01 / EVALUATION
When a good score is the wrong answer
Choose a test that reflects the decision your model will support. Accuracy is a number; usefulness is a question.
Read the note ↗A SMALL READING ROOM FOR MACHINE LEARNING
Models meet
the real world.
Clear notes on data, experiments and the work that happens after a notebook looks convincing.
Explore the field notes ↗EXPERIMENT / 001
signal ≠ certainty
DATA → MODEL → EVALUATE↺
Good questions before bigger models.Measure carefully. Keep the baseline.
FIELD NOTES
Three useful places
to start.
Small, self-contained guides.
No leaderboard required.
02 / DATA
The quiet work of checking the dataset
Duplicates, leakage and changing labels can matter more than another round of hyperparameter tuning.
Read the note ↗03 / RETRIEVAL
Start with retrieval, not a larger prompt
Separate finding evidence from generating an answer. Inspect each step before combining them.
Read the note ↗WHY OUTSIDE?
A model is only part
of the system.
ML Outside is an independent collection of educational notes about machine learning in practice. The focus is on understandable experiments, honest evaluation and useful limitations — not claims of perfect intelligence.
The site is deliberately simple: no accounts, advertising, tracking scripts or external fonts.