Five core ideas, or: what is actually happening inside a clanker

A transformer explained in five ideas: embeddings, MLPs, attention, decoding, and the quirks that fall out of how it actually runs.