HM

Generative AI in Business - Week 8 Notes

Learning Objectives

  • Define what Generative AI is.
  • Explain how generative AI works.
  • Differentiate between language models and diffusion models.

What is Generative AI?

  • Definition: Generative Artificial Intelligence (GenAI) refers to AI that creates new content, which can include text, images, code, music, etc.
  • Basis: It creates content based on patterns it has learned from large datasets.

Examples of Generative AI:

  • Text Generators:

    • ChatGPT
    • Copilot
    • Claude
    • Gemini

  • Image Generators:

    • DALL·E
    • MidJourney
    • Stable Diffusion
    • Adobe Firefly

  • Code Generators:

    • GitHub Copilot
    • Codeium

  • Voice Generators:

    • ElevenLabs
    • Speechify

  • Writing Assistants:

    • Grammarly

  • Research Assistants:

    • NotebookLM

How Generative AI Works

  1. Training: GenAI models are trained on very large datasets, known as big data.

    • Datasets Include:
      • Text: Books, articles, websites, conversations.
      • Images: Photographs, paintings, illustrations.
      • Audio: Speech recordings, music tracks.
      • Video: Movies, YouTube clips, surveillance footage.

  2. Learning: The model learns patterns, structures, and relationships in the data.

  3. Content Generation: It creates new content based on the learned patterns.

Example Output Process:

  • The AI processes input data through various stages leading to the generation of new content.

Model Types

1. Language Models

  • Functionality:
    • Example Prompt: "The cat is on __?"
    • Predict the next word or phrase based on previous context.
    • Examples: Generates human-like text for chatbots, translation, writing aids.

2. Diffusion Models

  • Functionality:
    • Create images from random noise, employing a step-by-step refinement process.
    • Applications include art, marketing, and design.
    • Examples: Stable Diffusion, Mid Journey, DALL·E.

Deep Learning Foundations

  • Utilizes neural networks with many layers to detect complex patterns in data.
  • Powers generation of text, image, and audio content.

Transformers: The AI Game Changer

  • Mechanism:
    • Processes entire inputs at once rather than sequentially.
    • Employs self-attention to identify key information within the data.
    • Foundation for advanced models such as GPT (Generative Pre-trained Transformer).

How ChatGPT Works

  1. Input Processing:

    • User prompts are input as text (e.g., "Once upon a time, in a distant kingdom").
    • The prompt is tokenized into individual components (e.g., "Once", "upon", "a", etc.).

  2. Embedding Process:

    • Tokens are converted to numerical representations.

  3. Attention Mechanism:

    • Weights are assigned to tokens according to their importance (e.g., location descriptors and narrative indicators).

  4. Prediction:

    • The model predicts the next token in the sequence.

  5. Repetition:

    • This process is repeated until the content generation is completed.
    • Example Generated Text: "Once upon a time, in a distant kingdom, there lived a brave knight who sought to slay a fearsome dragon."