Training method helps AI reasoning survive out-of-distribution examples

Original: Invariant Gradient Alignment for Robust Reasoning Distillation

Source: arxiv.org ↗