Section 3

Probability & Statistics

Section 1 answered "what is this piece of data." Section 2 answered "how does a model get better at its job." Section 3 answers a third question: "how sure is the model, and how sure should you be about it?" AI systems operate in environments with uncertain data, labels, and future outcomes, but not every output should be interpreted as calibrated uncertainty. A classifier's softmax vector parameterizes a categorical prediction; a language model does the same over tokens; and a confidence interval describes the repeated-sampling behavior of an estimation procedure under stated assumptions. Probability and statistics make those claims precise and expose where they can fail.