Transplanting, inverting, and preventing a misalignment persona: method-conditional emergent misalignment in Qwen2.5 文章

ArXiv CS.CL2026-08-05PAPERen作者: Lyndon Drake (University of Oxford), Zandi Eberstadt (University of Oxford)

Transplanting, inverting, and preventing a misalignment persona: method-conditional emergent misalignment in Qwen2.5 · 相关公司