Identify neuron activations that are decisive in a model’s factual predictions. Rank-One Model Editing (ROME)