Stackflight

Most explanations of large language models are either a hand-wave or a research paper. Stackflight is the middle. It is a finite course — nine tracks, fifty short units, about three and a half hours of narration — that takes you from what a token is to why a well-sourced answer can still be wrong, and what to do about it in the code you actually ship. HOW A UNIT WORKS Every unit is the same short loop, because the loop is what makes it stick: • Learn — a few paragraphs in plain language. Some units ask you to predict the answer before they explain it, because being wrong first is what makes the explanation land. • Try — split a word into tokens by hand. Trace a claim back to its evidence. Decide which text is allowed to give orders. Small hands-on exercises, not multiple choice about a diagram. • Check — a short check with the reasoning shown for every option, right and wrong. Then it…