Machine Learning
also called ML
Programs that improve from examples instead of explicit instructions.
A field of AI where a system learns a statistical pattern from data rather than being hand-coded with rules. You supply examples; the algorithm finds the mapping.
Classical ML — regression, decision trees, gradient boosting — still runs a huge amount of production software: fraud scoring, ranking, forecasting. It is cheaper, smaller and far more explainable than an LLM, and for a lot of problems it is the correct answer.
The distinction that matters in practice is not supervised versus unsupervised, it is whether you can write down your objective as examples. If you can, you can learn a mapping. If you cannot, you are back to rules — and rules break quietly on the inputs you did not enumerate.
When a rule beats a model
A team wants a model to decide whether an invoice total is arithmetically valid.
You do not need learning here: adding the line items is exact, cheap and explainable. Reserve the model for the judgement calls — is this line item plausible for this vendor — where no rule survives contact with reality.
Reaching for a generative model when the task has a fixed label set and good training data.
- guideIntroduction to large language models ↗
A plain-language official introduction to models, tokens and context. Teaching material rather than origin research — good for the definition, not for a claim about the frontier.
- paperScaling laws for neural language models ↗
Empirical power laws relating loss to parameters, data and compute — the justification for the "bigger is better" era of training.
Deep Learning
Machine learning with many-layered neural networks — the technique behind modern AI.
AI · Artificial Intelligence
Software that does things we used to think only people could do.
Overfitting
Memorising the training data instead of learning the pattern.
Evaluation · Evals
How you know the system is actually working.