Paper Reveals Third Stage of Grokking: Late-Stage Generalization Collapse
A new arXiv paper by Prakash, Charles Martin and the CalcCon team identifies a third stage of grokking in which late-stage training generalization collapses back to random performance, and proposes detecting overfitting directly from weight matrices without any data.
2026-09-20 ~ 2026-09-20 · 2 related posts
- New paper detects overfitting directly from weight matrices without any data — burny_tech · 2026-09-20
- New paper finds 'anti-grokking': test accuracy collapses back to chance after successful generalization — sytelus · 2026-09-20