Decoder. plain-English AI glossary

Pairwise Comparison

● Core

Comparing two models side-by-side on the same input and asking judges which output is better.

Think of it like

Like a taste test where you try both sodas and pick your favorite.

Example

You show judges the same prompt and outputs from Model A and Model B. They pick which response is clearer, more helpful, more accurate.

How it actually works

Pairwise comparison is more reliable than absolute scoring because judges anchor on relative differences, not absolute scales. Psychologically, humans are better at comparisons than independent ratings. Downside: it's expensive (n models = n² comparisons) and you can't measure degradation on single metrics (only relative ranking).

For product teams

Pairwise comparison is the most reliable way to measure relative model quality.

For engineers

Collect multiple judgments per pair to reduce noise; record model order (A vs B vs B vs A) to detect biases.

Related

Read anything AI without the jargon

Look up any term in plain English, or save terms as you read with the free Chrome extension.

Open DecoderAdd to Chrome