I think it largely doesn’t matter, models can (and do) often come out erring on the side of considering things evals.
I think it largely doesn’t matter, models can (and do) often come out erring on the side of considering things evals.