Comment by red2awn

The "distillation attacks" are mostly using Claude as LLM-as-a-judge. They are not training on the reasoning chains in a SFT fashion.