AI mechanics

Training Data

The pile of text a model read before it existed.

Books, code, websites, conversations. Trillions of tokens, assembled at a point in time and then closed.

It has a cutoff and it has a shape. It over-represents whatever gets written down, in the languages it gets written down in.

Also called: training set, the pile, cutoff, data.

Badge not earned yet

Finish the AI 101 card that teaches it and the badge is yours. Reading this page earns nothing.

Learn this in AI 101