Introduction to Generative AI

Jul 4, 2024

Introduction to Generative AI

Presented by Dr. Gwendolyn Stripling, AI Technical Curriculum Developer at Google Cloud

Course Objectives

  • Define generative AI
  • Explain how generative AI works
  • Describe generative AI model types
  • Describe generative AI applications

Key Concepts

What is Artificial Intelligence (AI)?

  • Discipline of computer science
  • Creation of intelligent agents
  • Systems that reason, learn, and act autonomously
  • AI vs. Machine Learning (ML)
    • AI: theory and methods to build intelligent machines
    • ML: a subfield of AI; programs/systems that train models from data

Machine Learning (ML)

  • Supervised ML: Uses labeled data (e.g., predicting tips in a restaurant)
  • Unsupervised ML: Uses unlabeled data to discover patterns (e.g., employee clustering)
  • Semi-Supervised ML: Combines small amounts of labeled data with large amounts of unlabeled data
  • Deep Learning
    • Subset of ML
    • Uses artificial neural networks
    • Processes complex patterns; many layers of neurons
    • Can use labeled and unlabeled data

Generative AI

  • Subset of deep learning
  • Uses neural networks
  • Processes both labeled and unlabeled data using supervised, unsupervised, and semi-supervised methods
  • Generative Models vs. Discriminative Models
    • Generative: Generates new data instances (e.g., creating images of dogs)
    • Discriminative: Classifies data instances (e.g., identifying if an image is of a dog)

Applications of Generative AI

  • Generative models can generate text, imagery, audio, etc.
  • Types of Generative Models
    • Text-to-Text
    • Text-to-Image
    • Text-to-Video
    • Text-to-3D
    • Text-to-Task

Evolution of AI Models

  • Traditional Programming
    • Hardcoding rules
  • Neural Networks
    • Predictive models using images and data
  • Generative Models
    • Generate new content based on large data sets

Large Language Models (LLMs)

  • Example Models: Palm, Lambda
  • Use vast amounts of data from the internet
  • Transformers
    • Encode input sequences
    • Use encoder and decoder architecture
    • Prone to generating nonsensical outputs (hallucinations)
  • Prompt Designing
    • Creating prompts to generate desired outputs

Applications of Generative AI in Google Cloud

  • Vertex AI
    • Offers a model garden including foundation models
  • Generative AI Studio
    • Explore and customize generative AI models
    • Tools for fine-tuning and deploying models
  • Generative AI App Builder
    • Create gen AI apps with drag-and-drop interface
  • Palm API
    • Test and experiment with Google’s LLMs and generative AI tools