A Geometric Calculator Inside a Neural Network

{
  "doi": "10.48550/arxiv.2605.01148",
  "pdf": "https://arxiv.org/pdf/2605.01148",
  "url": "https://arxiv.org/abs/2605.01148",
  "tags": [
    "mechanistic-interpretability",
    "circuits",
    "geometry-of-representations",
    "Llama-3",
    "circular-features",
    "addition-circuit",
    "goodfire"
  ],
  "year": 2026,
  "saved": "2026-05-14",
  "title": "A Geometric Calculator Inside a Neural Network",
  "venue": "arXiv preprint",
  "status": "full-text-saved",
  "authors": [
    "Sheridan Feucht",
    "Ekdeep Singh Lubana",
    "Tal Haklay",
    "Thomas Fel",
    "Usha Bhalla",
    "Atticus Geiger",
    "Daniel Wurgaft",
    "Can Rager",
    "Raphaël Sarfati",
    "Jack Merullo",
    "Thomas McGrath",
    "Owen Lewis"
  ],
  "landing": "https://www.goodfire.ai/research/a-geometric-calculator",
  "arxiv_id": 2605.01148,
  "published": "2026-05-14",
  "affiliation": "Goodfire (+ collaborators)",
  "openalex_id": "W7160247613",
  "openalex_year": 2026,
  "openalex_enriched_at": 1778977723,
  "openalex_match_title": "Arithmetic in the Wild: Llama uses Base-10 Addition to Reason About Cyclic Concepts",
  "openalex_cited_by_count": 0
}