cannot "guess" your intentions. Providing detailed information in your queries improves the quality and richness of responses.

• Case Comparisons:
  • Generic Query: "List some thinking models." The response is often vague.
  • Detailed Query: Specify context (e.g., "for business analysis"), quantity (3 models), and known references (e.g., "SWOT"), leading to more precise responses.
  • When asking the model to help memorize an abstract concept, adding details like "for teaching" or "needs analogies to daily life" yields more tailored answers.
  • For questions about prompts, contextualizing them (e.g., "for social media writing" or "distinguish between beginner and advanced scenarios") ensures more relevant responses.

Strategy 2: Craft Clear Instructions - Role-Playing

Core Point: Assigning a role to the model provides a clear framework, enabling it to generate responses aligned with the role’s characteristics.

• Examples of Role Influence:
  • Role: "Education Expert" – Rewrites text in a通俗 (accessible) and structured manner for educational audiences.
  • Role: "Artist" – Rewrites text with creativity and openness.
  • Role: "Renowned Literary Master" – Rewrites text with literary elegance and refined vocabulary.
• Dialogue Comparison: For the same topic (e.g., "evaluating technological development"), different roles (e.g., "environmental activist" vs. "tech entrepreneur") yield vastly different perspectives.

Strategy 3: Craft Clear Instructions - The Critical Role of Delimiters

Core Point: Use delimiters to mark independent blocks of content and prevent the model from confusing different sections.

• Common Delimiters:
  • Triple quotes: """Content"""
  • XML tags: <citation>Content</citation>
  • Markdown code blocks: Content
  • Section titles (e.g., "Chapter 1"), or continuous symbols (e.g., --, ···).
• Use Cases:
  • Separate prompt settings from referenced text to avoid misinterpretation.
  • Isolate specific format requirements (e.g., "generate JSON structure") from other content to ensure accuracy.
  • Designate placeholders for user input (e.g., "Insert text here") to clarify processing scope.

Strategy 4: Craft Clear Instructions - The Power of Examples

Core Point: Providing examples constrains generated content, ensuring the model uses trusted information to formulate answers.

• Operation Method:
  • Supply reference text and a question, instructing the model to "Answer using only the provided document and cite relevant paragraphs," formatted as {"引用": "..."}.
• Effect:
  • The model rigorously follows examples, responding with "Insufficient information" if the document lacks relevant data, enhancing answer credibility.

Strategy 5: Craft Clear Instructions - Specify Task Steps

Core Point: Break down complex tasks into steps to improve reasoning reliability and traceability.

• Advantages:
  1. Clarity: Eliminates misunderstandings by defining each step explicitly.
  2. Structure: Ensures logical flow and task coherence.
  3. Efficiency & Monitoring: Facilitates efficient execution and progress tracking.
  4. Error Troubleshooting: Enables step-by-step analysis of issues.
• Case:
  • When calculating a complex ticket purchase plan, asking for the final result directly often leads to errors. Breaking it into steps ("Calculate total number of people → Filter ticket types → Compare prices") yields more accurate responses.

Strategy 6: Craft Clear Instructions - Specify Format/Length Requirements

Core Point: Request outputs in specific formats (e.g., word count, paragraphs, bullet points).

• Considerations:
  • Precise word count control is unstable for Chinese (due to token calculation), while constraining sentence/paragraph counts is more reliable.
  • ChatGPT 4 demonstrates better format stability than version 3.5.
• Use Cases:
  • Summarization: "Summarize the text in triple quotes into 2 paragraphs."
  • Bullet Point Summaries: "List key points in 3 bullet points."
  • Example: Requesting "a 5-sentence product introduction" prioritizes sentence count over word count.