Suppressed, Not Erased: A Representational Trace of Edited Facts Survives Even Weight-Free Knowledge Editing
arXiv:2609.18985v1 Announce Type: new Abstract: Knowledge-editing benchmarks certify local correctness, whether an edited model produces the new fact on near-edit prompts but not how much of the original fact remains decodable inside the model. We study residu
arXiv cs.AI··Updated just now·34 sightings