LLM Comparator: Visual Analytics for Side-by-Side Evaluation of Large Language Models
This tool addresses scalability and interpretability challenges for researchers and engineers evaluating LLMs, though it is incremental as it builds on existing evaluation approaches.
The authors tackled the problem of analyzing results from automatic side-by-side evaluation of large language models (LLMs) by developing LLM Comparator, a visual analytics tool that enables interactive workflows for understanding model performance differences and qualitative response variations, as validated through an observational study with industry participants.
Automatic side-by-side evaluation has emerged as a promising approach to evaluating the quality of responses from large language models (LLMs). However, analyzing the results from this evaluation approach raises scalability and interpretability challenges. In this paper, we present LLM Comparator, a novel visual analytics tool for interactively analyzing results from automatic side-by-side evaluation. The tool supports interactive workflows for users to understand when and why a model performs better or worse than a baseline model, and how the responses from two models are qualitatively different. We iteratively designed and developed the tool by closely working with researchers and engineers at a large technology company. This paper details the user challenges we identified, the design and development of the tool, and an observational study with participants who regularly evaluate their models.