Generative AI is a type of artificial intelligence that can produce, or generate, new content. Generative AI uses unsupervised learning algorithms to create novel text, imagery and more. It does this by detecting underlying patterns related to what is input by a user and creating similar content. Modern generative AI models are trained on massive datasets and utilize several techniques to function:
- Generative Adversarial Networks (GAN): GANs utilize two neural networks—a generative network and a discriminatory network—to simulate conceptual tasks. The generative network generates synthetic data similar to the source data, while the discriminatory network tries to distinguish between the synthetic data and the original.
- Variational Auto-Encoders (VAE): Broadly, VAEs encode and compress the source input, then decode and attempt to reconstruct the information from the original source. This process helps generative AI models “fine tune” their outputs.
- Transformers: Transformers are deep learning models that mimic cognitive attention to classify and gauge the significance of input data. Transformers are trained on large language and/or imagery datasets and are an increasingly popular method for generating text and imagery.