论文Agent 与开发者09:58Forking: Sudden Overfitting Under ReplayDeep学习又发现新现象'分叉',比Grokking和Double Descent还奇怪,AutoResearch agents的技巧可能带来意外副作用。#NanoGPT#DeepSeek#Engram#n-gramaarXiv: DeepSeek@Shanbin Yu 等 5 人原文稍后读已读值得跟进有用关注 NanoGPT