How ChatGPT and Other LLMs Work—and Where They Could Go Next | WIRED

All of this text data, wherever it comes from, is processed through a neural network, a commonly used type of AI engine made up of multiple nodes and layers. These networks continually adjust the way they interpret and make sense of data based on a host of factors, including the results of previous trial and error. Most LLMs use a specific neural network architecture called a transformer, which has some tricks particularly suited to language processing. (That GPT after Chat stands for Generative Pretrained Transformer.)

Specifically, a transformer can read vast amounts of text, spot…