I think that has to do more with the thinking "train of thought" that some models show as what the model is processing before making the response. There shouldn

I think that has to do more with the thinking "train of thought" that some models show as what the model is processing before making the response. There shouldn't be a distillation risk with actually asking the model to explain why it made a decision and getting the response.