AI Glossary / Training data

🏗️ Models and training

Training data is the material the model learned from: texts from websites, books, code, images with descriptions, recordings. Their quantity, diversity, and quality largely determine the model's capabilities - garbage data produces garbage answers. The most important legal disputes in the industry revolve around training data: creators and publishers ask on what basis their work ended up in training. Hence the growing role of licensed datasets, synthetic data, and mechanisms allowing publishers to block content-gathering bots.