Training Data
Domain: Data & Datasets
🎯 Motivation
The examples a model learns from. Quality and quantity of training data directly determine model performance.
📋 Concrete Example
GPT-4 was trained on millions of documents: books, articles, code, conversations. The diversity and quality of this data shaped its capabilities.
🔗 Analogy
A student’s textbooks and lecture notes - the quality of these learning materials directly determines how well they understand the subject. Training data is the model’s textbook.