Dataset

A dataset is the collection of material used to train an AI. For language models, that means trillions of words from web pages, books, and code. For image models, it means billions of photos.

"What it's fed determines what kind of AI it becomes" sums up nearly every dataset story out there. Feed it low-quality data, and bias and errors get baked right in. Feed it someone else's copyrighted work, and lawsuits follow. And when good data starts running out, the conversation turns to synthetic data.