⌘K
ENES
GitHub
conceptLearning Foundations~1 min readUpdated 2026-06-07#foundations#systems#evaluation

Mental Models for AI Systems

An AI system is a loop: inputs, model behavior, product constraints, evaluation signals, and human feedback. The model is important, but it is not the whole system.

Useful Frames

  • Model as component: treat the model as one fallible service in a larger product.
  • Distribution shift: ask what changes between the examples you test and the users you serve.
  • Error budget: decide which failures are tolerable, visible, recoverable, or unacceptable.

Use this note before designing RAG First-Pass Design or writing an evaluation playbook.

Related notes

Learning FoundationsEvaluation metrics & what they hideAccuracy lies on imbalanced data. Precision, recall, F1, ROC-AUC — what each measures, and how to pick the metric that matches the cost of being wrong.Learning FoundationsTrain/validation/test splits & data leakageHow to estimate real-world performance honestly — and the subtle ways information leaks and inflates your scores.Learning FoundationsFeatures, representations & the curse of dimensionalityModels don't see the world — they see features. How representation quality decides everything, and why high dimensions are weird.Learning FoundationsGeneralization, overfitting & the bias–variance tradeoffA model is only useful if it works on data it never saw. Overfitting, underfitting, and the tradeoff that governs both.