Investigating Non-Transitivity in LLM-as-a-Judge

TL;DR

This paper reveals that LLM-based evaluation judges often exhibit non-transitive preferences, affecting the reliability of model rankings, and proposes methods like round-robin and Swiss-Wise tournaments to improve evaluation consistency.

Contribution

It uncovers non-transitivity in LLM judges and introduces tournament-based methods to enhance the reliability of model rankings in LLM evaluation.

Findings

LLM judges show non-transitive preferences affecting rankings.

Round-robin tournaments improve correlation with benchmark rankings.

Swiss-Wise tournaments offer a computationally efficient alternative.

Abstract

Automatic evaluation methods based on large language models (LLMs) are emerging as the standard tool for assessing the instruction-following abilities of LLM-based agents. The most common method in this paradigm, pairwise comparisons with a baseline model, critically depends on the assumption of transitive preferences. However, the validity of this assumption remains largely unexplored. In this study, we investigate the presence of non-transitivity within the AlpacaEval framework and analyze its effects on model rankings. We find that LLM judges exhibit non-transitive preferences, leading to rankings that are sensitive to the choice of the baseline model. To mitigate this issue, we show that round-robin tournaments combined with Bradley-Terry models of preference can produce more reliable rankings. Notably, our method increases both the Spearman correlation and the Kendall correlation with Chatbot Arena (95.0% -> 96.4% and 82.1% -> 86.3% respectively). To address the computational cost of round-robin tournaments, we propose Swiss-Wise Iterative Matchmaking (Swim) tournaments, using a dynamic matching strategy to capture the benefits of round-robin tournaments while maintaining computational efficiency.