Language Models Can Rewrite Their Own Weights Without Training
A new arXiv paper shows a 27B language model can rewrite its own weights without training or gradients, boosting unseen-task accuracy by 6.6 points and repairing a corrupted copy from 7% to 70% accuracy.
2026-11-18 ~ 2026-11-18 · 3 related posts
- Language Models Can Rewrite Their Own Weights: no gradients, no training — jiqizhixin · 2026-11-18
- 27B Model Rewrites Its Own Weights Without Training or Gradients, Paper Claims — udmrzn · 2026-11-18
1 near-duplicate retellings: jiqizhixin