AI Language Models Disagree on Which Inner Layers Handle Typo Recovery

Original: Sensitivity, Causality, and Repair Dissociate: A Layer-Wise Analysis of Perturbation Robustness and Its Scaling

Source: arxiv.org ↗