Evaluation

Safety reward model rejection of context distillation errors

Using the safety reward model to reject context-distillation errors (cases where distillation degraded response quality on adversarial prompts); safety context distillation applied only to adversarial prompts since helpful prompts degrade