GLOSSARY
Data Labeling
The human work underneath every AI system: tagging text, images and audio so models have ground truth to learn from — an industry of millions of workers.
Models learn from labeled examples, and someone has to create them. Data labeling spans gig workers clicking bounding boxes, domain experts annotating medical scans, and the elite tier — PhDs writing RLHF comparisons — that frontier labs compete for. The industry's geography and wages became a fairness issue in themselves, and synthetic data now substitutes for some of it.
Quality is the whole game: ambiguous guidelines, rater fatigue and inconsistent annotators quietly poison models, which is why serious programs measure inter-annotator agreement, run gold-standard checks and pay for expertise where it matters. “Garbage labels, garbage model” is the labeling-era version of the oldest rule in computing.