Researchers ran a controlled experiment placing LLM agents on a simulated social platform to study how peer-ranked feeds shape agent behavior. They found that showing agents a feed of prior-round peer posts ranked by peer-generated likes increased lexical similarity among agents’ final-round posts, indicating that ranking mechanisms alone can drive linguistic convergence. However, exposing agents to multiple distributed information sources showed no reliable advantage over single sources, suggesting exposure diversity alone doesn’t consistently shift agent stances.
