Prompts as Programs

We have the board. Now: the most direct way you write on it, the prompt, the words you actually type.

Here is the shift that makes prompting click. The words you give a model are not a request. They are an input. A request goes to a listener who works out what you meant and fills in the gaps. An input goes into a machine that continues it. And we know exactly what machine: the next-token predictor from the last module, the thing that reads what is in front of it and writes the most plausible next piece, over and over.

So your prompt is not a question handed to a clerk. It is the opening stretch of text that the model continues. Change the opening and you change what plausibly comes next. That is the whole reason phrasing matters. Not because certain words are magic spells, but because different words are genuinely different inputs, and different inputs run forward into different outputs.

This is why prompting is a real skill, and not a bag of tricks. It is just understanding how the run-up shapes the run.