Researchers have developed a new benchmark called Multimodal Interactive Deception Assessment (MIDA) to evaluate the ability of Multimodal Large Language Models (MLLMs) to detect deception in social interactions. The benchmark includes synchronized video and text data with ground truth labels. Current state-of-the-art MLLMs, including GPT-4o, show a significant performance gap in distinguishing truth from deception, indicating a struggle to integrate multimodal social cues and model others' intentions. The study proposes a Social Chain-of-Thought (SoCoT) reasoning pipeline and a Dynamic Social Epistemic Memory (DSEM) module to improve performance, with DSEM showing consistent benefits. AI
IMPACT Highlights a significant gap in current MLLM capabilities for understanding nuanced social cues, suggesting future research directions for more human-like AI interaction.
RANK_REASON The cluster describes a new academic paper introducing a benchmark and methods for evaluating LLM capabilities. [lever_c_demoted from research: ic=1 ai=1.0]
- arXiv
- Caixin Kang
- Dynamic Social Epistemic Memory
- GPT-4o
- Hugging Face
- MIDA
- MLLMs
- Multimodal Interactive Deception Assessment
- Sema2a
- Socotrella
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
