Self-Rewriting paper lets LLMs modify their own weights without backpropagation
kalyan_kpl · x · 2026-11-19
A new paper introduces Self-Rewriting, an approach enabling LLMs to directly modify their own neural network weights using their reasoning capabilities—no gradients, backpropagation, or conventional training required.
- Problem: frozen weights mean improvement needs extra fine-tuning; existing self-improvement methods only tweak prompts, generate data, or rely on external models
- Pipeline: diagnose wrong answers and recurring failure patterns; inspect internal representations, hidden states, and weight matrices; hypothesize which components cause failures and rewrite them directly
Related event: Research Shows LLMs Can Rewrite Their Own Weights Without Training(4 posts)→
More from Research
- SE3 Labs Releases First Open YAM Real-Robot Teleop Dataset: 500 Bimanual Episodes — paigeinsf · 2026-11-19
- 1.5 years under review: researchers urge releasing AI papers immediately after submission — carlosdponx · 2026-11-19
- Berkeley's StreamEmbed indexes your screen history on-device, 3.7x faster at matched accuracy — berkeley_ai · 2026-11-19
- NAACL RAF opens 2026-2027 grant proposals for NLP initiatives, due May 6 — keviv9 · 2026-11-19
- Open-sourced Indic profanity detector covers 12 Indian languages, runs at 4ms per comment on CPU — sudalairajkumar · 2026-11-19
- MAVE reconstructs editable 3D blockouts from film shots, works inside Blender — bdsqlsz · 2026-11-19