What Is a Large Language Model and How Does It Work?

GPT-3 is a large language model in sort of bite sized pieces and where does the magic come from? I mean it's gotten everyone's imagination like everyone out there is playing with GPT. What makes it possible to do such an impressive thing from such simple code is that those linear algebraic operations have all of these sort of free parameters. You're defining this very flexible, under-determined computation that takes place. And then you do this training process where you show the model enormous amounts of data that gradually, incrementally, sets all of those free parameters.