Fitting a model's parameters to data so it performs a task no one wrote rules for. The training run adjusts parameters against a loss; the fitted parameters carry the behaviour into inference.
Machine Learning is the practice of fitting a model's parameters to data so that it performs a task nobody wrote explicit rules for. A training procedure adjusts those parameters to reduce a loss function measured on examples, and the fitted parameters carry the learned behaviour into inference.
Supervised learning fits inputs to labelled outputs, which covers classification and regression. Unsupervised learning finds structure in unlabelled data, as in clustering and embedding methods. Reinforcement learning fits a policy against a reward signal rather than labels, and it is the setting behind RLHF.
The measure that counts is performance on data the model has not seen. A model that scores well on its training set and poorly on held-out data is overfitting, which is why training splits the data, watches a validation curve, and stops when that curve turns (early stopping). Hyperparameters such as learning rate and batch size are set before training starts and govern how the fit proceeds.
A large language model is one branch of the field: a deep learning model trained by supervised next-token prediction over text, then adjusted with reinforcement learning from human feedback. The vocabulary of epochs, checkpoints and loss curves is shared with every other branch.
