One way to look at how an LLM works is that it takes some string of natural language S, converts S into a vector embedding that represents S which we can denote here as VS, and runs VS through some functions to predict the next token that should come after