AI RESEARCH
State Transfer Reveals Reuse in Controlled Routing
arXiv CS.AI
•
ArXi:2604.18158v1 Announce Type: new Prompt-based interventions can change model behavior, but trained success alone does not identify where the behaviorally relevant state is represented. We study this question in controlled routing tasks using interfaces chosen on data, held-out query evaluation, and matched necessity, sufficiency, and wrong-interface controls. On GPT-2 triop, an early interface s exact transfer under these tests.