judge bias
Systematic errors of model graders, such as favouring the first answer shown, longer answers or their own model's style.
In more detail
Zheng and colleagues documented position, verbosity and self-enhancement bias. Swap the order of pairs, cap length, use a different model family as judge and check against human labels.
Related terms
- LLM-as-judge — Using a language model to grade another model's output against a rubric, for qualities that code cannot check.
- Cohen's kappa — A measure of agreement between two raters that corrects for the agreement expected by chance, from 1 for perfect to 0 for chance level.
Taught in
- Model-Graded Evals — AI Engineering: LLM Apps, RAG & Agents