Generative AI

The category of artificial intelligence that creates new content (text, images, audio, video, and code) from natural-language instructions, by learning patterns from large volumes of data.

Generative AI is the umbrella covering AI systems capable of *creating* content, rather than only classifying or predicting. These models are trained on huge volumes of examples (text, images, audio) and learn the statistical patterns that let them generate something new and coherent from an instruction: the *prompt*.

The main fronts:

Text: LLMs like those behind ChatGPT, Claude, and Gemini write, summarize, translate, and code.
Images: diffusion models like Midjourney and Stable Diffusion create illustrations and photos from descriptions.
Audio: speech synthesis (text-to-speech), voice cloning, and music generation.
Video: generating and editing scenes from text or images, a fast-evolving area.
Code: assistants that complete and generate programs.

The common thread: the interface is natural language. Describing what you want replaces mastering the tool, which explains the explosive adoption since 2022, when ChatGPT brought the technology to the mainstream. The limitations are shared too: models can "hallucinate" (confidently generate false content), reproduce biases from their training data, and raise copyright questions about generated material.

Concrete example: a small brand launches a product and uses generative AI across the whole campaign: the ad copy comes from an LLM, the images from Midjourney, and the video narration from a text-to-speech tool.