Mechanical Interpretability

• How GPT learns layer by layer