What is Supervised Learning?

During training, the model makes a prediction on each labeled example, compares it against the true label, and adjusts its internal parameters to reduce the error, repeating over many examples and passes until performance converges. The two broad task types are classification (predicting a discrete category, like spam vs. not-spam) and regression (predicting a continuous value, like a price). The quality of a supervised model is bounded by the quality and coverage of its labeled training data — a dataset that's biased, mislabeled, or missing important cases produces a model with the same blind spots. Supervised learning sits alongside unsupervised learning (finding structure in unlabeled data) and reinforcement learning (learning from reward signals through interaction) as one of the three classical ML paradigms. In modern large language model development, supervised fine-tuning (SFT) is a specific supervised-learning step — training a pretrained model on labeled instruction-response pairs to shape its behavior — that typically precedes reinforcement-learning-based alignment steps.