AI eval

A repeatable test of an AI feature: a set of inputs, a way to grade each output, and a score you track across changes.

In more detail

Evals replace "it looked fine" with numbers. Grade with code where you can, with a model judge where you must, and compare versions case by case, since a higher average can hide a regression. Without a way to judge answers there is no eval, and the feature cannot be shown to work.

Related terms

Taught in

References

Certification courses