Note_Tech

All technological notes.

Project maintained by simonangel-fong Hosted on GitHub Pages — Theme by mattgraham

AI Fundamental - Prompt Engineering

Back

AI Fundamental - Prompt Engineering

Prompt Engineering

Prompt engineering
- the practice of designing and refining inputs (prompts) to guide an AI model (like an LLM) to produce accurate, relevant, and structured outputs.
It focuses on:
- Task Definition: Clearly defining the objective.
- Contextualization: Providing the right background information.
- Behavioral Control: Controlling the format and tone of the response.

Attention Mechanism

Attention Mechanism
- a machine learning technique that directs deep learning models to focus on the most relevant words when generating each token
- by assigning weights (importance) to different parts of the input
Important parts get more influence
- Clear instructions → higher attention → better results
- Noisy/long context → diluted attention → worse results
Position bias exists
- Tokens at:
  - Beginning
  - End
- tend to receive stronger attention
Long prompts can degrade quality
- Attention is spread across many tokens
- Model may:
  - Miss key info
  - Ignore earlier instructions
Structured prompts improve attention
- Good structure helps the model:
  - Locate key signals
  - Prioritize correctly

Common Techniques

Zero-shot Prompting
- a technique where a Large Language Model (LLM) is given a task without any prior examples, relying entirely on its pre-trained knowledge to generate a response.

Example: “Summarize the following text in one sentence: [Insert Text]”

One-shot prompting
- a technique where you provide a Large Language Model (LLM) with a single example (or “shot”) of a task within the prompt before asking it to perform a new, similar task.

Few-shot Prompting
- a prompt engineering technique that provides an AI model with a few examples (or “shots”) in the prompt to demonstrate the desired behavior before asking it to perform a new task.

Example: > Translate English to French:

Hello → Bonjour

Goodbye → Au revoir

Thank you →

Role/Instruction Prompting
- a technique where you assign a specific persona, profession, or character to an AI to guide its tone, expertise, and style.

Example: “You are a senior DevOps engineer. Explain CI/CD pipelines clearly.”

Step-by-step (Chain-of-Thought) Prompting
- a technique that improves large language model (LLM) reasoning by guiding it to generate intermediate, step-by-step logic before providing a final answer.

Example: “Solve the following math problem step by step:”

Output Formatting

Specifying the exact structure of the response.

Example: Explain Kubernetes in:

1 sentence summary

3 bullet points

Context Injection

Providing relevant data to improve accuracy and grounding.

Example: “Based on the following logs, identify the issue: <logs here>”

Constraint-based Prompting

Setting specific rules or limitations.

Example: “Answer in less than 50 words. Do not use technical jargon.”

Recommended Pattern

You are a senior DevOps engineer.          ← Role

Your task is to analyze a CI/CD failure.   ← Task

Context:
<logs / data here>                         ← Context

Instructions:
- Think step by step
- Identify root cause first                ← Reasoning guidance

Available tools:
- log_parser                              ← (optional)

Output format:
- Root Cause
- Fix                                     ← Structure

Constraints:
- Max 100 words
- Be concise                              ← Limits

Example

You are a senior DevOps engineer.

Your task is to analyze the following CI/CD pipeline failure, identify the root cause, and suggest a fix.

Context:
ERROR: Docker build failed
Step 3/5 : COPY requirements.txt .
COPY failed: file not found in build context

Instructions:
- Think step by step
- Identify the root cause first
- Then suggest the most likely fix

Available tools:
- None

Output format:
- Root Cause
- Fix

Constraints:
- Keep the explanation under 80 words
- Avoid unnecessary jargon