Introduction to Generative AI

Course Overview

• Instructor: Dr. Gwendolyn Stripling, AI Technical Curriculum Developer at Google Cloud
• Key Learning Objectives:
  • Define generative AI
  • Explain how generative AI works
  • Describe generative AI model types
  • Describe generative AI applications

What is Artificial Intelligence?

• AI is a discipline within computer science focused on creating intelligent agents.
• Key Questions:
  • What is AI?
  • Difference between AI and Machine Learning (ML):
    • AI: Theory and methods to build machines that think and act like humans.
    • ML: Subfield of AI that involves training models from input data for predictions.

Machine Learning Overview

• Types of Machine Learning Models:
  • Supervised Learning:
    • Uses labeled data (tags with names, types, numbers).
    • Example: Predicting tips based on bill amount and order type.
  • Unsupervised Learning:
    • Deals with unlabeled data to discover patterns or clusters.
    • Example: Grouping employees based on tenure and income.

Deep Learning and Its Relation to Generative AI

• Deep learning uses artificial neural networks to process complex patterns.
• Neural Networks:
  • Inspired by the human brain with interconnected nodes (neurons).
  • Can process both labeled and unlabeled data (semi-supervised learning).

Generative AI Defined

• Generative AI is a subset of deep learning that generates new data instances based on learned patterns from existing data.
• Difference between Generative and Discriminative Models:
  • Discriminative Models: Classify data points based on features.
  • Generative Models: Generate new content based on probability distributions of existing data.

Generative Models Output

• Not generative AI if output is numerical (e.g., predicted sales).
• Generative AI if output is natural language, images, or audio.

Generative AI Process

• Foundation Model:
  • Uses training code, labeled and unlabeled data to generate new content (text, images, audio, etc.).
  • Example: Asking a question generates a relevant response based on learned data.

Types of Generative AI Models

• Generative Language Models:
  • Generate natural language responses.
• Generative Image Models:
  • Convert images to text or other images/videos.
• Generative Video Models:
  • Create videos from text input.
• Generative 3D Models:
  • Generate 3D objects from text descriptions.

Transformers in Generative AI

• Key innovation in natural language processing since 2018.
• Consists of encoder and decoder components.
• Hallucinations: Nonsensical outputs that can occur due to insufficient training data or noise.

Prompt Engineering

• Design prompts to control model output.
• Types of input-output mappings:
  • Text to Text
  • Text to Image
  • Text to Video
  • Text to Task

Foundation Models and Applications

• Foundation Models:
  • Pre-trained on large datasets and adaptable for various tasks.
  • Applications in healthcare, finance, customer service, etc.
• Generative AI Studio:
  • Tools for exploring, customizing, and deploying generative AI models.
• Generative AI App Builder:
  • Create applications without coding using a drag-and-drop interface.
• Palm API:
  • Allows quick prototyping and integration of generative AI tools.

Example Application

• Code generation (e.g., converting Python to JSON).
• Tools for debugging and generating documentation.

Conclusion

• Generative AI offers powerful capabilities for creating content based on learned data patterns.
• Importance of training data, model architecture, and user prompts in generating desired outputs.