Metraipolis › archive › dc_d910e5d91e97
Not Too Hard, Not Too Easy: Learning from Intermediate States for LLM Structured Reasoning — arXiv:2609.33149
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.33149
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-07T20:56:29.943Z
- SHA-256 of the archived text
f4fd67885de9b40d5d5ebf7f5ef101116fed32a07cda62a0dc9a132996f35440- Stable URL of this page
https://metraipolis.com/documento/dc_d910e5d91e97
How to cite
Not Too Hard, Not Too Easy: Learning from Intermediate States for LLM Structured Reasoning — arXiv:2609.33149 — arXiv (en), https://arxiv.org/abs/2609.33149 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_d910e5d91e97 (Metraipolis archive dc_d910e5d91e97, sha256 f4fd67885de9b40d5d5ebf7f5ef101116fed32a07cda62a0dc9a132996f35440)
BibTeX
@misc{dcd910e5d91e97,
title = {Not Too Hard, Not Too Easy: Learning from Intermediate States for LLM Structured Reasoning — arXiv:2609.33149},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.33149},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_d910e5d91e97; sha256 f4fd67885de9b40d5d5ebf7f5ef101116fed32a07cda62a0dc9a132996f35440},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.33149. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
A common principle of effective learning is to practice material that is neither already mastered nor too difficult to permit progress. We ask how to apply this principle to structured reasoning tasks such as Sudoku and maze solving. In these tasks, a model can repeatedly revise an incomplete or incorrect candidate solution until it satisfies the problem's constraints. The intermediate candidate solutions along this trajectory provide natural training examples: some are already solved, some cannot yet be repaired by the model, and others lie at its current frontier of achievable progress. We therefore investigate whether pretrained language models can learn to revise such states and whether training on states at this frontier improves reasoning more broadly. To achieve this, we couple a pretrained language-model backbone with a recurrent updater that repeatedly revises an explicit solution state, using the same parameters at every update step. We further introduce Frontier-Oriented Curation Using Self-trajectories (FOCUS), which selects training states from trajectories generated by the current model. FOCUS measures how much the model improves each state within a fixed number of recurrent updates and prioritizes states from which it can make substantial progress. With Qwen3-1.7B, FOCUS achieves 64.4% exact solve accuracy on Sudoku-Extreme and 91.1% on Maze-Hard, with similar gains observed across five Qwen and Llama backbones spanning 1.7B to 8B parameters. We further observe zero-shot transfer in the adapted LLM to mathematical reasoning and code execution, even when the recurrent updater is disabled and no downstream fine-tuning is performed. Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-29 in cs.AI). Categorie arXiv: Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2609.33149 — 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 2026-10-07T20:29:01.169Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-29?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).