Core ML Concepts · beginner · concept 23 of 311 · Updated Nov 4, 2026
Data Labeling
Data labeling is attaching the correct answer to examples so a supervised model can learn from them. It is often the most expensive and most underrated step in a project, because a model cannot be more consistent than its labels. Clear instructions, checks for agreement between labelers, and tools that pick the most useful examples to label all matter more than they sound.
Key terms
AnnotationGround truthInter-annotator agreementActive learningWeak supervision
Learn these first
Where you meet it in the real world
Training classifiers, evaluation sets for LLMs, medical image annotation, RLHF preference data
Guides and articles
A Recipe for Training Neural Networks ↗
Andrej Karpathy
What is data labeling? ↗
Snorkel AI
This unlocks