LATIDIA · Investigación

Aprender más formando menos: contracción progresiva guiada por la incertidumbre para un razonamiento eficiente después de la formación

arXiv: 2610.12933v1Tipo de anuncio: nuevo Resumen: Los modelos de lenguaje grandes han logrado avances notables en el razonamiento complejo. Sin embargo, el razonamiento de escala posterior a la capacitación con cuerpos cada vez más grandes puede aumentar sustancialmente el

La noticia

arXiv:2610.12933v1 Announce Type: new Abstract: Large language models have achieved remarkable advances in complex reasoning. However, scaling reasoning post-training with ever-larger corpora can substantially increase training cost without proportional performance gains. In this study, we therefore ask: how far can reasoning supervision be compressed without sacrificing capability? To this end, we propose an uncertainty-guided progressive contraction framework that concentrates supervision and optimization on the reasoning trajectories most valuable for learning. Specifically, rather than treating a teacher-generated chain-of-thought (CoT) as a single realized path, we construct uncertainty-profiled reasoning supervision by building Monte Carlo reasoning trees around intermediate reasoning states to expose process-level uncertainty. This uncertainty first contracts the original corpus into