Slide 1: Title Slide
Slide 1: Title Slide
Title: "Prompt Engineering: Lecture and Hands-on"
Subtitle: Understanding and Practicing with Large Language Models
Your Name / Institution
Title: "Prompt Engineering: Lecture and Hands-on"
Subtitle: Understanding and Practicing with Large Language Models
Your Name / Institution
Slide 2: Introduction to Prompt Engineering
Title: "What is Prompt Engineering?"
Content:
Brief introduction on how prompt engineering improves the performance of LLMs.
Key objectives: clarity, context control, output specification.
Image: A simple diagram showing input-prompt-output flow for LLMs.
Slide 3: AUTOMAT Framework Overview
Title: "AUTOMAT Framework for Prompt Engineering"
Content:
Act as a...
User Persona & Audience
Targeted Action
Output Definition
Mode / Tonality / Style
Atypical Cases
Topic Whitelisting
Example: “Act as a medical expert to answer patient queries.”
Image: Flowchart of AUTOMAT with key elements(
Read Medium articles with AI
)(
Read Medium articles with AI
).
Slide 4: CO-STAR Framework
Title: "CO-STAR Framework"
Content:
Context
Objective
Style & Tone
Audience
Response Format
Example: A chatbot for e-commerce answering customer questions about a product's availability.
Image: CO-STAR framework example diagram(
KDnuggets
).
Slide 5: Few-Shot Learning Technique
Title: "Few-Shot Learning"
Content:
Providing examples within the prompt for the model to learn.
Standard and edge cases.
Example: "Translate the following sentences: 1. Bonjour -> Hello 2. Merci -> Thank you."
Image: Few-shot learning example with input-output flow(
Read Medium articles with AI
).
Slide 6: Chain of Thought (CoT) Prompting
Title: "Chain of Thought (CoT)"
Content:
Forces model to reason step-by-step.
Useful for logical or mathematical problems.
Example: “If the train travels 60 km in 1 hour, how far will it travel in 4 hours? (Explain step-by-step).”
Image: Flowchart showing CoT reasoning(
Read Medium articles with AI
).
Slide 7: Retrieval-Augmented Generation (RAG)
Title: "RAG: Retrieval-Augmented Generation"
Content:
Combining external data with model responses for up-to-date information.
Overcoming model knowledge limitations.
Example: "Generate a report on recent market trends by retrieving the latest data from an external source."
Image: Diagram illustrating RAG process(
Read Medium articles with AI
).
Slide 8: Hands-on Session Overview
Title: "Hands-on: Building Your Prompts"
Content: Overview of what will be practiced, including AUTOMAT, CoT, Few-Shot, and RAG.
Image: Example of a prompt structure being built.
Slide 9: Practice with AUTOMAT Framework
Title: "Practice: Using AUTOMAT Framework"
Content:
Task: Create a prompt using AUTOMAT for a chatbot designed to help with health-related queries.
Ask participants to define all seven elements of AUTOMAT for the chatbot.
Image: A sample of an AUTOMAT-based prompt output from a chatbot.
Slide 10: Practice with Few-Shot Learning
Title: "Practice: Few-Shot Learning"
Content:
Task: Provide 2-3 examples within a prompt for translation tasks or other language-related tasks.
Encourage participants to provide examples for normal and edge cases.
Image: A prompt with few-shot learning examples filled in.
Slide 11: Practice with Chain of Thought
Title: "Practice: Chain of Thought"
Content:
Task: Create a prompt that asks the model to solve a problem step-by-step, such as solving a math problem or analyzing a scenario.
Image: A flowchart showing the thought process of solving a math problem with CoT.
Slide 12: Practice with RAG
Title: "Practice: Retrieval-Augmented Generation"
Content:
Task: Create a prompt that retrieves specific external data, such as current news or market data, and then uses the model to generate a report based on the retrieved information.
Image: Example of an LLM generating a report from retrieved data.
Slide 13: Summary and Q&A
Title: "Key Takeaways"
Content:
Recap the techniques learned: AUTOMAT, CO-STAR, Few-shot, CoT, and RAG.
Encourage participants to ask questions or share their experiences from the hands-on activities.
Image: Summary table comparing all techniques discussed.