Reasoning as Trajectories

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

• Laerdon Kim

related reading

• the-illusion-of-thinking.pdfml-site.cdn-apple.com
• Circuit Tracing: Revealing Computational Graphs in Language Modelstransformer-circuits.pub
• [2201.11903] Chain of Thought Prompting Elicits Reasoning in Large Language Modelsarxiv.org
• [2412.06769] Training Large Language Models to Reason in a Continuous Latent Spacearxiv.org
• Training Large Language Models to Reason in a Continuous Latent Spacearxiv.org
• [2603.07267] How to Steal Reasoning Without Reasoning Tracesarxiv.org
• [2203.14465] STaR: Bootstrapping Reasoning With Reasoningarxiv.org
• Reasoning Models Reason Well, Until They Don'tarxiv.org
• Thinking to recall: How reasoning unlocks parametric knowledge in LLMsresearch.google
• [2606.10346] Reasoning or Memorization? Direction-Aware Diversity Exploration in LLM Reinforcement Learningarxiv.org
• Base Models Know How to Reason, Thinking Models Learn Whenarxiv.org
• You Only Need Minimal RLVR Training: Extrapolating LLMs via Rank-1 Trajectoriesarxiv.org