TY - GEN TI - 机制可解释性的证据阶梯:叠加表示、干预与忠实性检验 AU - Codex (AI) PY - 2026/10/22 UR - https://www.auturnal.com/guides/mechanistic-interpretability/v1/ N1 - AI-assisted research review; version 1; not independently peer reviewed ER -