Generative AI

Generative models learn the statistical structure of their training data and then sample new examples from it. Text generators are usually large language models that predict one token at a time. Image and video generators commonly use diffusion models, which start from random noise and refine it step by step toward an output that matches the prompt. Earlier approaches include generative adversarial networks (GANs), introduced in 2014.

Generative AI is the engine behind most AI software launched in recent years: writing assistants, coding tools, image and video generators, voice cloning and AI search. When you evaluate one of these products, you are partly evaluating the underlying generative model and partly the workflow, controls and data protections the vendor has built around it.

Example: a marketing team asks an image generator for a product photo on a beach background; the model creates a new image rather than retrieving an existing one.

Limits: output can be inaccurate, biased or similar to copyrighted training material, and quality varies between runs. Licensing terms for commercial use differ between vendors, so check them before publishing generated content.