LLMs generate social networks influenced by culture and language

Researchers have investigated how large language models (LLMs) generate social networks, finding that prompt design, cultural context, and language significantly influence the outcomes. Their study, using 50 personas across various conditions, revealed that political affiliation often dominates tie formation, while prompt architecture can act as a sociological variable. Although LLM-generated networks exhibit better clustering than standard baselines, they also encode demographic biases exceeding empirical levels. AI

Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →

IMPACT Reveals how prompt engineering choices embed sociological assumptions in LLM outputs.

RANK_REASON Academic paper detailing a study on LLM capabilities. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• LLMs
• Dalal Alharthi Dr.
• GPT-4.1

• paper
• other