Orientation — concepts before the maths

Your phone finishes your sentence before you do. Type "see you" and it offers "tomorrow"; type "thank you so" and "much" is already waiting. Is that understanding — or just very good counting? Sit with the question a moment, because your professional instinct pulls both ways: the suggestion is often exactly what you meant, yet you know perfectly well the phone has no idea what a meeting is. This track takes that everyday puzzle seriously and uses it as the doorway to everything else on the platform.

Four short walks, no equations anywhere. The first pins down what a model actually is — a compact guessing machine — and discovers that models are far older than computers: a weather rule of thumb, a grammar paradigm, a phrase-book are all models. The second turns the lens on you: when you guess the end of a colleague's sentence, or pick the right register for a difficult email, you are running a language model of your own, and naming what you already do makes the machine version far less mysterious. The third walk tells the field's one great plot twist — how software went from hand-written rules to patterns learned from examples, and why the rule-writing approach lost. The fourth asks why even the best model runs into limits it cannot think its way past: memory, compute, and how much text fits in its attention at once.

By the end, the autocomplete question has a real answer — one that is more interesting than either "it understands" or "it's just statistics", and that you can defend in conversation.

Take this track first if formulas put you off, or as a warm-up before the mathematical route: every idea met here — model, prediction, embedding, context window — returns later with numbers attached, and it helps to have met each one in plain words first.