Training data

The examples a model learns from, such as text, images, code, or labeled records. Its quality and coverage shape what the model knows, what it gets wrong, and where debates about bias and copyright begin.

“It writes great Python and mediocre COBOL, which says a lot about its training data.”