Improving Language Model Behavior by Training on a Curated Dataset

Our latest research finds we can improve language model behavior with respect to specific behavioral values by fine-tuning on a small, curated dataset.