Deviation Ratings: A General, Clone-Invariant Rating Method

TL;DR

This paper introduces deviation ratings, a novel clone-invariant rating method for multi-agent and multi-task evaluations modeled as N-player general-sum games, addressing redundancy issues in strategic interactions.

Contribution

It presents the first N-player general-sum clone-invariant rating method based on coarse correlated equilibria, extending previous two-player zero-sum approaches.

Findings

Effective in N-player general-sum game scenarios

Reduces redundancy distortions in ratings

Applicable to large language model evaluations

Abstract

Many real-world multi-agent or multi-task evaluation scenarios can be naturally modelled as normal-form games due to inherent strategic (adversarial, cooperative, and mixed motive) interactions. These strategic interactions may be agentic (e.g. players trying to win), fundamental (e.g. cost vs quality), or complementary (e.g. niche finding and specialization). In such a formulation, it is the strategies (actions, policies, agents, models, tasks, prompts, etc.) that are rated. However, the rating problem is complicated by redundancy and complexity of N-player strategic interactions. Repeated or similar strategies can distort ratings for those that counter or complement them. Previous work proposed ``clone invariant'' ratings to handle such redundancies, but this was limited to two-player zero-sum (i.e. strictly competitive) interactions. This work introduces the first N-player general-sum clone invariant rating, called deviation ratings, based on coarse correlated equilibria. The rating is explored on several domains including LLMs evaluation.