Generative AI is the branch of artificial intelligence concerned with producing new content: text, images, audio, video, code or 3D assets. Where a classifier maps an input to a label, a generative model learns the distribution of its training data well enough to sample new examples from it. The term became mainstream during the AI boom that followed the launch of ChatGPT in November 2022, and by 2024 it was shorthand for an entire product category and investment wave.
Technical foundations
Modern generative AI rests on a small set of model families. Autoregressive models such as large language models generate sequences one token at a time and power chat assistants and code generation. Diffusion models dominate image and video synthesis, learning to reverse a gradual noising process. Earlier approaches include the generative adversarial network (2014), which drove the first wave of photorealistic synthesis and deepfakes, and the variational autoencoder. Nearly all current systems are built on the Transformer (architecture) architecture and trained with self-supervised learning over web-scale training data.
Modalities
Text generation is the most widely deployed form, through assistants like ChatGPT, Claude and Gemini. Text-to-image systems such as Stable Diffusion and Midjourney brought generation to visual art in 2022; text-to-video followed from 2024 with Sora, Veo and Kling. Other active modalities include speech synthesis, music generation and 3D generation.
Impact and debate
Generative AI reshaped software development through AI coding assistants, marketing, design and education, while raising disputes over copyright in training data, job displacement in creative industries, misinformation and low-quality synthetic content at scale, sometimes called AI slop. Critics such as the authors of the stochastic parrot paper question whether fluent generation reflects understanding. Proponents argue the technology augments rather than replaces human creativity, a debate that remains unresolved.