Introduction to Generative AI

Lecturer: Dr. Gwendolyn Stripling, AI Technical Curriculum Developer, Google Cloud

Course Overview

• Topics Covered:
  • Definition of generative AI
  • How generative AI works
  • Types of generative AI models
  • Applications of generative AI

What is Artificial Intelligence (AI)?

• Definition: AI is a branch of computer science dedicated to creating intelligent agents capable of reasoning, learning, and acting autonomously.
• Comparison to Machine Learning (ML):
  • Machine Learning: A subfield of AI, involves systems training a model from input data to make predictions on new data.
  • Supervised vs. Unsupervised ML:
    • Supervised Learning: Uses labeled data (e.g., predicting tip amount based on bill).
    • Unsupervised Learning: Uses unlabeled data (e.g., clustering employees based on tenure and income).

Deep Learning

• Relationship to ML: A subset of ML that uses artificial neural networks to process complex patterns.
• Artificial Neural Networks: Inspired by the human brain, consists of neurons (interconnected nodes).
• Semi-Supervised Learning: Utilizes both labeled and unlabeled data to train neural networks.

Generative AI (Gen AI)

• Subset of Deep Learning: Uses neural networks to generate new content (text, imagery, audio, etc.).
• Types of Models:
  • Generative Models: Generate new data instances based on learned distributions.
  • Discriminative Models: Classify or predict labels for data points.

Key Distinctions

• Traditional ML Models vs. Generative AI Models:
  • Traditional models predict/classify based on input data.
  • Generative models produce new, creative outputs from learned patterns.
• Output Characteristics:
  • Non-Gen AI: Outputs are labels, numbers, probabilities (e.g., spam or not spam).
  • Gen AI: Outputs are natural language, images, audio, and more.

Applications and Examples

• Large Language Models: Generate novel text based on training data.
• Generative Image Models: Convert text or images into other formats (e.g., image completion, animation).
• Transformers: Key architecture utilizing encoder-decoder setups to process and generate content.
• Use of Prompts: Directs models to produce specific outputs, useful for generating varied content.

Challenges

• Hallucinations: Incorrect or nonsensical outputs due to inadequate training data or context.

Tools and Platforms

• Generative AI Studio: Facilitates exploration and customization of Gen AI models on Google Cloud.
• Generative AI App Builder: Allows creation of Gen AI apps without coding, using drag-and-drop interfaces.
• Palm API and Maker Suite: Supports quick prototyping and experimentations with large language models.

Application Examples

• Code Generation: Converts code formats, explains code, generates SQL queries, etc.
• Sentiment Analysis, Occupancy Analytics: Specialized models for varied tasks.

Conclusion

• Generative AI Definition: A type of AI that learns from existing content to create new content based on statistical models.
• Power and Flexibility: From traditional programming to generative models, facilitating user-driven content creation across multiple formats.

Thank you for watching this course on Introduction to Generative AI!