LLM Prompt Engineering — Advanced Techniques

# Bad prompt
"Tell me about AI"

# Good prompt
"""Explain artificial intelligence in 2-3 paragraphs.
Focus on: definition, applications, limitations.
Use simple language suitable for a high school student."""

# Without CoT
prompt = "What is 15 * 8 * 3?"

# With CoT
prompt = """Let me think through this step by step:
15 * 8 = 120
120 * 3 = 360
What is 15 * 8 * 3?"""

prompt = """Classify sentiment as positive or negative.

Examples:
"I love this!" -> positive
"This is terrible" -> negative
"The product is okay" -> neutral

Now classify: "Amazing quality and fast shipping!"
"""

system_prompt = """You are an expert Python programmer.
Provide only working code.
Include docstrings for all functions.
Use modern Python 3.10+ features."""

user_prompt = "Write a function to check if a string is a palindrome"

prompt = """You are a venture capital investor with 20 years experience.
Evaluate this startup pitch: [pitch]
Provide feedback on market opportunity, team, and financial projections."""

prompt = """Extract information from this text and return as JSON.

JSON format:
{
  "name": "string",
  "age": "number",
  "skills": ["string", "string"],
}

Text: John is 30 years old. He knows Python and JavaScript."""

prompt = """Write product description.
DON'T use clichés like "revolutionary" or "game-changing".
DON'T make unsupported claims.
DO focus on specific features and benefits."""

from openai import OpenAI

client = OpenAI()

# For deterministic outputs (classification, extraction)
response = client.chat.completions.create(
    model="gpt-4",
    messages=[{"role": "user", "content": "Classify this sentiment"}],
    temperature=0  # Most deterministic
)

# For creative outputs (writing, brainstorming)
response = client.chat.completions.create(
    model="gpt-4",
    messages=[{"role": "user", "content": "Write a story"}],
    temperature=1.0  # More creative
)

def refine_output(initial_prompt, feedback):
    """Refine output based on feedback."""
    refined_prompt = f"""{initial_prompt}

User feedback: {feedback}

Please revise your response addressing the feedback."""

    return refined_prompt

You are an expert in [domain].
[Task description]
[Specific requirements]

Situation: [Background]
Complication: [Problem]
Resolution needed: [Desired outcome]

Use this template for all responses:
[Structure defined]

Now: [Actual task]

def evaluate_prompt_quality(responses: list) -> dict:
    """Evaluate quality of responses."""
    return {
        "consistency": len(set(responses)) == 1,  # All same = good
        "length_variance": max(len(r) for r in responses) / min(len(r) for r in responses),
        "relevance": evaluate_relevance(responses)  # Custom function
    }