Mental Models for Modern AI

Given software examples, distinguish rules-based automation, predictive machine learning, generative AI, and foundation-model use without treating overlapping categories as mutually exclusive.

Distinguish training, validation, deployment, and inference in the lifecycle of an AI model.

Trace one neural-network training step from input and prediction through loss calculation to a parameter update.

Interpret a language model's next-token output as a probability distribution and explain why sampling creates variability and probability does not guarantee truth.