Training data

Training data is the material a model learns from. For large language models it usually includes text gathered from the public web, books, code, licensed collections, and data created by people or other models for the purpose.

What goes into training data shapes what a model knows, how it behaves, and what biases it may carry. It is also at the center of many copyright lawsuits, which ask whether copying works to train a model is lawful, and of privacy disputes over personal information.

Most developers disclose only general descriptions of their training data. Some laws, including the EU AI Act, now require providers of general-purpose models to publish a summary of the content used.