New LLM framework CogWM tracks cognitive states for social influence evaluation

Researchers have developed a new LLM-based framework called CogWM to evaluate social influence in dialogues. This model focuses on tracking changes in a user's internal cognitive states, such as beliefs, desires, intentions, and emotions, rather than just surface-level text metrics. CogWM acts as both a user simulator and an evaluation platform, utilizing a three-tier framework for comprehensive assessment. The model was trained on over 150,000 user-turn samples and demonstrated improved accuracy in predicting emotional states, outperforming GPT-5.5. In discrimination trials, CogWM successfully identified commercial agents based on their cognitive influence, with Llama-4-Scout performing best. AI

IMPACT This research introduces a novel method for evaluating AI's impact on human cognitive states in dialogues, potentially improving AI's ability to understand and influence users.

RANK_REASON The cluster contains an academic paper detailing a new model and evaluation framework. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• CogWM
• GPT-5.5
• Llama-4-Scout

• paper
• model release

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