Types of Generative AI

1. Generativw Adversarial Networds (GANs): A GAN is an AI model where two neural networks compete with each other to create realistic new data.
   Generator: Creates fake data.
   Discriminator: Checks whether the data is real or fake.
   The Generator keeps improving until it can fool the Discriminator.
   
   Example- Creating realistic human faces, AI-generated artwork, etc.
   
   Another Example- Imagine a student and a teacher:
   The student draws a fake ₹500 note.
   The teacher checks whether it is real or fake.
   If the teacher finds mistakes, the student improves the drawing.
   After many attempts, the fake note looks almost real.
2. Variational Auto Encoders (VAEs): A VAE is an AI model that learns the important features of data and then creates new, similar data. It works in two steps:
   Encoder: Learns the important information.
   Decoder: Uses that information to generate a new example.
   
   Real-life Examples- Creating new handwritten digits, Generating new faces, etc.
   
   Another Example- Imagine you want to draw a flower.
   First, you learn the common features like petals, stem, and leaves.
   Then, you draw a new flower that looks different but still resembles a real flower.
3. Transformers: A Transformer is an AI model that understands the meaning and relationships between words or other data, allowing it to generate accurate and meaningful content.
   
   Real-life Examples - ChatGPT, Google Translate, AI coding assistants, Text summarization, etc.
   
   Another example - When you read the sentence:
   "Riya dropped the glass because it was slippery."
   You understand that "it" refers to the glass, not Riya.
   A Transformer learns this relationship and uses it to generate meaningful text.
4. Diffusion Models: A Diffusion Model is an AI model that starts with random noise and slowly removes the noise until a clear image is created.
   
   Real-life examples- AI image generation, Photo editing, AI art creation, etc.
   
   Another Example - Imagine a TV showing only static noise. As you adjust the signal, the picture gradually becomes clear.
   
   A Diffusion Model works in the same way:
   Starts with random dots (noise), Removes the noise step by step, and Produces a realistic image.