With LLM-as-a-judge, one language model rates another model's answers, for example for accuracy or tone.
Example
A model gives each of 500 customer replies a score from 1 to 5 against a fixed checklist.
For pros
It scales better than human rating but has known biases: favouring longer answers, its own model family or the first answer. Clear grading instructions and human spot checks are a must.
