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。
OpenAI secures up to $110bn in record funding deal
，这一点在safew官方版本下载中也有详细论述
union object_info *h;
The Endowment is a very promising project. I am happy to join and pretty excited to