Metraipolis › archive › dc_47faff97972c
Exploring Weight-Space Programming: Capabilities, Generalization, and Limitations — arXiv:2611.09382
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2611.09382
- License (as archived)
- CC0-1.0
- Archived
- 2026-11-09T22:34:04.679Z
- SHA-256 of the archived text
46d42f357a92be5d03ab41ddc62f3f7143065e8b0882e207acca3627ca0474d2- Stable URL of this page
https://metraipolis.com/documento/dc_47faff97972c
How to cite
Exploring Weight-Space Programming: Capabilities, Generalization, and Limitations — arXiv:2611.09382 — arXiv (en), https://arxiv.org/abs/2611.09382 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_47faff97972c (Metraipolis archive dc_47faff97972c, sha256 46d42f357a92be5d03ab41ddc62f3f7143065e8b0882e207acca3627ca0474d2)
BibTeX
@misc{dc47faff97972c,
title = {Exploring Weight-Space Programming: Capabilities, Generalization, and Limitations — arXiv:2611.09382},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2611.09382},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_47faff97972c; sha256 46d42f357a92be5d03ab41ddc62f3f7143065e8b0882e207acca3627ca0474d2},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.09382. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Weight-Space Programming (WSP) enables reasoning agents to directly inspect and modify neural network parameters, offering an alternative to gradient-based model adaptation. However, its capabilities, cross-task generalization, and practical limitations remain insufficiently understood. We systematically investigate WSP through 16 editing trajectories across seven benchmarks and two model scales (Qwen3.5-0.8B and Qwen3.5-9B). Our findings reveal three key insights. First, WSP can achieve substantial performance gains through highly localized interventions. On full held-out evaluations, accuracy improves by up to 29.45, 4.42, 36.72, and 17.11 percentage points on MATH-500, GPQA Diamond, IFEval, and SciBench, respectively. Notably, modifying only 1,023 parameters yields a 29.45-point improvement. Second, WSP exhibits promising but uneven cross-task generalization. A Math-edited checkpoint improves GPQA Diamond, IFEval, and SciBench by 3.03, 12.18, and 20.10 points, respectively, without further editing, while leaving SciCode performance unchanged. Third, effective editing remains challenging and inconsistent. Outcomes vary substantially across trajectories, and six of the 16 trajectories retain the original weights when candidate improvements cannot be reliably validated. Together, these results highlight WSP's potential for highly localized model adaptation while revealing important challenges in intervention discovery, reliability, and cross-task transfer. Homepage: this https URL GitHub: this https URL Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno). Categorie arXiv: Artificial Intelligence (cs.AI) ; Machine Learning (cs.LG). Pagina del lavoro: https://arxiv.org/abs/2611.09382 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito). Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina. Letto il 2026-11-09T22:30:47.278Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).