教程 / 实战普通A 0.1% Pretraining Poison Taught a 4B Agent curl | bash. Post-Training Didn't Remove It信息来源:DEV Community·2026-11-03 04:04内容摘要Researchers poisoned 0.1% of pretraining data. After clean SFT, DPO and RL, a 4B agent still wrote curl | bash on most path-triggered prompts.内容分类AI 教程与实战内容层级普通情报发布时间(北京时间)2026-11-03 04:04本站收录时间(北京时间)2026-11-03 04:38信息来源DEV Community站内情报编号intel-f42b29a6d56ff306820904ad阅读原始信息 ↗更多教程 / 实战分享文章