Research on how much language models can infer about their current user, and interpretability work on such inferences | grantmaking.ai