Generative AI refers to a class of artificial intelligence models that can generate new content — text, images, audio, video, code, and more — that resembles human-created data. Unlike discriminative models that classify or predict, generative models learn the underlying distribution of training data and sample from it to create novel outputs.
Generative vs Discriminative Models
Aspect
Discriminative Models
Generative Models
Goal
Learn decision boundary between classes
Learn the data distribution itself
Output
Class labels, probabilities
New data samples, probabilities
Examples
Logistic Regression, SVM, CNN classifiers
GANs, VAEs, GPT, Stable Diffusion
Uses
Classification, regression, detection
Content creation, synthesis, augmentation
Types of Generative Models
Generative Adversarial Networks (GANs) — Two networks (generator + discriminator) compete in a minimax game
Variational Autoencoders (VAEs) — Encode data to latent space, decode to reconstruct; adds probabilistic regularization
Autoregressive Models — Predict next token/ pixel given previous ones (GPT, PixelCNN)
Flow-Based Models — Invertible transformations for exact likelihood estimation
Diffusion Models — Gradually add noise then learn to reverse the process (Stable Diffusion, DALL-E)
Transformer-Based — Attention mechanisms that generate sequences (GPT-4, Claude, Gemini)
Why Now?
Three factors converged: massive datasets (the internet), powerful hardware (GPUs/TPUs), and algorithmic breakthroughs (attention, diffusion, RLHF). The result is models capable of human-like text generation, photorealistic image synthesis, and even music and video creation.
Practice Task: Use a free generative AI tool (ChatGPT, Claude, Gemini, or DALL-E) to generate something — a poem, an image, or a piece of code. Analyze the output: what's impressive? What's wrong? Compare the output with what a human might produce. Write a paragraph reflecting on the experience.