Why can't transformers learn multiplication? reverse-engineering reveals long-range dependency pitfalls

{
  "doi": "10.48550/ARXIV.2510.00184",
  "year": 2025,
  "title": "Why can't transformers learn multiplication? reverse-engineering reveals long-range dependency pitfalls",
  "venue": "CoRR",
  "authors": [
    "Xiaoyan Bai",
    "Itamar Pres",
    "Yuntian Deng",
    "Chenhao Tan",
    "Stuart M. Shieber",
    "Fernanda B. Viégas",
    "Martin Wattenberg",
    "Andrew Lee"
  ],
  "arxiv_id": "2510.00184",
  "s2_paper_id": null,
  "ingest_status": "referenced-only"
}