3
Generative AI
AI systems that create new content (text, images, audio) by learning patterns from existing data.
DALL·E 3 creating images from text descriptions
GANs
Generative Adversarial Networks: Two neural networks (generator & discriminator) competing to create realistic data.
Generating photorealistic human faces for virtual avatars
Transformers
Architecture using self-attention mechanisms to process sequential data efficiently.
ChatGPT writing a 500-word essay in seconds
Diffusion Models
Generate data by iteratively denoising through a reverse diffusion process.
Stable Diffusion creating art from text prompts
VAEs
Variational Autoencoders: Encode data into latent space and decode to generate new samples.
Generating synthetic medical images for research
StyleGAN
GAN variant that separates content and style for controlled image generation.
MidJourney generating concept art for games
Tokenization
Breaking text into smaller units (tokens) for processing by language models.
Splitting “Hello World” into [“Hello”, “World”]
Prompt Engineering
Crafting text inputs to guide AI models toward desired outputs.
Using “cyberpunk cityscape at night” for image generation
Latent Space
A compressed representation of data where meaningful transformations occur.
Morphing a cat image into a dog via latent vectors
Neural Style Transfer
Applying artistic styles from one image to another using CNNs.
Turning a photo into a Van Gogh-style painting
Autoregressive Models
Generate data by predicting next token/element based on previous context.
GPT-4 writing code from natural language instructions
CLIP
Contrastive Language-Image Pretraining: Links text and images for multimodal understanding.
Finding images of “sunset over mountains” via text input
