{
  "id": "2511.16886",
  "title": "Latent Reasoning in TRMs is Secretly a Policy Improvement Operator",
  "authors": "Arip Asadulaev, Rayan Banerjee, Fakhri Karray, Martin Takac",
  "first_public": "2025-11-21",
  "added_at": "2026-11-02",
  "venue": "arXiv",
  "arxiv": "2511.16886",
  "paper_url": "https://arxiv.org/abs/2511.16886",
  "html_url": "https://arxiv.org/html/2511.16886v5",
  "lens": [
    "recurrent",
    "analysis"
  ],
  "representation": [
    "hidden-state",
    "continuous-state"
  ],
  "computation": [
    "recurrent"
  ],
  "training": [
    "supervised",
    "analysis-only"
  ],
  "questions": [
    "theory",
    "interpretability",
    "capability"
  ],
  "evidence": [
    "formal",
    "empirical"
  ],
  "modality": [
    "symbolic"
  ],
  "summary": "Interprets recursive state updates as changes to a model's answer distribution. Trains intermediate updates toward progressively less corrupted puzzle solutions, testing whether recursion improves the answer at each step.",
  "provenance": "Primary-source literature review, 2026-11-02",
  "author_names": [
    "Arip Asadulaev",
    "Rayan Banerjee",
    "Fakhri Karray",
    "Martin Takac"
  ],
  "record_url": "https://latent-implicit-thinking.github.io/papers/2511.16886/",
  "legacy_record_url": "https://latent-implicit-thinking.github.io/papers/#paper-2511-16886",
  "citation_date": "2025-11-21",
  "json_url": "https://latent-implicit-thinking.github.io/papers/2511.16886/record.json",
  "markdown_url": "https://latent-implicit-thinking.github.io/papers/2511.16886/index.md",
  "bibtex_url": "https://latent-implicit-thinking.github.io/papers/2511.16886/citation.bib",
  "csl_url": "https://latent-implicit-thinking.github.io/papers/2511.16886/citation.csl.json",
  "pdf_url": "https://arxiv.org/pdf/2511.16886"
}