Raters or automated evaluators compare candidates, assign rankings, or provide scalar feedback. Training methods convert those comparisons into a signal, so unclear rubrics or rater bias directly shape learned behavior.
Preference data records judgments that one model output is better than another under a stated criterion.
Raters or automated evaluators compare candidates, assign rankings, or provide scalar feedback. Training methods convert those comparisons into a signal, so unclear rubrics or rater bias directly shape learned behavior.