Evaluation & Data42

How we measure whether AI is any good, and what data we teach it with. Accuracy, precision and recall, benchmarks, labeling, bias, data augmentation: the words you need to read a results table.

Level
A/B Test

A test that shows two versions to real people and compares results

Beginner
Accuracy

The share of all judgments a model got right

Beginner
AI Bias

Skew in the training material showing up unchanged in an AI's judgment

Ethics Beginner
AI Content Detection

Working out whether something was made by an AI

Ethics Beginner
Benchmark

A shared set of problems used to measure candidates the same way

Beginner
Bounding Box

The smallest box around an object, and where that box sits

Vision Beginner
Calibration

Making a model's stated confidence match how often it's right

Advanced
Class Imbalance

A dataset where one category vastly outnumbers the rest

Intermediate
Confidence Score

The number a model attaches to how sure it is about its own answer

Vision Beginner
Confusion Matrix

A table that sorts right and wrong outcomes into four boxes

Intermediate
Cross-Validation

Splitting data into pieces, checking each in turn, then averaging

Training Advanced
Data Augmentation

Transforming data you already have to multiply how much of it you have

Training Intermediate
Data Cleaning

Fixing collected data up into a usable state

Beginner
Data Leakage

When information that should stay separate crosses into training

Intermediate
Data Quality

How trustworthy the data used for training actually is

Beginner
Data Science

Turning scattered data into an answer worth acting on

Basics Beginner
Dataset

A bundle of material gathered and organized to teach an AI

Basics Beginner
Dataset Bias

When the data you collected already leans to one side

Beginner
Elo Rating

A skill number built up from head-to-head results

Intermediate
Evaluation Metric

The yardstick that decides what counts as doing well

Beginner
Explainability

How much an AI can show for why it decided what it did

Ethics Intermediate
F1 Score

Precision and recall folded into a single score

Intermediate
Feature

One ingredient an AI looks at when it makes a judgment

Basics Beginner
Ground Truth

The real answer training and grading are measured against

Beginner
Human Evaluation

Evaluation where a person looks and scores it directly

Beginner
Label

The answer a person attaches to data ahead of time

Basics Beginner
Labeling

The work of a person attaching an answer to each piece of data

Beginner
Leaderboard

A table that lines evaluation scores up in order

Beginner
Learning Curve

A line showing how the score changes over training

Training Beginner
LLM-as-a-Judge

Letting a language model rate an answer instead of a person

Intermediate
Mean Squared Error

A penalty that squares how far off you were, then averages

Intermediate
Outlier

A single value sitting far apart from the rest

Beginner
Overfitting

Memorizing the practice answers and failing on anything new

Training Intermediate
Perplexity

A score for how many directions a model is torn between

LLM Advanced
Precision & Recall

How many picks were right, and how many real ones got caught

Intermediate
Prompt Adherence

The yardstick for whether a result matches what was requested

Generative Intermediate
Robustness

Still holding up when conditions get shaken

Intermediate
ROC & AUC

Scores across every cutoff, folded into one curve and one number

Advanced
Sampling Bias

A skew in data caused by who got picked

Intermediate
Synthetic Data

Practice data made up in place of the real thing

Intermediate
Underfitting

Too simple to get even the practice data right

Training Intermediate
Validation Set

Data set aside for checking only, never used to train

Training Beginner