Mastering Prompt Engineering: Techniques, Tips, and Real-World Examples

Background

Prompt engineering is a relatively new discipline that develops and optimizes prompts to effectively use large language models (LLMs) for various applications and research topics. It helps researchers improve LLM performance on tasks such as question answering and arithmetic reasoning, and enables developers to design powerful interactions between LLMs and tools.

Fundamental Knowledge

LLMs assume different roles in different scenarios, requiring tailored prompts. Prompt sensitivity varies across models, so prompts must be adapted to the specific model and task. The evolution of NLP can be seen as moving from feature engineering to structural engineering and finally to prompt engineering.

Temperature : Lower values produce more deterministic outputs; higher values increase randomness and creativity.

Top_p : Controls nucleus sampling; lower values favor accurate, factual answers, while higher values encourage diverse responses.

Prompt Elements

Instruction : The specific task you want the model to perform.

Context : External information that guides the model.

Input Data : The question or data you want answered.

Output Indicator : The desired format of the response.

Construction Methods & Tips

1. Write clear and specific instructions

Use delimiters to separate different parts of the input, preventing prompt injection. Example:

text = f"""You should express what you want a model to do by providing instructions that are as clear and specific as possible."""
prompt = f"""Summarize the text delimited by triple backticks into a single sentence.
```{text}```"""
response = get_completion(prompt)
print(response)

2. Ask for structured output

Request JSON or other structured formats so the result can be directly consumed by programs.

prompt = f"""Generate a list of three made‑up book titles with their authors and genres. Provide them in JSON format with keys: book_id, title, author, genre."""
response = get_completion(prompt)
print(response)

3. Let the model think step‑by‑step (Chain‑of‑Thought)

Break complex tasks into sequential sub‑tasks, allowing the model to reason before answering.

text = f"""In a charming village, siblings Jack and Jill ..."""
prompt = f"""Perform the following actions:
1. Summarize the text delimited by triple backticks in one sentence.
2. Translate the summary into French.
3. List each name in the French summary.
4. Output a JSON object with keys french_summary and num_names.
```{text}```"""
response = get_completion(prompt)
print(response)

4. Mitigate hallucinations

Ask the model to first retrieve relevant information and then answer based on that information.

Advanced Prompt Types

Zero‑shot prompting : Directly ask the model without examples.

Few‑shot prompting : Provide a few examples to guide the model.

Zero‑shot Chain‑of‑Thought (CoT) : Add “Let’s think step by step” to zero‑shot prompts.

Program‑Aided Language Models (PAL) : Generate code as intermediate reasoning steps.

Practical Applications

Examples include classification, translation, domain‑specific Q&A, and independent question rewriting. Prompts are adapted for general tasks and business‑specific scenarios such as local‑life recommendation systems.

Evaluation Metrics

Prompt quality can be assessed from dimensions such as correctness, relevance, robustness, and efficiency, following guidelines from public courses and research papers.

Conclusion & Outlook

Prompt engineering still relies heavily on manual design and debugging, though automated approaches like Automatic Prompt Engineers (APE) are emerging. The article invites further discussion and experimentation.

References

Prompt‑Engineering‑Guide (GitHub)

Learn Prompting

Wu En Da’s Prompt Engineering Course

Various Chinese and English prompt guides