Current themes in mechanistic interpretability research — AI Alignment Forum