Re-evaluating Open-ended Evaluation of Large Language Models

TL;DR

This paper critiques current Elo-based open-ended evaluation methods for large language models, proposing a game-theoretic approach to improve robustness against data biases and redundancies, thereby offering more reliable model assessments.

Contribution

It introduces a novel 3-player game framework for LLM evaluation, addressing biases in existing rating systems with new solution concepts for enhanced robustness.

Findings

The proposed method yields more intuitive and reliable ratings.

It reveals insights into the competitive landscape of LLMs.

The approach mitigates bias amplification in model evaluation.

Abstract

Evaluation has traditionally focused on ranking candidates for a specific skill. Modern generalist models, such as Large Language Models (LLMs), decidedly outpace this paradigm. Open-ended evaluation systems, where candidate models are compared on user-submitted prompts, have emerged as a popular solution. Despite their many advantages, we show that the current Elo-based rating systems can be susceptible to and even reinforce biases in data, intentional or accidental, due to their sensitivity to redundancies. To address this issue, we propose evaluation as a 3-player game, and introduce novel game-theoretic solution concepts to ensure robustness to redundancy. We show that our method leads to intuitive ratings and provide insights into the competitive landscape of LLM development.