{
  "id": 6542,
  "url": "https://arxiv.org/abs/2603.29735v2",
  "title": "Unveiling the Reasoning Process of Large Language Models",
  "summary": "Large language models often reason beyond surface tokens, but the internal stage at which token-level information becomes abstract relational structure remains unclear. We investigate this question by analyzing how attention heads and layers transform information during autoregressive reasoning. Across mathematical and symbolic reasoning tasks, we observe a consistent layer-wise division of labor: outer layers mainly preserve and route input-related features, whereas middle layers reorganize the",
  "authors": "Junjie Zhang, Zhen Shen, Xisong Dong, Gang Xiong",
  "category": "research",
  "topics": "jobs-economy,finance-investment",
  "orgs": null,
  "regions": null,
  "published_at": "2026-03-31T13:36:08.000Z",
  "fetched_at": "2026-07-14T16:32:33.102Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/6542",
  "original_url": "https://arxiv.org/abs/2603.29735v2",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}