Zheng and colleagues: MT-Bench, Chatbot Arena and model-judge evaluation

What changed

Combined multi-turn questions, crowdsourced pairwise preferences and analysis of strong language models as judges for open-ended assistant evaluation.

Scientific scope

Combined multi-turn questions, crowdsourced pairwise preferences and analysis of strong language models as judges for open-ended assistant evaluation. Position, verbosity and self-preference biases limit model judges. Agreement with sampled human preferences is not universal answer correctness or a current ranking.

Evidence

The linked original sources describe the stated contribution and its evidence limits. WRS has not independently rerun the experiments or certified a proof. Position, verbosity and self-preference biases limit model judges. Agreement with sampled human preferences is not universal answer correctness or a current ranking.

Contributors

• Source-listed authors and technical contributors · SCIENCE: Source-listed authors or explicitly credited technical contributors for the cited versions. Family entries retain source-specific attribution; a combined list is not one original-paper byline or equal individual work. Names alone do not establish resolved identities.

• Judging LLM-as-a-Judge with MT-Bench and Chatbot Arena

Provisional WRS assessment, edition 2026-10-16.1. Independent expert calibration is pending. This grade does not certify correctness or assign ownership.

Full scientific record · Browse contribution grades · Request a correction