Training data is what an AI system learns from — the text, documents, and examples that teach the model how language works, what facts are true, and how to handle different types of inputs.
The quality, diversity, and representativeness of training data are among the most significant factors in how well a model performs. Biased training data produces biased models. Narrow training data produces models that fail on anything outside that narrow range. Low-quality training data produces models with unreliable outputs. The model is a reflection of what it learned from.
