x
Applying Network Motif Analysis to Transformer Attribution Graphs — LessWrong