Agreed, reward models are great in general. Just not as flexible for some evals.

→ View original post on X — @maximelabonne