Boosting Developer Productivity with AI Training

Introduction to Large Language Models

• What is Generative AI?
• A Bit of History …
• … and Then …
• RNNs
• Problems with RNNs
• Transformers
• Encoders and Decoders
• Generative AI and LLMs
• Training the Model to Predict the Next Word Visually
• The LLMs Landscape
• The Evolutionary Tree of LLMs
• The Microsoft 365 Copilot Ecosystem
• The LLM Capabilities vs LLM Size (in Parameters)
• Does the Model Size Matter?
• Inference Accuracy vs LLM Size
• Open AI GPT Models
• Llama
• The LLaMA Family of LLMs
• LLaMA 2
• The AI-Powered Chatbots
• How Can I Access LLMs?
• Options for Accessing LLMs
• Cloud Hosting
• Opinions about LLMs
• Multimodality of LLMs
• Infographic of Multimodality Tasks
• Example of an LLM Explaining a Joke
• Example of Cause & Effect Reasoning
• Inferring Movie from Emoji
• Prompt Engineering
• The Right People, with the Right Skills, for the Right Time …
• Context Window and Prompts
• Zero- and Few-Shot Prompting
• The Training Datasets
• The RedPajama Project (OSS LLaMA Dataset)
• AI Alignment
• Reinforcement Learning with Human Feedback (RLHF)
• Problems with RLHF
• Ethical AI

LLMs, a Technologist’s Perspective

• LLM Operational Aspects
• Understanding Model Sizes
• Physical Model Sizes
• The Training and Inference Costs
• The Model Training Phase’s Carbon Footprint
• Quantization
• Model Formats
• LLM Accuracy Benchmarks
• Open and Closed Book Benchmarks
• The Perplexity Performance Metric
• Embeddings
• Where are Embeddings Used?
• The Vector Databases
• LLM Concerns
• Ways to Interface with Local LLMs
• Using a Supported Programming API (Binding)
• UI Options
• Customization Options for LLMs
• Customization Options: Top-p and Top-k
• Customization Options: Temperature and Repetition Penalty
• Customization Option: The Turn Template
• Configuration Presets

Introduction to ChatGPT

• A Stylized OpenAI ChatGPT Logo
• OpenAI GPT Models
• OpenAI Models
• ChatGPT 4.0
• ChatGPT Prompts
• ChatGPT Prompts Strategies, Tactics, and Best Practices
• Prompt Engineering: Dealing with ChatGPT’s Hallucination Syndrome
• Prompt Engineering: Break Down the Complex Tasks into Smaller Ones
• Prompt Engineering: Examples of Prompts
• OpenAI API
• GPT Embeddings
• Embedding Models’ Risks and Limitations
• OK. How Can I Get My OpenAI Embedding
• Tokens, Take 1
• Tokens, Take 2
• The Tokenizer UI
• Prompts, Embeddings, and Tokens

AI-Powered Developer Productivity

• Generative AI and LLMs for Developers
• How to Become a Technologies and Philosopher All in One
• Gartner on AI-augmented Development Tools
• Developer-AI Pair Programming Paradigm
• The Tooling
• Some Facts …
• Code Generation: SQL Example
• Code Generation: Using ThreadLocal Storage in Java
• Code Generation: Thread-safe Singleton Design Pattern in C#
• Code Generation: Bash Scripting
• Code-to-Code Translation
• Code Llama
• Fine-Tuning Llama 2 Workflows
• GitHub Copilot
• Can I Trust AI-Generated Code?
• The Safeguards
• The General Recommendations …

Introduction to GitHub Copilot

• What is GitHub Copilot?
• Copilot Chat
• IDE and REPL Integrations
• Will Copilot Replace Developers?
• Can I Trust Code Generated by GitHub Copilot Code?
• GitHub Copilot’s Modus Operandi
• The Life of a Code Completion: The Big Picture
• Code Suggestions are Not Copy & Paste from Other Peoples’ Code
• The Shebang Prologue Hint
• Getting Started with GitHub Copilot
• GitHub Copilot Plans
• Copilot for Individuals
• Copilot for Businesses
• GitHub Copilot Security
• Responsible Copilot

Lab Exercises

• Lab 1. Learning the Colab Jupyter Notebook Environment
• Lab 2. Hello, AI!
• Lab 3. OpenAI Platform Overview
• Lab 4. Using OpenAI API
• Lab 5. Understanding Embeddings
• Lab 6. OpenAI API Project
• Lab 7. Copilot Environment Setup
• Lab 8. Hello, Copilot!