Generative artificial intelligence (GenAI) - tools that use algorithms, data, and statistical models to create content of its own or that's similar to the data it was trained on. They are not search engines but rather trained chatbots. This is what we know as a technology of ChatGPT. Using a prompt, a chatbot generate the answer predicting the next missing content piece.
GenAI which generates text is trained on the large amounts of text from books, articles, and websites. It's analyzing the text to find patterns and relationships in human language. Once it is trained, it can create new text based on an understanding of human language. It predicts the next words, even if doesn't understand anything.
Generative AI image tools can produce diverse images in a range of mediums, everything from photorealistic oil painting style to anime. This type of AI learns through analysing datasets of images with captions or text descriptions. The image generation uses different technics.
Generative Adversarial Networks is based on the work of two neural networks. One creates fake images, while another classifies the images as real or generated.
Diffusion models imitate the images that they are trained on. The process is a back and forth imitation until the result is the same as given first.
Neural Style Transfer - knowing what two different concepts are, such as a cat and roller skates, dog and flowers, it can merge those concepts together when prompted to create an image of a dog lying on the board of the pond.
Similar principles are applied to train AI to generate video, music, ppt, codes and other information.
