rome-rank-one-weight-update-layer-9

Summary

ROME fixes a single wrong fact in a language model by making a tiny, targeted tweak to the weights at one specific depth (layer 9), rather than retraining or broadly modifying the model. This matters because the edit is cheap and surgical, meaning the model can correct one piece of knowledge without risking damage to everything else it already "knows."