flâneur

Reasoning as Trajectories

slhleosun.github.io · 1,061 words · saved by 1 readers

LLM Reasoning as Trajectories: Step-Specific Representation Geometry and Correctness Signals

Reasoning as Trajectories LLM Reasoning as Trajectories : Step-Specific Representation Geometry and Correctness Signals Lihao Sun , Hang Dong , Bo Qiao , Qingwei Lin , Dongmei Zhang , Saravan Rajmohan Microsoft ACL 2026 (Main) Paper arXiv Code Chain-of-thought reasoning in LLMs traces structured trajectories through representation space. Step-specific regions become linearly separable with depth, late-step geometry predicts correctness, and trajectory-based steering enables both error correction and reasoning length control at inference time. Step-specific representation structure across layer

saved by

related reading