标签
This paper proposes a three-stream detector that combines residual-stream motion with coarse regions and fine directions to better identify reasoning errors in LLMs, improving selection accuracy by up to 12% over state-of-the-art displacement-only methods.
提出了流形引导的注意力转向(MAGS),这是一种轨迹感知的推理时干预方法,通过将注意力输出投影回学习的正确性流形(当偏差超过阈值时)来纠正LLM中的推理错误,在数学、代码和分子基准测试中优于静态转向方法。