AI evals shouldn't be done by underpaid students but by stable, tenured research staff

tallinzen · x · 2026-09-17

Continuing the discussion: the author agrees it doesn't make sense for AI evals to be done by underpaid students who will soon be job-hunting; evals should be run by people with stable, well-paid staff positions, essentially tenured research scientists, to reduce incentive distortions.

Related event: Scholars debate who should run independent AI model evaluations(3 posts)→

Original post →

More from Safety

Safety channel →