蒸馏是模仿，学强模型的输出，把它的「答案形状」复制过来；RL 是探索，模型必须大量自己推理、自己生成、在错误里反复迭代，从试错中提炼能力。
Democrats said that the incident underscored how important it was for there to be a clear public record of the deposition. Rep. Robert Garcia, the top Democrat on the Oversight panel, said that Hillary Clinton, after the incident, repeated her longstanding demand that the deposition be made public, and Democrats called for a video and transcript of the complete proceedings to be released quickly.
，推荐阅读搜狗输入法下载获取更多信息
Жители Санкт-Петербурга устроили «крысогон»17:52
It’s time to go short German government debt, according to strategists at Barclays.
Медведев вышел в финал турнира в Дубае17:59