So Many Opinions, So Many LLMs: Comparing Large Language Models to Traditional Machine Learning for Open- Ended Survey Analysis

Collection of recent arXiv papers (15 July 2026) on large language models, agent-based systems, and socio-technical impacts. Key themes include: LLMs outperforming traditional ML on open-ended survey analysis but presenting explainability/consistency trade-offs; AgentSociety 2 for integrated, auditable agent-based social science experiments; an AI insurance “stack” blueprint addressing correlated losses, rising incident severity, and tail risks from frontier models (including discussion of AI CAT scenarios such as CBRN and critical-infrastructure collapse); risks of using ‘policy-as-prompt’ LL