[rramadass]

Any article titled "How LLMs work" (or similar) and which does not start with conditional and joint probability (high-level and not necessarily detailed) and then show by hand how a trivial language with tokens (eg: "the", "mat", "cat", "sat", "on") can produce the semantically coherent and most likely sentence (eg. "The Cat sat on the Mat") is no good. The intuition for the whole should be built using the above before diving into details of transformers etc.