Generative models

Generative models: AI systems that learn the underlying distribution of a dataset to produce novel, high-quality content: text, images, audio, and code.

This technology is the core engine for modern AI creation. Generative models capture the complex patterns within massive datasets, then synthesize new data points that share the original characteristics. Key architectures drive this capability: Generative Adversarial Networks (GANs) use two competing neural networks (generator and discriminator) to refine output quality; Variational Autoencoders (VAEs) focus on efficient data representation; and Transformer models, like those powering GPT-4, excel at sequential data tasks. Diffusion models, which generate realistic images by reversing a noise process, power tools like DALL-E and Midjourney. These models are not classifiers; they are creators, rapidly accelerating productivity across design, software development, and content strategy.

https://www.ibm.com/topics/generative-ai