JuStRank: Benchmarking LLM Judges for System Ranking
This work addresses the need for reliable system comparisons in generative AI, though it is incremental by extending evaluation from instance-based to system-level assessments.
The paper tackles the problem of evaluating LLM judges for ranking AI systems by conducting a large-scale study that assesses judges based on system-level rankings compared to human rankings, revealing biases and decisiveness in judge behavior.
Given the rapid progress of generative AI, there is a pressing need to systematically compare and choose between the numerous models and configurations available. The scale and versatility of such evaluations make the use of LLM-based judges a compelling solution for this challenge. Crucially, this approach requires first to validate the quality of the LLM judge itself. Previous work has focused on instance-based assessment of LLM judges, where a judge is evaluated over a set of responses, or response pairs, while being agnostic to their source systems. We argue that this setting overlooks critical factors affecting system-level ranking, such as a judge's positive or negative bias towards certain systems. To address this gap, we conduct the first large-scale study of LLM judges as system rankers. System scores are generated by aggregating judgment scores over multiple system outputs, and the judge's quality is assessed by comparing the resulting system ranking to a human-based ranking. Beyond overall judge assessment, our analysis provides a fine-grained characterization of judge behavior, including their decisiveness and bias.