EMNLP'26 Werewolf benchmark tests multimodal LLMs on social deception
Overview
Researchers from Dongxi NLP and collaborators built a Werewolf (狼人杀) game benchmark to test whether multimodal large models can detect deception from video, audio and text, according to AI Era (新智元).
The report is a short lead, and it gives no benchmark scores, model versions or findings. A headline in the same report says GPT models miss social cues in deception, but the body text does not confirm this result.
Written by AI from the articles below · updated Oct 28, 9:06 PM ET
Check the sources:
Article timeline
The articles in this story. Times are ET.
- AI EraNewsWerewolf benchmark shows GPT models miss social cues in deception
AIResearchers from Dongxi NLP and collaborators built a Werewolf game benchmark to test whether multimodal large models can detect deception from video, audio, and text. The source text provided is only a short lead, so no benchmark scores, model versions, or findings are confirmed here.
Heat Trend
Not enough continuous observations to show a trend yet.