Causally Grounded Mechanistic Interpretability for LLMs with Faithful Natural-Language Explanations

Ajay Pravin Mahale

Mechanistic interpretability identifies internal circuits responsible for model behaviors, yet translating these findings into human-understandable explanations remains an open pro…