Exploring the Residual Stream of Transformers for Mechanistic Interpretability — Explained — LessWrong