LLMs understand nullability

Steering Characters with Interpretability