Papers09:56Chinese content强化学习提升大语言模型选择性采纳污染检索证据Qwen3.5用强化学习学会在污染检索中挑有用证据,成功率从22%提到26%,还不掉通用分。#SelectBench#DAPO#Qwen3.5#RAG2 sourcesEventaarXiv: DeepSeek@Yanyu Chen et al. (5)SourceRead laterReadWorth followingUsefulFollow SelectBench