Probing Multimodal Large Language Models on Cognitive Biases in Chinese Short-Video Misinformation
arXiv:2601.06600v4 Announce Type: replace Abstract: Short-video platforms have become major channels for misinformation, where deceptive claims frequently leverage visual experiments and social cues. While Multimodal Large Language Models (MLLMs) have demonstrated impressive reasoning capabilities, their robustness against misinformation entangled with cognitive biases remains under-explored. In this paper, we introduce a comprehensive evaluation framework using a high-quality, manually annotated dataset of 200 short videos spanning four health domains. This dataset provides fine-grained annot
The rapid proliferation of short-video misinformation, particularly leveraging visual and social cues, necessitates urgent investigation into the vulnerabilities of advanced AI models like MLLMs.
This research reveals critical vulnerabilities in leading AI models concerning cognitive biases within misinformation, which could undermine trust and accelerate the spread of deceptive content at scale.
Our understanding of MLLM robustness is updated, highlighting the need for more sophisticated bias mitigation and ethical deployment strategies for AI in content moderation.
- · AI ethics researchers
- · Social media platforms investing in robust content moderation AI
- · Developers focused on explainable and bias-aware AI
- · Platforms with unmitigated MLLM deployments
- · Users susceptible to sophisticated misinformation
Increased scrutiny and demand for 'robustness' assessments of MLLMs against social engineering and cognitive biases.
Development of new open-source datasets and benchmarks specifically designed to stress-test MLLMs for misinformation vulnerability.
Potential for regulatory frameworks to mandate specific bias detection and mitigation capabilities for AI deployed in public information spaces.
This signal links to a primary source. Continuum Brief monitors and indexes it as part of the live intelligence stream — we do not republish source content.
Read at arXiv cs.CL