Research focused on developing strategies to align machine learning systems with human interests, including formalizing explanations of neural network behaviors.

Combining mechanistic interpretability with formal verification to understand and verify the internal workings of neural networks.