{
  "id": "2604.07822",
  "title": "Loop, Think, & Generalize: Implicit Reasoning in Recurrent-Depth Transformers",
  "first_public": "2026-04-09",
  "paper_url": "https://arxiv.org/abs/2604.07822",
  "authors": "Harsh Kohli, Srinivasan Parthasarathy, Huan Sun, Yuekun Yao",
  "arxiv": "2604.07822",
  "added_at": "2026-11-09",
  "venue": "arXiv",
  "provenance": "Primary-source literature review, 2026-11-09",
  "lens": [
    "recurrent",
    "analysis"
  ],
  "representation": [
    "hidden-state"
  ],
  "computation": [
    "recurrent",
    "depth-wise"
  ],
  "training": [
    "supervised"
  ],
  "questions": [
    "generalization",
    "capability",
    "scaling"
  ],
  "evidence": [
    "empirical",
    "mechanistic"
  ],
  "modality": [
    "language",
    "symbolic"
  ],
  "summary": "Trains transformers to answer multi-hop knowledge queries without writing intermediate steps. Tests whether repeated shared layers help combine unfamiliar facts and solve longer chains, and examines when extra inference loops help or cause overthinking.",
  "author_names": [
    "Harsh Kohli",
    "Srinivasan Parthasarathy",
    "Huan Sun",
    "Yuekun Yao"
  ],
  "record_url": "https://latent-implicit-thinking.github.io/papers/2604.07822/",
  "legacy_record_url": "https://latent-implicit-thinking.github.io/papers/#paper-2604-07822",
  "citation_date": "2026-04-09",
  "json_url": "https://latent-implicit-thinking.github.io/papers/2604.07822/record.json",
  "markdown_url": "https://latent-implicit-thinking.github.io/papers/2604.07822/index.md",
  "bibtex_url": "https://latent-implicit-thinking.github.io/papers/2604.07822/citation.bib",
  "csl_url": "https://latent-implicit-thinking.github.io/papers/2604.07822/citation.csl.json",
  "pdf_url": "https://arxiv.org/pdf/2604.07822"
}