Skip to main content

COMPEL Glossary / llm-as-judge

LLM-as-judge

An evaluation technique using a large language model to score outputs from another LLM on quality dimensions — helpfulness, correctness, safety — scaling evaluation beyond human-rater capacity.

What this means in practice

Strengths: scalability, consistency. Weaknesses: judge-model biases, verbosity preference, and self-preference when the judge and candidate share architecture.

Synonyms

model-graded evaluation , LLM judge , judge model

See also

  • Evaluation harness — The infrastructure that runs capability, regression, safety, and human-review evaluations on an LLM feature on a defined cadence.
  • Benchmark contamination — The presence of benchmark test data in foundation-model training corpora — whether through web crawling or deliberate inclusion — inflating reported benchmark scores and breaking the comparability of benchmark results across models.
  • Red-team experiment — An adversarial experiment designed to probe failure modes rather than validate desired behavior — structured, hypothesis-driven exploration of safety bypass, goal mis-specification, jailbreak, and harm.

Related articles in the Body of Knowledge