Re-evaluating Open-ended Evaluation of Large Language Models
This addresses evaluation challenges for researchers and developers of large language models, but is incremental as it builds on existing open-ended evaluation systems.
The paper tackles the problem of bias in open-ended evaluation of large language models caused by Elo-based rating systems' sensitivity to redundancies, and proposes a game-theoretic approach that leads to intuitive ratings and insights into the competitive landscape.
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.