“The "Scaling Monosemanticity" paper provided significant evidence that dictionary learning works on large, production-grade models like Claude 3 Sonnet, not just on small, one-layer models.”