Evaluation
LLM-as-Judge
LLM-as-judge uses another LLM to score a model's outputs against a rubric, at scale.
It is useful for grading subjective qualities like tone, relevance, or faithfulness to a source.
It inherits the judging model's own biases and blind spots, so it needs calibration.
It is usually calibrated against a smaller human-labeled sample rather than trusted blind.
Treat it as one signal in an eval suite, not the whole suite.
← Back to the full glossaryRelated terms
More in Evaluation
Hallucination A hallucination is a confident, fluent output that is factually wrong or unsupported. Prompt Injection Prompt injection is untrusted input crafted to override a model's system prompt or task. Shadow Deployment A shadow deployment runs a candidate release against real production traffic, offline, before release.