Metraipolis › archive › dc_ed1858aea079
The Imitation Game: When LLMs Learn to Reason Like Programs via Code-Centric Reasoning Data Synthesis — arXiv:2609.16076
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2609.16076
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-15T18:32:38.215Z
- SHA-256 of the archived text
8187e4987e690cfcae8a9c649fc764f2f7406f421f610d886f84fde87566849c- Stable URL of this page
https://metraipolis.com/documento/dc_ed1858aea079
How to cite
The Imitation Game: When LLMs Learn to Reason Like Programs via Code-Centric Reasoning Data Synthesis — arXiv:2609.16076 — arXiv (en), https://arxiv.org/abs/2609.16076 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_ed1858aea079 (Metraipolis archive dc_ed1858aea079, sha256 8187e4987e690cfcae8a9c649fc764f2f7406f421f610d886f84fde87566849c)
BibTeX
@misc{dced1858aea079,
title = {The Imitation Game: When LLMs Learn to Reason Like Programs via Code-Centric Reasoning Data Synthesis — arXiv:2609.16076},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2609.16076},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_ed1858aea079; sha256 8187e4987e690cfcae8a9c649fc764f2f7406f421f610d886f84fde87566849c},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2609.16076. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Large Language Models (LLMs) excel at programming tasks but frequently fail at deterministic, fine-grained reasoning in natural language, relying heavily on semantic approximations rather than robust symbolic execution. To bridge this gap, we propose MIMIC, a framework that leverages executable code as a rigorous medium for reasoning data synthesis. MIMIC fundamentally transforms algorithms into verifiable reasoning trajectories through narrative fusion, code-guided test synthesis, and dynamic code instrumentation. Crucially, these explicit intermediate execution states naturally form a Code-Instrumented Reward (CIR), providing dense, high-fidelity process supervision for reinforcement learning without external reward models. Extensive evaluations reveal that models trained via SFT and GRPO on our synthesized dataset achieve substantial, consistent gains. Our method significantly elevates accuracy across general reasoning, complex mathematical benchmarks, and fine-grained deterministic tasks, demonstrating that the procedural rigor of executable code can effectively unlock and enhance the generalized reasoning capabilities of LLMs. Our code and data are available at this https URL . Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-16 in cs.AI). Categorie arXiv: Computation and Language (cs.CL) ; Artificial Intelligence (cs.AI). Pagina del lavoro: https://arxiv.org/abs/2609.16076 — 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 2026-10-15T18:03:09.581Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-09-16?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).