Dual-Reasoning Framework for Robust and Interpretable LLMs
Опубліковано: 2025-12-05
Переглянути на arXiv →Анотація
This work introduces a dual-reasoning training framework that integrates affirmative generation with structured counterfactual denial, leading to more robust, interpretable, and human-reasoning-aligned Large Language Models.