Tag
The paper introduces PV-SST, a peer-voted social-platform testbed for LLM agents, and finds that peer-ranked feeds increase lexical convergence but do not reliably produce opinion capture or coordination advantages across model families and topics.