Drift and selection in LLM text ecosystems
arXiv:2604.08554v1 Announce Type: new
Abstract: The public text record — the material from which both people and AI systems now learn — is increasingly shaped by its own outputs. Generated text enters the public record, later agents learn from it, a…