Using Game Play to Investigate Multimodal and Conversational Grounding in Large Multimodal Models

TL;DR

This paper introduces a goal-oriented game evaluation paradigm for multimodal models, assessing their visual understanding and conversational grounding, revealing performance gaps between large closed and open models.

Contribution

It adapts a text model evaluation method to multimodal models, providing a new benchmark for assessing visual and conversational grounding capabilities.

Findings

Largest models perform well on the games

Open-weight models struggle with the tasks

Deep captioning abilities influence performance

Abstract

While the situation has improved for text-only models, it again seems to be the case currently that multimodal (text and image) models develop faster than ways to evaluate them. In this paper, we bring a recently developed evaluation paradigm from text models to multimodal models, namely evaluation through the goal-oriented game (self) play, complementing reference-based and preference-based evaluation. Specifically, we define games that challenge a model's capability to represent a situation from visual information and align such representations through dialogue. We find that the largest closed models perform rather well on the games that we define, while even the best open-weight models struggle with them. On further analysis, we find that the exceptional deep captioning capabilities of the largest models drive some of the performance. There is still room to grow for both kinds of models, ensuring the continued relevance of the benchmark.