What is mechanistic interpretability in LLMs?

A A method for mechanical explanation of model predictions B A research program reverse-engineering the specific circuits and algorithms implemented by transformer weights to understand exactly how models compute their outputs C A framework for interpretable model design D An automated tool for generating model documentation