AdamMeme: Adaptively Probe the Reasoning Capacity of Multimodal Large Language Models on Harmfulness

TL;DR

AdamMeme introduces an adaptive, multi-agent evaluation framework for assessing multimodal large language models' ability to understand meme harmfulness, overcoming static dataset limitations and revealing model weaknesses.

Contribution

It presents a novel, flexible evaluation method that adaptively probes mLLMs' reasoning on harmful memes through multi-agent collaboration and iterative data updating.

Findings

Systematically reveals performance variations among mLLMs

Provides in-depth analysis of model-specific weaknesses

Offers a dynamic, up-to-date assessment framework

Abstract

The proliferation of multimodal memes in the social media era demands that multimodal Large Language Models (mLLMs) effectively understand meme harmfulness. Existing benchmarks for assessing mLLMs on harmful meme understanding rely on accuracy-based, model-agnostic evaluations using static datasets. These benchmarks are limited in their ability to provide up-to-date and thorough assessments, as online memes evolve dynamically. To address this, we propose AdamMeme, a flexible, agent-based evaluation framework that adaptively probes the reasoning capabilities of mLLMs in deciphering meme harmfulness. Through multi-agent collaboration, AdamMeme provides comprehensive evaluations by iteratively updating the meme data with challenging samples, thereby exposing specific limitations in how mLLMs interpret harmfulness. Extensive experiments show that our framework systematically reveals the varying performance of different target mLLMs, offering in-depth, fine-grained analyses of model-specific weaknesses. Our code is available at https://github.com/Lbotirx/AdamMeme.