Introduction to Generative AI

Lecturer

• Name: Dr. Gwendolyn Stripling
• Position: 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

What is Artificial Intelligence (AI)?

• Discipline: Branch of computer science creating intelligent agents.
• Objective: Build machines that think and act like humans.

AI vs. Machine Learning (ML)

• AI: Broad field like physics.
• ML: Subfield of AI, trains models from input data to predict new data.

Machine Learning Models

Types

• Supervised ML Models: Uses labeled data for training and prediction (e.g., predicting tips based on bill amount).
• Unsupervised ML Models: Uses unlabeled data, focuses on discovery and grouping (e.g., clustering employees by tenure and income).

Deep Learning

• Subset of ML: Uses artificial neural networks to process complex patterns.
• Neural Networks: Inspired by the human brain, made of interconnected nodes (neurons).
• Semi-Supervised Learning: Combines labeled and unlabeled data for training.

Generative AI

Definition

• Type of AI: Produces new content based on learned data (text, images, audio, etc.).
• Uses: Supervised, unsupervised and semi-supervised learning methods.

Generative vs. Discriminative Models

• Discriminative Models: Classify or predict labels for data points.
• Generative Models: Generate new data instances based on learned patterns.

Examples

• Discriminative Model Task: Classify dog vs. cat.
• Generative Model Task: Generate an image of a dog.

Large Language Models and Transformers

• Models: Transformers are composed of encoders and decoders.
• Function: Encoders encode input, decoders predict relevant task outputs.
• Issue: Hallucinations, nonsensical output caused by inadequate training or context.

Prompt Designing

• Prompt: Short text input guiding model output.
• Importance: Controls the generative model's response.

Generative AI Model Types

Model Types

• Text to Text: Converts natural language into text output (e.g., translation).
• Text to Image: Generates images from text descriptions (e.g., diffusion method).
• Text to Video: Produces video from textual input (e.g., script to video).
• Text to 3D: Generates 3D objects from text descriptions.
• Text to Task: Performs defined tasks based on text input (e.g., SQL query generation).

Foundation Models

• Definition: Large pre-trained models adapted for various tasks (e.g., sentiment analysis).
• Use Cases: Fraud detection, personalized support, etc.
• Google Tools: Vertex AI's Model Garden (e.g., PaLM API for chat/text, Stable Diffusion).

Generative AI Applications

Examples

• Code Generation: Convert Python to JSON using Google BARD in a prompt.
• Gen AI Studio: Tool for exploring and customizing generative models.
• Gen AI App Builder: No-code environment for building gen AI applications.

PaLM API

• Usage: Test and experiment with generative AI tools.
• Maker Suite Integration: Access API via a graphical interface for model training, deployment, and monitoring.

Conclusion

• Summary: Generative AI creates new content learned from existing data, transforming multiple industries through various applications.
• Thank You: Course: Introduction to Generative AI