Metraipolis › archive › dc_2dedb22c0f33
Selection Is Not Learning: Choosing Model Futures — arXiv:2610.22029
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2610.22029
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-16T09:30:55.302Z
- SHA-256 of the archived text
f04fbe2c74ccff3320b1bff9cb37e74f53219f0be10888ea84b57fbf6a287168- Stable URL of this page
https://metraipolis.com/documento/dc_2dedb22c0f33
How to cite
Selection Is Not Learning: Choosing Model Futures — arXiv:2610.22029 — arXiv (en), https://arxiv.org/abs/2610.22029 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_2dedb22c0f33 (Metraipolis archive dc_2dedb22c0f33, sha256 f04fbe2c74ccff3320b1bff9cb37e74f53219f0be10888ea84b57fbf6a287168)
BibTeX
@misc{dc2dedb22c0f33,
title = {Selection Is Not Learning: Choosing Model Futures — arXiv:2610.22029},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2610.22029},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_2dedb22c0f33; sha256 f04fbe2c74ccff3320b1bff9cb37e74f53219f0be10888ea84b57fbf6a287168},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.22029. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Self-evolving language models train on tasks they generate, so every round must decide which updated model the next round inherits. Current loops such as R-Zero, Agent0, and Skill Self-Play make this decision through the task, admitting tasks that look learnable and inheriting every resulting update. We identify a selection-to-update gap in these loops, where a task that looks learnable need not produce a useful update. Most updates on verified tasks fail to make correct answers more likely, yet these loops inherit them all. HeirCommit freezes the updated model and uses fresh evaluation questions to decide whether to inherit it or retain the current model. EvoHeir separates parameter inheritance from experience inheritance, using current and archived experience to repair failures and preserve successes before final confirmation. Our analysis characterizes selection-induced optimism and decomposes regret into selection, absorption, and confirmation errors. HeirCommit cuts the harmful share of accepted updates by 68-83% relative to random inheritance at the same acceptance rate. Across four backbones, EvoHeir achieves the best mathematics and general-reasoning averages, raising the math average by 5.3% relative to the strongest baseline, Agent0. The framework lets useful experience persist even when the model that produced it is not inherited. Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno). Categorie arXiv: Machine Learning (cs.LG) ; Artificial Intelligence (cs.AI); Computation and Language (cs.CL). Pagina del lavoro: https://arxiv.org/abs/2610.22029 — 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 2026-10-16T08:44:18.021Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).