method · first seen 2 Nov 2026
weight-space programming
Also: WSP
Using an LLM to directly edit neural network weights to diagnose and fix model behaviour without gradient training.
Weight-space programming (WSP) has a frontier LLM read and write the raw parameters of another neural network directly, with no gradient descent, to diagnose bugs, repair behaviour, or build small networks from a task description. It traces to an MIT blog post and paper by Takehi et al., in which a GPT-6 Astra agent fixed reasoning loops in Qwen3.5-0.8B. Within two weeks, independent groups extended it to image-restoration diffusion models, clinical network auditing, quantum circuit angles, RL policies checked with model checking, tabular foundation models, and medical classifiers. Others produced a benchmark (WeightBench), a backdoor security study, and a discrete-optimization alternative. Two small tooling repos have also appeared.
If it holds up, LLM-driven weight editing could become a practical tool for auditing and patching small models without training data or retraining. It also opens a new attack surface, since the same ability could be used to insert backdoors.
Analyst note (emerging), 17 Nov 2026.
Mentions per week
PapersCodeDiscussion
Rank 1 on 16 Nov 2026 · 22 mentions in 28 days · 21 independent groups · discussion → papers → code · first seen 2 weeks ago · acceleration +1.71
Earliest sources
- Astra can read and write weights to improve QwenDiscussion · 2 Nov 2026 · dbreunig · 2,971 points
- Weight-Space Programming for Improving Image-Restoration Diffusion Models: Technical Conjectures and a Recursive Improvement LoopPapers · 3 Nov 2026 · Zhuoran Zheng, Doubao
- Weight-space programming for auditing clinical decision networksPapers · 3 Nov 2026 · Humza Nusrat, Omar Nusrat
- felipepenha/pi-weight-space-programmingCode · 3 Nov 2026 · felipepenha · 15 stars
- AuremHQ/weight-space-programmingCode · 4 Nov 2026 · AuremHQ · 18 stars
- WeightLoop: Weight-Space Programming with Experimental FeedbackPapers · 5 Nov 2026 · Mengran Li, Bo Li, Weijing Zhao, Chengyang Zhang, Boyu Zhang, Jinfeng Xu, Zelin…
Latest sources
- Replicating and Extending Weight-Space Programming: Frontier LLMs Compile Programs into Weights and Read Input LayersPapers · 16 Nov 2026 · Roksana Goworek, Alexander Hickey, Federico Nanni, Martin O'Reilly
- Weight-Space Programming Changes How an LLM Uses Python for Computational BiologyPapers · 16 Nov 2026 · Makoto Sato
- Weights & Caches: LLMs Can Program Tabular Foundation ModelsPapers · 15 Nov 2026 · Shi Bin Hoo, Magnus Buehler, Omar Swelam, Frank Hutter, Lennart Purucker
- Zero-Shot Weight-Space Programming for Prediction and Classification of Target Variables in Structured DatasetsPapers · 14 Nov 2026 · Eduardo C. Garrido-Merchán
- WeightBench: Evaluating Language Models as Weight-Space ProgrammersPapers · 13 Nov 2026 · Aayush Marasini, Barsat Khadka, Nick Rahimi
- Can LLMs Write Backdoors? An Empirical Security Study of Weight-Space ProgrammingPapers · 12 Nov 2026 · Zhaojia Wang, Jiucui Lu, Zheng Li
- Compiling Known Rules into Compact Neural Agents: Weight-Space Programming on Unseen GamesPapers · 12 Nov 2026 · Jens Lehmann
- Tree-Ising: Repairing Language Models with Bit-Level Ising EditsPapers · 10 Nov 2026 · Hadi Salloum, Ammar Ali
- Weight-Space Programming for Medical Imaging: Do We Still Need Training Data?Papers · 9 Nov 2026 · Maximilian Lindholz, Richard Ruppel, Annette Milnik, Sebastian Halskov, Semil Em…
- Weight-Space Programming of Policies and Their Verification via Model CheckingPapers · 8 Nov 2026 · Dennis Gross
- Weight-Space Programming of Variational Quantum Circuits: Large Language Models Write Quantum Circuit Parameters Without an OptimizerPapers · 8 Nov 2026 · Debarshi Kundu, Swaroop Ghosh, Vasant G Honavar
- Physics2Weight: Direct Neural Network Architecture and Weight Generation from Physical Laws Using Large Language ModelsPapers · 7 Nov 2026 · Shogo Sato, Tomoyasu Shimada, Kazuhiko Murasaki, Takashi Sano
Mentioned together with
In the ledger
- 2026-W47 · call #1 · open