Assessment of Multimodal Large Language Models in Alignment with Human Values
This addresses the gap in evaluating MLLM alignment with human values for researchers and developers, though it is incremental as it builds on existing hhh principles with new multimodal data.
The paper tackled the problem of assessing Multimodal Large Language Models (MLLMs) for alignment with human values like helpfulness, honesty, and harmlessness, by introducing Ch3Ef, a comprehensive evaluation dataset and strategy, resulting in over 10 key findings on MLLM capabilities and limitations.
Large Language Models (LLMs) aim to serve as versatile assistants aligned with human values, as defined by the principles of being helpful, honest, and harmless (hhh). However, in terms of Multimodal Large Language Models (MLLMs), despite their commendable performance in perception and reasoning tasks, their alignment with human values remains largely unexplored, given the complexity of defining hhh dimensions in the visual world and the difficulty in collecting relevant data that accurately mirrors real-world situations. To address this gap, we introduce Ch3Ef, a Compreh3ensive Evaluation dataset and strategy for assessing alignment with human expectations. Ch3Ef dataset contains 1002 human-annotated data samples, covering 12 domains and 46 tasks based on the hhh principle. We also present a unified evaluation strategy supporting assessment across various scenarios and different perspectives. Based on the evaluation results, we summarize over 10 key findings that deepen the understanding of MLLM capabilities, limitations, and the dynamic relationships between evaluation levels, guiding future advancements in the field.