AI界的“学霸”和“学神”：差的不是智商，是训练方法

[CL] OctoThinker: Mid-training Incentivizes Reinforcement Learning Scaling
[Shanghai Jiao Tong University]
arxiv.org