Back to All Terms
AI Definition

Generative AI

AI systems that create new text, images, audio, code, or video rather than simply classifying or predicting.

Generative AI refers to models that produce new, original content, text, images, audio, video, or code, rather than only analyzing or classifying existing data. This distinguishes it from traditional "discriminative" machine learning, whose job is typically to predict a label or number from inputs (like flagging spam or forecasting sales). Generative AI systems learn the underlying probability distribution of their training data well enough to sample entirely new, coherent examples from it. Large Language Models (like GPT and Claude) are the text-generating branch of this field, while diffusion models (like Stable Diffusion and Midjourney) are the dominant approach for image generation, gradually removing noise from a random starting point until a coherent image emerges.

Practical Implementation & Example

Generative vs. Traditional ML, Side by Side:

  • Traditional ML task: "Given this customer's browsing history, will they churn next month?" → Output: a probability score (0.83).
  • Generative AI task: "Write a personalized retention email for this customer based on their browsing history." → Output: entirely new text that didn't exist before.

The Two Dominant Generative Architectures:

  1. Autoregressive Transformers (text/code): Generate one token at a time, each new token conditioned on everything generated so far. Powers ChatGPT, Claude, and GitHub Copilot.
  2. Diffusion Models (images/audio/video): Start from random noise and iteratively "denoise" it over dozens of steps, guided by a text prompt, until a realistic image or audio clip forms. Powers Stable Diffusion, Midjourney, and Sora.

Both approaches share the same core idea: learn the statistical structure of real-world data well enough to generate convincing new samples from it.