XFlowMap: Cross-Scale Generalization and Mapping of Massive Origin-Destination Data

Collection of recent arXiv CS/social-informatics papers. Key security-relevant findings: (1) “LLM Agents Make Collective Belief Dynamics Programmable” demonstrates proof-of-concept that coordinated LLM agents can steer population beliefs quickly and stably, and identifies four properties (indistinguishability, persistence, contextuality, configurability) that make detection and defense difficult. (2) “Can LLMs Emulate Human Belief Dynamics?” shows LLMs systematically differ from human belief distributions (e.g., higher conformity), cautioning against using LLMs as human proxies in social-sim;(