{"ID":2878364,"CreatedAt":"2026-06-01T04:54:23.091178241Z","UpdatedAt":"2026-06-01T04:54:23.091178241Z","DeletedAt":null,"paper_url":"https://arxiv.org/abs/2508.17627","arxiv_id":"2508.17627","title":"The Evolution of Thought: Tracking LLM Overthinking via Reasoning Dynamics Analysis","abstract":"Test-time scaling via explicit reasoning trajectories significantly boosts large language model (LLM) performance but often triggers overthinking. To explore this, we analyze reasoning through two lenses: Reasoning Length Dynamics, which reveals a compensatory trade-off between thinking and answer content length that eventually leads to thinking redundancy, and Reasoning Semantic Dynamics, which identifies semantic convergence and repetitive oscillations. These dynamics uncover an instance-specific Reasoning Completion Point (RCP), beyond which computation continues without further performance gain. Since the RCP varies across instances, we propose a Reasoning Completion Point Detector (RCPD), an inference-time early-exit method that identifies the RCP by monitoring the rank dynamics of termination tokens (e.g., \u003c/think\u003e). Across AIME and GPQA benchmarks using Qwen3 and DeepSeek-R1, RCPD reduces token usage by up to 44% while preserving accuracy, offering a principled approach to efficient test-time scaling.","short_abstract":"Test-time scaling via explicit reasoning trajectories significantly boosts large language model (LLM) performance but often triggers overthinking. To explore this, we analyze reasoning through two lenses: Reasoning Length Dynamics, which reveals a compensatory trade-off between thinking and answer content length that e...","url_abs":"https://arxiv.org/abs/2508.17627","url_pdf":"https://arxiv.org/pdf/2508.17627v2","authors":"[\"Zihao Wei\",\"Liang Pang\",\"Jiahao Liu\",\"Wenjie Shi\",\"Jingcheng Deng\",\"Shicheng Xu\",\"Zenghao Duan\",\"Fei Sun\",\"Huawei Shen\",\"Xueqi Cheng\"]","published":"2025-08-25T03:17:17Z","proceeding":"cs.CL","tasks":"[\"cs.CL\",\"cs.AI\"]","methods":"[\"Large Language Model\",\"Language Model\"]","has_code":false}