Generalization and Robustness: RECKONING Excels on Longer Reasoning Chains Unseen During Training

RECKONING demonstrates superior generalization capacity to longer reasoning chains unseen during training