{
  "id": 7651,
  "url": "https://arxiv.org/abs/2603.05228v3",
  "title": "The Geometric Inductive Bias of Grokking: Bypassing Phase Transitions via Architectural Topology",
  "summary": "Mechanistic interpretability typically relies on post-hoc analysis of trained networks. We instead adopt an interventional approach: testing hypotheses a priori by modifying architectural topology to observe training dynamics. We study grokking - delayed generalization in Transformers trained on cyclic modular addition (Zp) - investigating if specific architectural degrees of freedom prolong the memorization phase. We identify two independent structural factors in standard Transformers: unbounde",
  "authors": "Alper Yıldırım",
  "category": "research",
  "topics": "bias-fairness,safety-alignment,finance-investment",
  "orgs": null,
  "regions": null,
  "published_at": "2026-03-05T14:41:01.000Z",
  "fetched_at": "2026-07-14T16:33:21.052Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/7651",
  "original_url": "https://arxiv.org/abs/2603.05228v3",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}