IsoBench: Benchmarking Multimodal Foundation Models on Isomorphic Representations

TL;DR

IsoBench is a benchmark dataset that evaluates foundation models across different input modalities using isomorphic representations, revealing modality preferences and proposing techniques to improve multimodal performance.

Contribution

The paper introduces IsoBench, a novel benchmark with isomorphic input representations, and proposes prompting methods to enhance multimodal model capabilities.

Findings

Models prefer textual over visual representations.

Claude-3 Opus performs 28.7 points worse with images.

Proposed techniques improve performance by leveraging multiple representations.

Abstract

Current foundation models exhibit impressive capabilities when prompted either with text only or with both image and text inputs. But do their capabilities change depending on the input modality? In this work, we propose $\textbf{IsoBench}$, a benchmark dataset containing problems from four major areas: math, science, algorithms, and games. Each example is presented with multiple $\textbf{isomorphic representations}$ of inputs, such as visual, textual, and mathematical presentations. IsoBench provides fine-grained feedback to diagnose performance gaps caused by the form of the representation. Across various foundation models, we observe that on the same problem, models have a consistent preference towards textual representations. Most prominently, when evaluated on all IsoBench problems, Claude-3 Opus performs 28.7 points worse when provided with images instead of text; similarly, GPT-4 Turbo is 18.7 points worse and Gemini Pro is 14.9 points worse. Finally, we present two prompting techniques, $\textit{IsoCombination}$ and $\textit{IsoScratchPad}$, which improve model performance by considering combinations of, and translations between, different input representations.