Embeddings: In the end it is all about your relationships

Generative AI has been all the craze since the arrival of ChatGPT and Stable Diffusion. But the real promise of these tools is not that they can generate media but rather that they can find relationships between media and thus between the people who made the media. We all know that computers start by turning expressions into numbers, zeros and ones, ASCII numbers for letters, and maybe 0-255 for RGB pixel values. Machine Learning goes another step further by making a network between these numerical representations of human expression. The network encodes that cats are numerically closer to dogs than to antelopes. it allows them to have a better chance of responding to the question “What else did she bring to the vet?” and generate an answer. But conversations are not ultimately about generating words but finding relationships and connecting with people. Let’s use neural networks to make social networks. To do this you have to learn to play with embeddings which are coordinates in the neural network. In this session we will learn how to get embedding from an API for textual and visual media and then compare them to see if they are more or less alike.