Generative AI

Generative AI is a highly studied field at the moment due to the rising popularity of language models (LMs) and GANs. Certainly one of the flashiest types of ML algorithms, GenAI algorithms are different from traditional ML because they’re not meant to predict; they use training data to create semi-original content. GANs are commonly used for creating synthetic rows, for example, but other more famous algorithms like Stable Diffusion or GPT-4 allow for more varied and creative outputs.

Language models

With the rise of ChatGPT, the hype around language modeling has never been greater. However, language modeling originated quite a long time ago and has only recently come to the spotlight. LMs are simply ways of “understanding” human language—but they’re wildly inefficient. LMs simply take all the text they’re exposed to and convert them into numbers, which they store in parameters along with key items such as context, similar words, etc. Naturally, the more parameters a model has, the more information it’s able to store. They're essentially a large, cleverly arranged database of text.

When a query is made, it’s converted into numbers and the next word or sequence of words is predicted given the parameters in the model. LLMs are just language models that have so many parameters that they start displaying emergent properties that pass off as human qualities.