PulseAugur
EN
LIVE 06:49:08

LLMs struggle to balance trust and accusations in social deduction games

By PulseAugur Editorial · [1 sources] ·

Researchers have developed a new benchmark for evaluating Large Language Models (LLMs) in social deduction games like Werewolf, focusing on how their beliefs shift based on accusations. The study found that while larger LLMs can better distinguish between wolves and villagers, they are still heavily influenced by accusations, especially when made by a trusted source, even if that source is a wolf. The findings indicate that current open-weight LLMs struggle to balance source trust with accusation content in strategic communication. AI

IMPACT This research highlights limitations in LLM reasoning and trust assessment, suggesting areas for improvement in their ability to handle nuanced social interactions and information integration.

RANK_REASON Academic paper detailing a new benchmark for LLM evaluation. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →

LLMs struggle to balance trust and accusations in social deduction games

How we ranked this

Signal score
27 / 100
Composite score across the factors below. Higher = stronger signal that this story matters right now.
Newsworthiness bucket
Tool
Academic paper detailing a new benchmark for LLM evaluation. [lever_c_demoted from research: ic=1 ai=1.0]
Source corroboration
Single-source cluster
Only one publisher covered this so far. Single-source stories can still rank when the publisher is high-authority, but they lack cross-source corroboration.
Topics
paper, safety
Editorial topic classification. Feeds into how the story surfaces on /topic/<slug> hub pages and into the per-entity coverage mix.
AI-industry relevance
High
Clearly on-topic for AI-industry coverage.
Story freshness
Breaking (< 6h)
Fresh story with cross-source coverage still developing. Ranking may shift as more sources report.

Full methodology in our editorial standards.

COVERAGE [1]

  1. arXiv cs.CL TIER_1 English(EN) · Yu-Yu Yang, Ti-Rong Wu, Hung Guei, Hsing-Yu Chen, I-Chen Wu ·

    Do LLMs Trust the Accuser or the Accusation? Measuring Belief Shifts in Werewolf

    arXiv:2609.12446v1 Announce Type: cross Abstract: Social-deduction games such as Werewolf are increasingly used to evaluate LLM agents, but existing evaluations often rely on final game outcomes. We propose a belief-shift evaluation benchmark in Werewolf for analyzing communicati…

RELATED ENTITIES

RELATED TOPICS