LLM-as-a-judge
LLM-as-a-judge is the practice of using a language model to grade another model's outputs against a rubric or a reference answer, instead of a human grader.
Zheng et al. found strong judges such as GPT-4 agreed with human preferences over 80% of the time, about as often as humans agree with each other. Judges also show biases: they favour the first answer shown (position bias), longer answers (verbosity bias) and their own outputs (self-enhancement).
Mitigations include swapping answer order, specific rubrics, reference answers, and checking agreement against a human-labelled sample, for example with Cohen's kappa.
Learn more on Kojen
Related terms
- Cohen's kappaCohen's kappa measures agreement between two raters on categorical labels, corrected for the agreement expected by chance: 1 is perfect, 0 is chance level..
- Agent evalAn agent eval measures an AI agent on multi-step tasks, scoring both the final outcome and how it got there, including tool calls, turns, cost and reliability over repeated runs..
- HallucinationA hallucination is model output that is fluent and confident but unsupported by its source material or false about the world..
Sources
- Zheng et al., 2023. Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena (arXiv:2306.05685)
Put the definitions to work.
Kojen evaluates, tunes and red-teams your agents with these methods built in: Wilson intervals, pass^k, LLM judges with visible components. Free to start.
Start training free