arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2408.01091cs.AI
收藏

剖析不和谐:针对自相矛盾指令对大型多模态模型进行基准测试

Dissecting Dissonance: Benchmarking Large Multimodal Models Against Self-Contradictory Instructions

Shanghai Jiao Tong University(上海交通大学);Fudan University(复旦大学);Shanghai Artificial Intelligence Laboratory(上海人工智能实验室)

Jin Gao, Lei Gan, Yuankai Li, Yixin Ye, Dequan Wang

AI总结:

本文构建含20,000个语言与视觉冲突的 Self-Contradictory Instructions 基准,评估 LMM 对自相矛盾指令的识别能力,并提出 Cognitive Awakening Prompting 以提升检测效果。

摘要:

大型多模态模型(LMM)在遵循人类指令方面表现出色。然而,随着多模态交互和上下文长度的趋势不断增强,可能会出现自相矛盾的指令,这对语言初学者和弱势群体构成挑战。我们推出 Self-Contradictory Instructions 基准,用于评估 LMM 识别冲突命令的能力。该基准包含 20,000 个冲突,在语言和视觉范式之间均匀分布。它由一个新颖的自动化数据集创建框架构建而成,该框架加快了流程,并使我们能够涵盖广泛的指令形式。我们的综合评估显示,由于缺乏自我意识,当前 LMM 在识别多模态指令不一致方面始终存在困难。因此,我们提出 Cognitive Awakening Prompting,从外部注入认知,显著增强不和谐检测能力。数据集和代码位于:https://selfcontradiction.github.io/。

英文摘要:

Large multimodal models (LMMs) excel in adhering to human instructions. However, self-contradictory instructions may arise due to the increasing trend of multimodal interaction and context length, which is challenging for language beginners and vulnerable populations. We introduce the Self-Contradictory Instructions benchmark to evaluate the capability of LMMs in recognizing conflicting commands. It comprises 20,000 conflicts, evenly distributed between language and vision paradigms. It is constructed by a novel automatic dataset creation framework, which expedites the process and enables us to encompass a wide range of instruction forms. Our comprehensive evaluation reveals current LMMs consistently struggle to identify multimodal instruction discordance due to a lack of self-awareness. Hence, we propose the Cognitive Awakening Prompting to inject cognition from external, largely enhancing dissonance detection. The dataset and code are here: https://selfcontradiction.github.io/.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑