Research Scientists wanted to advance Ambitious Mechanistic Interpretability by building data structure models, quantifying how AI systems learn features from data, and creating synthetic datasets to benchmark interpretability tools.