Why this matters. In 1964, Richard Feynman used this exact demonstration at Cornell to explain why time runs one way. Mixing is easy — unmixing never happens. That one‑way fall from order to disorder is what we call entropy.
Every AI model trained today minimizes cross‑entropy loss — a direct descendant of the same idea. Shannon borrowed the word from physics in 1948 for the paper that made large language models possible. Feynman showed what it meant before it became a number on a dashboard. Watch Lecture 5: “The Distinction of Past and Future”