{
  "id": "2610.39016",
  "title": "Implicit Reasoning in Transformers: Bridging the Compositionality Gap in GPT-2-like Models",
  "authors": "Davide Venuto, Leonardo Venuto",
  "first_public": "2026-10-28",
  "added_at": "2026-11-02",
  "venue": "arXiv",
  "arxiv": "2610.39016",
  "paper_url": "https://arxiv.org/abs/2610.39016",
  "html_url": "https://arxiv.org/html/2610.39016v1",
  "lens": [
    "analysis"
  ],
  "representation": [
    "hidden-state"
  ],
  "computation": [
    "depth-wise"
  ],
  "training": [
    "pretraining",
    "analysis-only"
  ],
  "questions": [
    "generalization",
    "interpretability",
    "capability"
  ],
  "evidence": [
    "empirical",
    "mechanistic"
  ],
  "modality": [
    "language",
    "symbolic"
  ],
  "summary": "Investigates two-hop factual reasoning in a GPT-2-like model trained from scratch. Aligns the positions of atomic facts with the second hop of composite queries, then tests whether the model can answer unseen combinations without written intermediate steps.",
  "provenance": "Primary-source literature review, 2026-11-02",
  "author_names": [
    "Davide Venuto",
    "Leonardo Venuto"
  ],
  "record_url": "https://latent-implicit-thinking.github.io/papers/2610.39016/",
  "legacy_record_url": "https://latent-implicit-thinking.github.io/papers/#paper-2610-39016",
  "citation_date": "2026-10-28",
  "json_url": "https://latent-implicit-thinking.github.io/papers/2610.39016/record.json",
  "markdown_url": "https://latent-implicit-thinking.github.io/papers/2610.39016/index.md",
  "bibtex_url": "https://latent-implicit-thinking.github.io/papers/2610.39016/citation.bib",
  "csl_url": "https://latent-implicit-thinking.github.io/papers/2610.39016/citation.csl.json",
  "pdf_url": "https://arxiv.org/pdf/2610.39016"
}