Interpreting Large Language ModelsWe share initial thoughts on how to peer inside transformer networks.