What is Machine Learning?

The core workflow is training and evaluation: a model is fed labeled examples (in supervised learning) or raw data to find structure in (unsupervised learning), an optimization process adjusts the model's internal parameters to reduce error, and the resulting model is evaluated on data it hasn't seen to check that it generalizes rather than simply memorizing the training set. Overfitting — a model that performs well on training data but poorly on new data — is one of the field's most persistent practical failure modes, which is why held-out validation and test sets are standard practice. Machine learning spans a wide range of techniques beyond the large language models that dominate current attention: classification and regression models for structured tabular data, recommendation systems, anomaly detection, and reinforcement learning for sequential decision-making all remain in active production use and are often the better tool for a well-defined, narrow prediction task than a general-purpose LLM would be. Choosing the right technique for the problem — rather than defaulting to the most fashionable one — is itself a core ML engineering skill.