evhub comments on Discovering Language Model Behaviors with Model-Written Evaluations