How to Become a Data Scientist: Skills, Education and Entry Points
The math, coding and domain skills data scientists need, realistic entry points like data analyst roles, and a step-by-step plan to get your first job.
Artificial Intelligence
What is Model Evaluation?
Quick answer
The process of testing an AI model against defined tasks and criteria to measure how accurate, reliable, and safe it is.
Model Evaluation, explained
Model evaluation, often called evals in AI teams, checks whether a model or AI feature does what it is supposed to do. It involves a set of test inputs with expected results or scoring rules, plus metrics such as accuracy, error rate, response time, and cost. For generative AI, evaluation may also use human reviewers or another model to grade answers.
A company building a customer support assistant might collect a few hundred real questions, write the correct answers, and score each new model version against them before release. Evaluation should continue after launch, because user behaviour, data, and model versions change. A strong evaluation set is specific to the task, includes difficult edge cases, and is kept separate from any data used for training.
Related terms
Model Drift
The gradual decline in an AI model's accuracy as real-world data or conditions change from those it was trained on.
OverfittingWhen a machine learning model learns its training data too closely, including noise, and performs poorly on new, unseen data.
Quality AssuranceThe process of ensuring software meets requirements and quality standards through planning, testing, and continuous improvement.
Guides that mention Model Evaluation