Prompt Engineering Course by Anu Kubo

Course Overview

• Instructor: Anu Kubo
• Focus on prompt engineering for generating effective responses from AI/LLMs.
• Topics:
  • Introduction to AI & LLMs (like ChatGPT)
  • Text-to-image models (e.g., MidJourney)
  • Text-to-speech, text-to-audio, speech-to-text models
  • Prompt engineering techniques
  • Best practices
  • Zero-shot and few-shot prompting
  • AI hallucinations, vectors, text embeddings
  • Intro to ChatGPT

What is Prompt Engineering?

• A career that emerged with AI rise
• Involves writing, refining, and optimizing prompts to improve AI interactions
• Requires ongoing monitoring and maintaining of prompt libraries
• Companies paying up to $335,000/year (source: Bloomberg)

AI Basics

• Artificial Intelligence: Simulation of human intelligence by machines
• Machine Learning: Using training data to predict outcomes (pattern detection)
• Language Models: GPTs, capable of understanding/generating human-like text

Language Models Development

• Eliza (1960s): Early NLP program simulating conversation (pattern matching)
• Shudlu (1970s): Simple command-based interaction
• GPT Evolution:
  • GPT-1 (2018)
  • GPT-2 (2019)
  • GPT-3 (2020): 175 billion parameters
  • GPT-4: Latest, trained on most of the internet

Prompt Engineering Mindset

• Similar to crafting effective Google searches
• Key: Write effective prompts that get desired results first time

Best Practices

• Clear Instructions: Specify details in prompts
• Avoid Leading Questions: Prevent bias
• Format Specification: Define response format (summary, list, detailed, etc.)

Persona Adoption

• Create prompts with character/persona for consistent results
• Example: Writing a poem as a specific persona/style

Advanced Prompting

Zero-Shot Prompting

• Leveraging pre-trained model capabilities without additional training examples
• E.g., Asking for dates like Christmas in America

Few-Shot Prompting

• Providing few example data for more accurate outputs
• E.g., Training the model with personal preferences for specific queries

AI Hallucinations

• Refers to unusual, incorrect outputs by AI models
• Can occur in both text and image models
• Highlights how AI interprets data

Vectors and Text Embeddings

• Text Embeddings: Convert text to high-dimensional vectors for semantic analysis
• Capture meaning behind words instead of lexicographical similarity
• Used in comparing the semantic similarity of texts
• Tools: OpenAI’s create embedding API

Use of ChatGPT (GPT-4)

• Getting Started:
  • Sign up/in at OpenAI website
  • Use the platform (ChatGPT)
  • Token management for interactions
• API Usage:
  • Obtain API keys
  • Create embedding via post requests

Conclusion

• Prompt Engineering is essential for maximizing interaction efficiency with AI
• Covers the essentials for getting the most out of language models like GPT-4